Multi-modal machine learning approach for early detection of neurodegenerative diseases leveraging brain MRI and wearable sensor data

Andrew Li; Jie Lian; Varut Vardhanabhuti

doi:10.1371/journal.pdig.0000795

. 2025 Apr 25;4(4):e0000795. doi: 10.1371/journal.pdig.0000795

Multi-modal machine learning approach for early detection of neurodegenerative diseases leveraging brain MRI and wearable sensor data

Andrew Li ¹, Jie Lian ², Varut Vardhanabhuti ^2,^3,^*

Editor: Martin G Frasch⁴

PMCID: PMC12027105 PMID: 40279355

Abstract

Neurodegenerative diseases, such as Alzheimer’s and Parkinson’s Disease, pose a significant healthcare burden to the aging population. Structural MRI brain parameters and accelerometry data from wearable devices have been proven to be useful predictors for these diseases but have been separately examined in the prior literature. This study aims to determine whether a combination of accelerometry data and MRI brain parameters may improve the detection and prognostication of Alzheimer’s and Parkinson’s disease, compared with MRI brain parameters alone. A cohort of 19,793 participants free of neurodegenerative disease at the time of imaging and accelerometry data capture from the UK Biobank with longitudinal follow-up was derived to test this hypothesis. Relevant structural MRI brain parameters, accelerometry data collected from wearable devices, standard polygenic risk scores and lifestyle information were obtained. Subsequent development of neurodegenerative diseases among participants was recorded (mean follow-up time of 5.9 years), with positive cases defined as those diagnosed at least one year after imaging. A machine learning algorithm (XGBoost) was employed to create prediction models for the development of neurodegenerative disease. A prediction model consisting of all factors, including structural MRI brain parameters, accelerometry data, PRS, and lifestyle information, achieved the highest AUC value (0.819) out of all tested models. A model that excluded MRI brain parameters achieved the lowest AUC value (0.688). Feature importance analyses revealed 18 out of 20 most important features were structural MRI brain parameters, while 2 were derived from accelerometry data. Our study demonstrates the potential utility of combining structural MRI brain parameters with accelerometry data from wearable devices to predict the incidence of neurodegenerative diseases. Future prospective studies across different populations should be conducted to confirm these study results and look for differences in predictive ability for various types of neurodegenerative diseases.

Author summary

Neurodegenerative diseases, like Alzheimer’s and Parkinson’s Disease, are a major health concern for the elderly worldwide. Our study aimed to improve the incidence prediction of such diseases by combining two types of data: MRI brain scans and accelerometry data collected from wearable devices. We analysed information from nearly 20,000 participants in the UK Biobank, including their MRI brain scans, accelerometry data, genetic risk scores, and lifestyle information. We recorded the development of neurodegenerative diseases among the participants and used machine learning algorithms to create prediction models. Our findings showed that the model incorporating all factors, including MRI brain parameters and accelerometry data, had the highest accuracy in predicting disease incidence. The results indicate that combining MRI brain scans with accelerometry data could be a powerful approach to predict the onset of neurodegenerative diseases. Further studies are needed to confirm these findings and explore their applicability to different types of neurodegenerative diseases.

Introduction

Neurodegenerative diseases, which encompass conditions such as Alzheimer’s Disease (AD), Parkinson’s Disease (PD), and various types of dementia, impose a substantial burden on patients and healthcare systems worldwide [1–4]. In 2019, neurological diseases as a whole surpassed cardiovascular diseases as the largest cause of global burden [1]. Neurodegenerative diseases, more specifically, was found to be the second leading cause of death among neurological disorders (after stroke), and the third leading cause of disability-adjusted life years (DALYs) (after stroke and migraine) among the adult population [1,4,5]. The incidence of neurodegenerative disorders in 2019 was 7,236.38 in thousands, while the prevalence was found to be 51,624.19 in thousands, both showing over 240% increase compared with data from 1990 [1,4,5]. With the continued growth of an aging population, the burden of neurodegenerative diseases is expected to at least further double in the next two decades [4], and at least triple in 2050 to over 150 million cases worldwide [4]. The escalating incidence and profound impact of these diseases highlight the critical need for early detection to identify individuals in the prodromal stage, where interventions hold the greatest potential for efficacy.

Structural MRI brain parameters as imaging biomarkers for neurogenerative diseases

Neuroimaging techniques have been heavily utilized, particularly in the last two decades, to screen for and monitor the progression of neurodegenerative diseases. Specifically, MRI has been one of the more popular imaging modalities employed by clinicians and researchers alike. Various MRI techniques, including structural MRI, functional MRI, diffusor tensor imaging and even neuromelanin-sensitive MRI, have been studied for potential use in diagnosing and prognosticating neurodegenerative changes [6]; however, structural MRI has remained the mainstay due to its relative ease of acquisition and strong evidence from the literature supporting its utility. Its strength lies in the ability to detect and differentiate atrophic morphological patterns in various neurodegenerative diseases. Scans of AD patients, for example, show characteristic disproportionate atrophy in the temporal lobe and medial parietal cortex [7]. Atrophic changes detected in structures of the medial temporal lobe, including the hippocampi, amygdala, cingulate cortex, parahippocampal gyrus and entorhinal cortex, have been consistently observed to be correlated with progression of AD [8–11].

Notably, hippocampal volume has been established as a reliable biomarker for AD diagnosis and severity, given ample evidence supporting its correlation with pathological findings and cognitive performance [12,13]. MRI volumetric measurements of the hippocampus were found to be correlated with neurofibrillary tangles accumulation in the same brain region [14]. In addition, several studies demonstrated that the degree of hippocampal atrophy was associated with poorer performance on memory tasks [15,16]. Patients with frontotemporal dementia (FTD), on the other hand, often demonstrate disproportionate frontal, insular and anterior temporal lobe atrophy [17]. Interestingly, previous research has found atrophic changes to be less uniform in PD patients, only observing volume loss in parahippocampal gyrus, temporal gyrus and occipital lobe in severe cases [18,19]. However, more recent large-cohort research using deformation-based volumetry has provided evidence for a widespread multi-region atrophic pattern in Parkinson’s patients that is readily detectable at the early stages of the disease [20]. Subcortical regions involved include all components of the basal ganglia, pedunculopontine nucleus, basal forebrain, hypothalamus, amygdala, hippocampus, parahippocampal gyrus, and the ventrolateral nucleus and pulvinar of thalamus. Cortical regions involved include the insula, occipital cortex Brodmann area 19, superior temporal gyrus, anterior cingulate cortex, premotor and supplementary areas, and lateral prefrontal cortex.

Besides volumetric measurements, cortical thickness measured on structural MRI has also been found to be predictive of the incidence and severity of neurodegenerative disorders. Cortical thickness at the entorhinal cortex, for instance, has been established as a sensitive biomarker for mild cognitive impairment and AD [21]. Moderate to severe PD cases have also been shown to have diffuse cortical thickness loss, especially those with disease duration of 5 years or more [18]. Interestingly, cortical thickness has also been demonstrated to reliably differentiate AD from FTD, in which AD patients consistently showed a greater degree of cortical thinning [22].

An additional imaging parameter detected on structural MRI that has strong associations with neurodegenerative diseases is white matter hyperintensities (WMHs). WMHs are one of the most common imaging features demonstrated on T2/FLAIR MRI sequences and are most commonly associated with cerebral small vessel disease. The most widely accepted pathological explanations for these lesions are chronic hypoperfusion and alterations in the blood-brain barrier [23], though other theories have been postulated, including demyelination, microglial activation, neuronal degeneration and even inflammation [24,25]. Numerous studies have shown an association between WMHs and a variety of neurodegenerative diseases, including MCI, AD, PD, Dementia with Lewy Bodies and FTD [26–32]. In fact, previous studies have shown that WMH burden is correlated with the severity of cognitive deficit in patients with neurodegenerative disease [33]. More recent studies have also lend support for WMH burden being associated with the severity of neuropsychiatric symptoms in this group of patients [34].

Though there is much evidence to support the utility of structural MRI brain parameters as biomarkers for neurodegenerative diseases, there are inherent imaging modality and patient-related factors that can affect the reliability of such brain parameters. Qualitative and quantitative assessment of MRI brain parameters, especially atrophy patterns and cortical thickness, may be affected by differences in scanner specifications and parameters, which can result in failure to detect disease progression or early changes. In addition, detection of these parameters may in some cases be only feasible with serial imaging, not just from imaging at one specific time point. Most importantly, patients with neurodegenerative diseases often have co-morbidities or multiple types of neurodegenerative diseases, causing difficulty in imaging interpretation and thus diagnosis [35,7]. White matter hyperintensities, for instance, are also present in patients with ischemic stroke. As a result, not all imaging biomarkers may be useful predictors for neurodegenerative disease incidence as a whole.

Accelerometry data from wearable devices as a diagnostic and prognostic tool for neurodegenerative diseases

Neurodegenerative diseases are characterized by neuronal loss, which leads to not only cognitive deficits but also motor impairments. AD patients may present with a slower gait, shorter stride length, and increased stride-to-stride variability [36,37]. PD, on the other hand, is typically associated with rigidity, tremors, and freezing, which can manifest in gait abnormalities such as lower walking speed, reduced step length, and impaired rhythmicity [37,38]. The advent of wearable devices containing accelerometers in the recent decade has made it increasingly accessible for researchers to track the motor and physical activity of patients across a 24-hour continuum. Initially, these wearable accelerometers were used to study physical activity levels, and large-scale studies found that self-reported as well as accelerometry-detected increase in total physical activity and less sedentary lifestyles are associated with a reduced risk of dementia and neurodegenerative diseases [39,40].

Research focus subsequently transitioned to more sophisticated data analysis, such as gait patterns, diurnal variation, and sleep, particularly in PD patients, as they often present with characteristic movement abnormalities, offering a window for early detection and tracking of disease severity [41]. Machine learning techniques have been employed to improve analysis and assess the predictive ability of accelerometry data. A small proof-of-concept study successfully employed support vector machine classifiers on wrist accelerometer data to detect PD with high accuracy (85%+/-15%), demonstrating the potential role of wearable devices in early detection [42]. Feature engineering methods, including epoch-based statistical feature engineering and the document-of-words method, were used to extract relevant predictive features from accelerometry data. A subsequent large-cohort study analysing UK Biobank data showed that wearable accelerometry features inputted into a Gaussian mixed model classifier could reliably diagnose PD, demonstrating an area under the curve (AUC) of 0.69 on gait data, 0.84 on low-movement data, and 0.85 on a fusion of both activities [43]. In addition, Schalkamp and colleagues (2023) demonstrated that accelerometry data from wearable devices, augmented by machine learning methods consisting of balanced random forests with Markov confusion matrices, were a significantly better predictor of established and prodromal PD than genetic markers, lifestyle factors, blood biochemistry or the presence of prodromal symptoms [44].

More recently, these machine learning techniques have also been used on accelerometry data collected from other neurodegenerative diseases. Accelerometry data has been employed to predict the diagnosis of both AD and PD. A large-cohort study conducted by Winer and colleagues (2024) also focused on UK Biobank data revealed that accelerometer-derived measures of activity levels, including interdaily stability, diurnal amplitude, and activity during the most active 10 hours, were predictive of both AD and PD incidence [45]. It is clear that while parameters derived from structural MRI may be reflective of cognitive deficits observed in patients with neurodegenerative diseases, patterns detected on accelerometry are more sensitive in capturing motor deficits associated with these diseases.

Study objective

No previous study has examined the combined predictive ability of structural MRI brain parameters and accelerometry data on the incidence of neurodegenerative diseases. Given the unique insights that structural MRI and accelerometry data can provide on cognitive deficits and motor impairments, respectively, both modalities should be considered useful tools for predicting and monitoring neurodegenerative diseases.

Combining accelerometry data with established structural MRI brain parameters, therefore, may offer a multimodal approach to improve early detection of neurodegenerative diseases. The primary goal of this study is to investigate whether the combination of these two modalities does improve prediction of neurodegenerative disease incidence, compared with structural MRI brain parameters alone. To achieve this, we will leverage the rich phenotypic and imaging data from the UK Biobank cohort to explore the predictive potential of this integrated approach.

Results

Demographics of study cohort

Table 1 provides an overview of the demographic characteristics of our study participants. Among the 19,793 individuals in our cohort, 56 were diagnosed with neurodegenerative diseases, while the remaining 19,737 were healthy. The mean age of the disease group was 69.9 years, significantly higher than that of the healthy group, which had a mean age of 64.1 years. Additionally, a higher proportion of men was observed in the disease group (60.7%) compared to the healthy group (45.4%).

Table 1. Participants demographic data with lifestyle factors.

	Healthy participants (n=19,737)	Disease participants (n=56)	P-values
Age (Years)	64.1 (7.8)	69.9 (6.2)	<0.0001
Sex (Female percentage %)	54.6%	39.3%	0.02
Current Smoker	63.4%	51.8%	0.113
Previous Smoker	33.6%	46.4%	0.113
Current Drinking	92.9%	83.9%	0.175
Previous Drinking	3.3%	8.9%	0.175
Nap during day (hours)	0.478 (0.604)	0.741 (0.757)	0.001
Daytime dozing (hours)	0.254 (0.485)	0.423 (0.572)	0.012

Open in a new tab

Regarding lifestyle factors, a higher percentage of current smokers was found in the healthy group (63.4%) compared to the disease group, where a larger proportion were previous smokers (46.4%). A similar trend was observed in alcohol consumption, with more healthy participants reporting current drinking habits. Notably, individuals in the disease group reported spending more time on napping and daytime dozing (0.741 hours and 0.423 hours, respectively) compared to those in the healthy group (0.478 hours and 0.253 hours, respectively).

Model performance

The comparison of model performances using the AUROC metric is listed in Table 2.

Table 2. The comparison of model performances in terms of AUROC.

Model	AUROC
Model 1 (all modalities)	0.819
Model 2 (without brain MRI)	0.688
Model 3 (without PRS)	0.758
Model 4 (without lifestyle information)	0.759

Open in a new tab

Model 1, which incorporated all modalities including MRI brain data, accelerometry data, PRS, and lifestyle information, achieved an AUC of 0.819. In comparison, Model 4, which excluded lifestyle information, attained an AUC of 0.759, indicating a decrease in predictive accuracy. Model 3, which omitted PRS information, yielded an AUC of 0.758, demonstrating a slight reduction in performance. Importantly, Model 2, excluding brain MRI data, exhibited the lowest AUC of 0.688.

The ablation studies demonstrated the varying contributions of each modality to the predictive performance. Notably, brain MRI data contributed the most to the final prediction, with its removal resulting in a significant decrease of 0.131 in AUC compared to Model 1. Conversely, the exclusion of PRS and lifestyle information led to relatively smaller reductions in AUC, decreasing by 0.06 and 0.061, respectively.

Feature importance

Detailed analysis of the contribution of the various modalities provided further insight into the predictive power of each feature. We reported feature importance scores calculated using the XGBoost gain-based algorithm. Gain is defined as the improvement in accuracy brought by a feature in the model. Specifically, it measures how much a feature contributes to reducing the loss function when used for a split in the tree. A higher gain value indicates that the feature plays a more important role in improving the model’s performance.

Fig 1 summarizes the top 20 important features identified by Model 1. Remarkably, 18 out of the 20 features were MRI-related, consisting of the grey matter volumes of various brain regions, as well as the total amount of white matter hyperintensities. In addition to these MRI features, two other features were based on accelerometry data: the average acceleration recorded by activity tracking during the time periods of 5 to 6 pm and 6 to 7 pm.

Discussion

This study aimed to investigate the predictive ability of combining accelerometry data from wearable devices with MRI-based brain parameters in predicting the incidence of neurodegenerative diseases. Our results show that a statistical model (Model 1) that combined MRI brain features, accelerometry data from wearable devices, polygenic risk scores of neurodegenerative diseases, and lifestyle factors predicted more reliably the incidence of neurodegenerative diseases in the UK Biobank cohort, compared with other statistical models (Model 2-4) that excluded either MRI brain data, PRS or lifestyle information. This finding not only reaffirms the importance of MRI brain parameters in predicting neurodegenerative diseases but also provides support for the combination of MRI brain parameters and accelerometry data from wearable devices in improving the early detection of the onset of neurodegenerative diseases.

To be specific, 18 out of the 20 most important predictive features for neurodegenerative diseases as derived from Model 1 were MRI brain parameters. They include total white matter and total periventricular white matter hyperintensities, as well as the grey matter volumes of bilateral hippocampi, left temporal fusiform cortex, bilateral ventral striatum, left posterior cingulate gyrus, left cuneal cortex, left pallidum, left amygdala, left middle temporal gyrus, right paracingulate gyrus, bilateral lateral occipital cortices/left occipital pole and right frontal medial cortex. It was anticipated that the majority of identified predictive features in the model were MRI brain parameters, given that MRI features have long been considered key diagnostic and prognostic biomarkers for neurodegenerative diseases. The improved predictive ability of the statistical model that incorporated both MRI brain parameters and accelerometry data suggests clear added value that accelerometry data brings to the detection of neurodegenerative diseases.

The identified brain regions that contributed significantly to the predictive ability of the model are consistent with previous literature on neurodegenerative diseases. Total periventricular white matter hyperintensities have been consistently linked with small vessel disease, which is found in most neurodegenerative diseases [27–29]. Multiple studies have reported significantly increased volume of white matter hyperintensities in patients with AD and PD [27–33].

As discussed, volume loss of bilateral hippocampi is well-established as a biomarker for mild cognitive impairment and AD [12–16]. An MRI volumetric study, in fact, demonstrated focal atrophy of the CA1 subfield in the early (predementia or even preclinical) stages of AD, before widespread atrophy of the whole hippocampus begins at the dementia stage [46]. In PD, the severity of hippocampal atrophy is correlated with the severity of cognitive impairment [47,48]. Volume loss in the temporal fusiform cortex, which has been implicated in the formation and storage of semantic information, is primarily associated with semantic dementia rather than AD or PD [49,50]. Volume reduction of the ventral striatum is a recognized biomarker for PD, as it is well established that loss of dopaminergic neurons in the striatal network forms the pathological basis of the disease [51]. Multiple studies have also reported significant atrophy in striatal structures in AD patients, potentially secondary to the degeneration of connected structures, such as the hippocampus [52,53].

A few studies have demonstrated atrophy of the posterior cingulate gyrus in AD and PD patients, though the underlying mechanism remains unclear [54–57]. Some have attributed the finding to the fact that the region is part of the default mode network, which is often impaired in the case of neurodegenerative diseases [55]. The cuneal cortex, known to be involved in cognitive and emotional coordination, has been shown to demonstrate early cortical atrophy in PD patients with cognitive impairment [58–60]. Volume loss of the pallidum, whose key subregions include the thalamus and globus pallidus, has also been associated with neurodegenerative disease progression. A study showed thalamic volume loss as one of the first signs of cognitive decline during early mild cognitive impairment, though no volume loss is observed during further progression to AD [61]. Other studies have also demonstrated thalamic volume loss in the early stage of PD [62,63], which could be attributed to the region’s strong connectivity with the striatal region [63].

The amygdala, regarded as the center for emotional information processing, shows significant volume loss in PD patients, especially in those with depressive and anxiety symptoms [64]. Some studies have also reported amygdala volume loss in AD patients [65,66]; atrophy of the amygdala could also be due to the regions’ strong connectivity with the hippocampus. The middle temporal gyrus, an important site for memory processing [67–70], has been shown to be one of the first neocortical regions in the temporal lobe to show volume loss in AD patients [67,68]. The paracingulate gyrus, responsible for mentalizing, inhibitory control, and guiding motor actions, exhibits prominent atrophy in frontotemporal dementia patients [70,71].

The occipital cortex, specifically the lateral occipital cortex, known to be involved in object recognition and visuospatial processing, has also been shown to undergo a reduction in volume among PD patients [72]. The frontal medial cortex has also been widely implicated in the progression of various neurodegenerative diseases. Overall frontal lobe volume reduction has been shown in late-stage PD by multiple studies [57,73–78].

Our study is the first to demonstrate that accelerometry data from wearable devices can serve as a significant predictor of neurodegenerative disorders as a whole. Previous studies have primarily focused on the predictive ability of accelerometry data in tracking and predicting PD. Winer and colleagues (2024) demonstrated that 24-hour rhythm integrity measured by wearable devices was an associated risk of AD, PD and cognitive decline [45]; however, it did not explore its association with all neurodegenerative diseases. As discussed, a likely explanation for the predictive ability of accelerometry data on all neurodegenerative diseases is the universal effect of neurodegenerative diseases on gait, which could be indirectly reflected in accelerometry [79]. Our study hence highlights the broad potential applicability of wearable device data in not only tracking but also predicting a wide range of neurodegenerative diseases.

Despite the significant findings of this study, several limitations should be acknowledged. Firstly, this study is an association study and cannot establish a causal relationship between MRI brain parameters and accelerometry data with the incidence of neurodegenerative diseases. Secondly, the UK Biobank cohort used in this study contains only a small number of positive cases for neurodegenerative diseases, resulting in an imbalanced dataset that could introduce statistical bias in the analysis. Thirdly, the study was performed on data from the UK Biobank cohort with a mean follow-up period of 5.9 years. A longer period of follow-up is always desirable as well as multiple time points longitudinal tracking, albeit may not be feasible at large scale. Finally, future prospective studies will be helpful, particularly focusing on specific neurodegenerative diseases (e.g. Alzheimer’s or Parkinson’s disease, for example) to also reduce the impact of different disease heterogeneity. Testing in different populations would also ensure the reproducibility of these findings. Nevertheless, further research in this area holds promise for improving the early detection, diagnosis and management of these debilitating conditions.

In conclusion, this study provides evidence for the strength of combining MRI brain parameters and accelerometry data from wearable devices to predict the incidence of neurodegenerative diseases. Using a cohort of 19,793 participants from the UK Biobank, free of neurodegenerative disease at baseline and followed for an average of 5.9 years, we employed the XGBoost machine learning algorithm to develop prediction models. The comprehensive model, incorporating MRI parameters, accelerometry data, polygenic risk scores, and lifestyle information, achieved the highest AUC value of 0.819. In contrast, the model excluding MRI parameters had the lowest AUC value of 0.688. Feature importance analysis highlighted that 18 of the top 20 predictors were MRI brain parameters, while 2 were from accelerometry data, emphasising the critical role of MRI parameters in predicting neurodegenerative diseases. Future studies should explore the predictive ability of other MRI brain parameters, such as cortical thickness and functional connectivity, in conjunction with accelerometry data from wearable devices. Additionally, studying cohorts with more time points and longer follow-up would allow for a better assessment of the temporal association between MRI brain parameters and accelerometry data with neurodegenerative disease incidence. It would also be valuable to investigate the differences in predictive ability for various types of neurodegenerative diseases, such as AD, PD and dementia.

Methods

Ethical approval

Ethical approval was obtained for this study from the UK Biobank study (UK North West Multi-Centre Research Ethics Committee under reference 11/NW/0382) and the University of Hong Kong (UW-20814). Written informed consent was secured from all individuals participating in the study. This research was conducted using the UK Biobank Resource Application Number 78730.

Participants and study design

This study utilized data from the UK Biobank, a comprehensive dataset containing biological and medical data from over 502,000 participants [80]. Our initial cohort consisted of 48,457 participants, who had both T1 and T2 weighted brain MRI data available. From this initial pool, exclusions were made for individuals lacking accelerometry data (N=28,560), those without an official standard PRS provided by the UK Biobank (N=30), and those with missing lifestyle information including smoking habit, alcohol consumption, duration of daytime napping, and duration of daytime dozing (N=74). These exclusions yielded a final sample size of 19,793 participants, as illustrated in Fig 2.

To identify individuals with neurodegenerative diseases (AD, PD and dementia), we employed algorithms developed by the UK Biobank Outcome. Cases were defined as individuals diagnosed with any of these diseases at least one year after their image screening. The mean follow-up time for our cohort is 5.9 years. This process identified 56 positive cases, comprising 19 with PD, 13 with AD, and 40 with dementia. Notably, 10 participants were diagnosed with all three diseases, while 27 exhibited two of the specified diseases.

Brain MRI pre-processing

The UK Biobank’s MRI brain data were acquired from three imaging centers (Cheadle, Newcastle, and Reading) using 3T Siemens Skyra scanners.

The imaging time period for our participants was between 05/2014 and 03/2020. We used the processed brain images as our input data. Specifically, for T1-weighted brain MRI analysis, we utilized the 139 regional grey matter volumes (GMV) generated by FSL FAST.

For T2-weighted MRI analysis, we focused on 5 features related to white matter hyperintensities (WMH) volumes. T2-weighted MRI is a valuable tool for detecting WMH in neurodegenerative diseases. WMH appears as areas of increased signal intensity on T2-weighted and fluid-attenuated inversion recovery (FLAIR) MRI scans.

For reproducibility and comparability, all segment volumes’ information was generated by the UK Biobank using field IDs 1101 and 112. Detailed information can be found in S1 and S2 Tables.

Polygenic risk score data

Polygenic Risk Scores (PRS) are increasingly recognized for their potential in understanding and predicting neurodegenerative diseases. A PRS quantifies an individual’s genetic predisposition to a disease based on the cumulative effect of numerous single nucleotide polymorphisms (SNPs) identified through genome-wide association studies (GWAS). In the context of neurodegenerative diseases, such as AD and PD, PRS can correlate with disease risk and progression, providing insights into genetic liability. For reproducibility and comparability, all PRS scores were calculated by the UK Biobank using field IDs 301. Detailed information can be found in S3 Table.

Accelerometry data

To investigate the association between physical activity patterns and neurodegenerative disease incidence, we focused on participants’ accelerometry measurements. Measurements were collected via wrist-worn accelerometers, with the primary data collection period spanning from June 2013 to January 2016. Derived accelerometry data provided key metrics, including the average proportion of time spent engaging in light activity, moderate-vigorous activity, sedentary behavior, and sleep per day. Also, average accelerations measured during the activity monitor’s wearing period were used to further characterize participants’ physical activity levels. Please refer to S4 and S5 Tables for details.

Lifestyle data

Neurodegenerative disorders are greatly affected by factors related to lifestyle, such as smoking, alcohol consumption, and sleep patterns. Studies suggest that people who adhere to a healthy lifestyle, which includes abstaining from smoking, consuming alcohol in moderation, engaging in regular physical activity, and getting sufficient sleep, have a reduced likelihood of developing diseases. We included smoking status, drinking status and frequencies, nap time during day, and daytime dozing/sleeping and lifestyle factors.

Input modalities

This study employed 4 modalities: (1) MRI: We utilized T1-weighted brain MRI segments to assess 139 regional grey matter volumes (GMVs), and 5 features related to white matter hyperintensities (WMH) volumes generated from T2-weighted MRI to focus. (2) Polygenic Risk Scores (PRS): We used 39 common diseases’ PRS scores. (3) Accelerometry Data: We utilized data reflecting the average proportion of time spent in various activities, including light activity, moderate-to-vigorous activity, sedentary behavior, and sleep per day, as well as average accelerations. (4) Lifestyle Information: Smoking status, drinking habits, and sleep patterns were included as input.

Classification performance

To assess the predictive power of disease incidence, we devised an experimental setup consisting of four models:

Model 1: Utilizing all modalities of interest, including MRI brain data, accelerometry data, PRS, and lifestyle information as input.
Model 2: Exclusively using accelerometry data, PRS, and lifestyle information.
Model 3: Utilizing MRI brain data, accelerometry data, and lifestyle information.
Model 4: Utilizing MRI brain data, accelerometry data and PRS.

XGBoost, an advanced machine learning algorithm, has shown promising results in medical classification tasks. It outperforms traditional methods like logistic regression in predicting myocardial infarction, achieving higher ROC scores [81]. XGBoost demonstrates robustness in handling missing data, maintaining high F1-scores even with 90% missing values in breast cancer and heart failure datasets [82]. The algorithm excels at managing complex, high-dimensional data, as evidenced by its superior performance in cancer stage prediction using multi-omics data [83]. Importantly, XGBoost offers interpretability through feature importance ranking and SHAP values, bridging the gap between medicine and data science [83,84]. Its ability to handle data complexities and provide interpretable results makes XGBoost a valuable tool for medical practitioners in diagnosis and treatment decision-making [84].

To evaluate the model’s performance, the dataset was randomly split into 80% for training and 20% for testing. The model was trained and tuned using the training dataset with the strategy of five-folder-cross-validation. We applied gride-search methods to find the best hyper-parameters for each model, includeding: “subsample”, “reg_lambda”, “reg_alpha”, “n_estimators”, “min_child_weight”, “max_depth”, “gamma”, “eta” and “colsample_bytree”.

We applied the Area Under the Receiver Operating Characteristic (AUROC) curve as our evaluation metric. AUROC measures a model’s ability to distinguish between classes across all possible classification thresholds. Given the significant class imbalance in this dataset, using AUROC for unbalanced classification tasks in disease classification offers a robust, comprehensive, and interpretable assessment of model performance. The results can be found in S6 Table.

Statistical analysis

Correlation analyses between the disease group and healthy controls were conducted using T-tests. All statistical analyses were performed using Python (version 3.10), with significance set at a two-sided level of 5%. Performance metrics were presented with 95% confidence intervals to accurately reflect the variability in the data and the precision of our estimates.

Supporting information

S1 Table. T1 Brain MRI features and the related field IDs in UK Biobank.

(DOCX)

pdig.0000795.s001.docx^{(21.8KB, docx)}

S2 Table. T2 Brain MRI features and the related field IDs in UK Biobank.

(DOCX)

pdig.0000795.s002.docx^{(14.7KB, docx)}

S3 Table. RPS scores and the related field IDs in UK Biobank.

(DOCX)

pdig.0000795.s003.docx^{(17.7KB, docx)}

S4 Table. Accelerometry average data and the related field IDs in UK Biobank.

(DOCX)

pdig.0000795.s004.docx^{(16.8KB, docx)}

S5 Table. Accelerometry-derived data and the related field IDs in UK Biobank.

(DOCX)

pdig.0000795.s005.docx^{(15.8KB, docx)}

S6 Table. Comparison with Common Models.

(DOCX)

pdig.0000795.s006.docx^{(14.7KB, docx)}

Data Availability

The data used for this work is from the UK Biobank Study, which can be accessed at https://www.ukbiobank.ac.uk/.

Funding Statement

The author(s) received no specific funding for this work.

References

1.Feigin VL, Vos T, Nichols E, Owolabi MO, Carroll WM, Dichgans M, et al. The global burden of neurological disorders: translating evidence into policy. Lancet Neurol. 2020;19(3):255–65. doi: 10.1016/S1474-4422(19)30411-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
2.Van Schependom J, D’haeseleer M. Advances in Neurodegenerative Diseases. J Clin Med. 2023;12(5):1709. doi: 10.3390/jcm12051709 [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Zahra W, Rai SN, Birla H, Singh SS, Dilnashin H, Rathore AS, et al. The global economic impact of neurodegenerative diseases: Opportunities and challenges. Bioeconomy for sustainable development. 2020:333–45. [Google Scholar]
4.Nichols E, Steinmetz J, Vollset S, Fukutaki K, Chalek J, Abd-Allah F. Estimation of the global prevalence of dementia in 2019 and forecasted prevalence in 2050: an analysis for the Global Burden of Disease Study 2019. The Lancet Public Health. 2022;7(2):e105–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Ding C, Wu Y, Chen X, Chen Y, Wu Z, Lin Z. Global, regional, and national burden and attributable risk factors of neurological disorders: The Global Burden of Disease study 1990–2019. Frontiers in Public Health. 2022;10:952161. doi: 10.3389/fpubh.2022.952161 [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Ehrenberg AJ, Khatun A, Coomans E, Betts MJ, Capraro F, Thijssen EH, et al. Relevance of biomarkers across different neurodegenerative diseases. Alzheimers Res Ther. 2020;12(1):56. doi: 10.1186/s13195-020-00601-w [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Young PNE, Estarellas M, Coomans E, Srikrishna M, Beaumont H, Maass A, et al. Imaging biomarkers in neurodegeneration: current and future practices. Alzheimers Res Ther. 2020;12(1):49. doi: 10.1186/s13195-020-00612-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Henley SMD, Bates GP, Tabrizi SJ. Biomarkers for neurodegenerative diseases. Curr Opin Neurol. 2005;18(6):698–705. doi: 10.1097/01.wco.0000186842.51129.cb [DOI] [PubMed] [Google Scholar]
9.Koikkalainen J, Rhodius-Meester H, Tolonen A, Barkhof F, Tijms B, Lemstra A. Differential diagnosis of neurodegenerative diseases using structural MRI data. NeuroImage: Clinical. 2016;11:435–49. [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Agosta F, Galantucci S, Filippi M. Advanced magnetic resonance imaging of neurodegenerative diseases. Neurol Sci. 2017;38(1):41–51. doi: 10.1007/s10072-016-2764-x [DOI] [PubMed] [Google Scholar]
11.Hansson O. Biomarkers for neurodegenerative diseases. Nat Med. 2021;27(6):954–63. doi: 10.1038/s41591-021-01382-x [DOI] [PubMed] [Google Scholar]
12.Frisoni GB, Jack CR Jr, Bocchetta M, Bauer C, Frederiksen KS, Liu Y, et al. The EADC-ADNI Harmonized Protocol for manual hippocampal segmentation on magnetic resonance: evidence of validity. Alzheimers Dement. 2015;11(2):111–25. doi: 10.1016/j.jalz.2014.05.1756 [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Risacher SL, Shen L, West JD, Kim S, McDonald BC, Beckett LA, et al. Longitudinal MRI atrophy biomarkers: relationship to conversion in the ADNI cohort. Neurobiol Aging. 2010;31(8):1401–18. doi: 10.1016/j.neurobiolaging.2010.04.029 [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Whitwell J, Shiung M, Przybelski S, Weigand S, Knopman D, Boeve B. MRI patterns of atrophy associated with progression to AD in amnestic mild cognitive impairment. Neurology. 2008;70(7):512–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Wei H, Kong M, Zhang C, Guan L, Ba M, for Alzheimer’s Disease Neuroimaging Initiative*. The structural MRI markers and cognitive decline in prodromal Alzheimer’s disease: a 2-year longitudinal study. Quant Imaging Med Surg. 2018;8(10):1004–19. doi: 10.21037/qims.2018.10.08 [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Risacher S, Anderson W, Charil A, Castelluccio P, Shcherbinin S, Saykin A. Alzheimer disease brain atrophy subtypes are associated with cognition and rate of decline. Neurology. 2017;89(21):2176–86. [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Rascovsky K, Hodges JR, Knopman D, Mendez MF, Kramer JH, Neuhaus J, et al. Sensitivity of revised diagnostic criteria for the behavioural variant of frontotemporal dementia. Brain. 2011;134(Pt 9):2456–77. doi: 10.1093/brain/awr179 [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Hanganu A, Bedetti C, Degroot C, Mejia-Constain B, Lafontaine A-L, Soland V, et al. Mild cognitive impairment is linked with faster rate of cortical thinning in patients with Parkinson’s disease longitudinally. Brain. 2014;137(Pt 4):1120–9. doi: 10.1093/brain/awu036 [DOI] [PubMed] [Google Scholar]
19.Mitchell T, Lehéricy S, Chiu SY, Strafella AP, Stoessl AJ, Vaillancourt DE. Emerging Neuroimaging Biomarkers Across Disease Stage in Parkinson Disease: A Review. JAMA Neurol. 2021;78(10):1262–72. doi: 10.1001/jamaneurol.2021.1312 [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Zeighami Y, Ulla M, Iturria-Medina Y, Dadar M, Zhang Y, Larcher K. Network structure of brain atrophy in de novo Parkinson’s disease. eLife. 2015;4:e08440. [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Velayudhan L, Proitsi P, Westman E, Muehlboeck J, Mecocci P, Vellas B, et al. Entorhinal cortex thickness predicts cognitive decline in Alzheimer’s disease. Journal of Alzheimer’s Disease. 2013;33(3):755–66. [DOI] [PubMed] [Google Scholar]
22.Du A-T, Schuff N, Kramer JH, Rosen HJ, Gorno-Tempini ML, Rankin K, et al. Different regional patterns of cortical thinning in Alzheimer’s disease and frontotemporal dementia. Brain. 2007;130(Pt 4):1159–66. doi: 10.1093/brain/awm016 [DOI] [PMC free article] [PubMed] [Google Scholar]
23.McAleese KE, Alafuzoff I, Charidimou A, De Reuck J, Grinberg LT, Hainsworth AH, et al. Post-mortem assessment in vascular dementia: advances and aspirations. BMC Med. 2016;14(1):129. doi: 10.1186/s12916-016-0676-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Abraham HMA, Wolfson L, Moscufo N, Guttmann CRG, Kaplan RF, White WB. Cardiovascular risk factors and small vessel disease of the brain: Blood pressure, white matter lesions, and functional decline in older persons. J Cereb Blood Flow Metab. 2016;36(1):132–42. doi: 10.1038/jcbfm.2015.121 [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Richard E, Gouw AA, Scheltens P, van Gool WA. Vascular care in patients with Alzheimer disease with cerebrovascular lesions slows progression of white matter lesions on MRI: the evaluation of vascular care in Alzheimer’s disease (EVA) study. Stroke. 2010;41(3):554–6. doi: 10.1161/STROKEAHA.109.571281 [DOI] [PubMed] [Google Scholar]
26.Prins ND, Scheltens P. White matter hyperintensities, cognitive impairment and dementia: an update. Nat Rev Neurol. 2015;11(3):157–65. doi: 10.1038/nrneurol.2015.10 [DOI] [PubMed] [Google Scholar]
27.Gaubert M, Lange C, Garnier-Crussard A, Köbe T, Bougacha S, Gonneaud J, et al. Topographic patterns of white matter hyperintensities are associated with multimodal neuroimaging biomarkers of Alzheimer’s disease. Alzheimers Res Ther. 2021;13(1):29. doi: 10.1186/s13195-020-00759-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Lee S, Viqar F, Zimmerman ME, Narkhede A, Tosto G, Benzinger TLS, et al. White matter hyperintensities are a core feature of Alzheimer’s disease: Evidence from the dominantly inherited Alzheimer network. Ann Neurol. 2016;79(6):929–39. doi: 10.1002/ana.24647 [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Yoshita M, Fletcher E, Harvey D, Ortega M, Martinez O, Mungas DM, et al. Extent and distribution of white matter hyperintensities in normal aging, MCI, and AD. Neurology. 2006;67(12):2192–8. doi: 10.1212/01.wnl.0000249119.95747.1f [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Damulina A, Pirpamer L, Seiler S, Benke T, Dal-Bianco P, Ransmayr G. White matter hyperintensities in Alzheimer’s disease: a lesion probability mapping study. Journal of Alzheimer’s Disease. 2019;68(2):789–96. [DOI] [PubMed] [Google Scholar]
31.Garnier-Crussard A, Cotton F, Krolak-Salmon P, Chételat G. White matter hyperintensities in Alzheimer’s disease: Beyond vascular contribution. Alzheimers Dement. 2023;19(8):3738–48. doi: 10.1002/alz.13057 [DOI] [PubMed] [Google Scholar]
32.Linortner P, McDaniel C, Shahid M, Levine TF, Tian L, Cholerton B, et al. White Matter Hyperintensities Related to Parkinson’s Disease Executive Function. Mov Disord Clin Pract. 2020;7(6):629–38. doi: 10.1002/mdc3.12956 [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Zou X, Dong Z, Chen X, Yu Q, Yin H, Yi L. White matter hyperintensities burden in the frontal regions is positively correlated to the freezing of gait in Parkinson’s disease. Frontiers in Aging Neuroscience. 2023;15:1156648. doi: 10.3389/fnagi.2023.1156648 [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Ozzoude M, Varriano B, Beaton D, Ramirez J, Adamo S, Holmes MF, et al. White matter hyperintensities and smaller cortical thickness are associated with neuropsychiatric symptoms in neurodegenerative and cerebrovascular diseases. Alzheimers Res Ther. 2023;15(1):114. doi: 10.1186/s13195-023-01257-y [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Chouliaras L, O’Brien JT. The use of neuroimaging techniques in the early and differential diagnosis of dementia. Mol Psychiatry. 2023;28(10):4084–97. doi: 10.1038/s41380-023-02215-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Beauchet O, Annweiler C, Dubost V, Allali G, Kressig RW, Bridenbaugh S, et al. Stops walking when talking: a predictor of falls in older adults?. Eur J Neurol. 2009;16(7):786–95. doi: 10.1111/j.1468-1331.2009.02612.x [DOI] [PubMed] [Google Scholar]
37.Cicirelli G, Impedovo D, Dentamaro V, Marani R, Pirlo G, D’Orazio TR. Human gait analysis in neurodegenerative diseases: a review. IEEE Journal of Biomedical and Health Informatics. 2021;26(1):229–42. [DOI] [PubMed] [Google Scholar]
38.Morris M, Iansek R, Matyas T, Summers J. The pathogenesis of gait hypokinesia in Parkinson’s disease. Brain. 1994;117(5):1169–81. [DOI] [PubMed] [Google Scholar]
39.del Pozo Cruz B, Ahmadi M, Naismith S, Stamatakis E. Association of daily step count and intensity with incident dementia in 78 430 adults living in the UK. JAMA Neurology. 2022;79(10):1059–63. [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Raichlen D, Aslan D, Sayre M, Bharadwaj P, Ally M, Maltagliati S. Sedentary behavior and incident dementia among older adults. JAMA. 2023;330(10):934–40. [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Del Din S, Godfrey A, Mazzà C, Lord S, Rochester L. Free‐living monitoring of Parkinson’s disease: Lessons from the field. Movement Disorders. 2016;31(9):1293–313. [DOI] [PubMed] [Google Scholar]
42.Rastegari E, Ali H, Marmelat V. Detection of Parkinson’s Disease Using Wrist Accelerometer Data and Passive Monitoring. Sensors (Basel). 2022;22(23):9122. doi: 10.3390/s22239122 [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Williamson J, Telfer B, Mullany R, Friedl K. Detecting Parkinson’s disease from wrist-worn accelerometry in the UK Biobank. Sensors. 2021;21(6):2047. [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Schalkamp A-K, Peall KJ, Harrison NA, Sandor C. Wearable movement-tracking data identify Parkinson’s disease years before clinical diagnosis. Nat Med. 2023;29(8):2048–56. doi: 10.1038/s41591-023-02440-2 [DOI] [PubMed] [Google Scholar]
45.Winer JR, Lok R, Weed L, He Z, Poston KL, Mormino EC, et al. Impaired 24-h activity patterns are associated with an increased risk of Alzheimer’s disease, Parkinson’s disease, and cognitive decline. Alzheimers Res Ther. 2024;16(1):35. doi: 10.1186/s13195-024-01411-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
46.De Flores R, La Joie R, Chételat G. Structural imaging of hippocampal subfields in healthy aging and Alzheimer’s disease. Neuroscience. 2015;309:29–50. [DOI] [PubMed] [Google Scholar]
47.Xu R, Hu X, Jiang X, Zhang Y, Wang J, Zeng X. Longitudinal volume changes of hippocampal subfields and cognitive decline in Parkinson’s disease. Quant Imaging Med Surg. 2020;10(1):220–32. doi: 10.21037/qims.2019.10.17 [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Calabresi P, Castrioto A, Di Filippo M, Picconi B. New experimental and clinical links between the hippocampus and the dopaminergic system in Parkinson’s disease. Lancet Neurol. 2013;12(8):811–21. doi: 10.1016/S1474-4422(13)70118-2 [DOI] [PubMed] [Google Scholar]
49.Galton C, Patterson K, Graham K, Lambon-Ralph M, Williams G, Antoun N. Differing patterns of temporal atrophy in Alzheimer’s disease and semantic dementia. Neurology. 2001;57(2):216–25. [DOI] [PubMed] [Google Scholar]
50.Chapleau M, Aldebert J, Montembeault M, Brambati S. Atrophy in Alzheimer’s disease and semantic dementia: an ALE meta-analysis of voxel-based morphometry studies. Journal of Alzheimer’s disease. 2016;54(3):941–55. [DOI] [PubMed] [Google Scholar]
51.Lee HM, Kwon K-Y, Kim M-J, Jang J-W, Suh S-I, Koh S-B, et al. Subcortical grey matter changes in untreated, early stage Parkinson’s disease without dementia. Parkinsonism Relat Disord. 2014;20(6):622–6. doi: 10.1016/j.parkreldis.2014.03.009 [DOI] [PubMed] [Google Scholar]
52.Pini L, Pievani M, Bocchetta M, Altomare D, Bosco P, Cavedo E. Brain atrophy in Alzheimer’s disease and aging. Ageing research reviews. 2016;30:25–48. [DOI] [PubMed] [Google Scholar]
53.Slachevsky A, Forno G, Barraza P, Mioshi E, Delgado C, Lillo P. Mapping the neuroanatomy of functional decline in Alzheimer’s disease from basic to advanced activities of daily living. Journal of Neurology. 2019;266(6):1310–22. [DOI] [PubMed] [Google Scholar]
54.Lee P, Chou K, Chung C, Lai T, Zhou J, Wang P. Posterior cingulate cortex network predicts Alzheimer’s disease progression. Frontiers in Aging Neuroscience. 2020;12:608667. doi: 10.3389/fnagi.2020.608667 [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Leech R, Sharp DJ. The role of the posterior cingulate cortex in cognition and disease. Brain. 2014;137(Pt 1):12–32. doi: 10.1093/brain/awt162 [DOI] [PMC free article] [PubMed] [Google Scholar]
56.Jones BF, Barnes J, Uylings HBM, Fox NC, Frost C, Witter MP, et al. Differential regional atrophy of the cingulate gyrus in Alzheimer disease: a volumetric MRI study. Cereb Cortex. 2006;16(12):1701–8. doi: 10.1093/cercor/bhj105 [DOI] [PubMed] [Google Scholar]
57.de Schipper L, van der Grond J, Marinus J, Henselmans J, van Hilten J. Loss of integrity and atrophy in cingulate structural covariance networks in Parkinson’s disease. NeuroImage: Clinical. 2017;15:587–93. [DOI] [PMC free article] [PubMed] [Google Scholar]
58.Zhornitsky S, Chaudhary S, Le T, Chen Y, Zhang S, Potvin S. Cognitive dysfunction and cerebral volumetric deficits in individuals with Alzheimer’s disease, alcohol use disorder, and dual diagnosis. Psychiatry Research: Neuroimaging. 2021;317:111380. doi: 10.1016/j.pscychresns.2021.111380 [DOI] [PMC free article] [PubMed] [Google Scholar]
59.Liu C, Jiang J, Zhou H, Zhang H, Wang M, Jiang J. Brain functional and structural signatures in Parkinson’s disease. Frontiers in Aging Neuroscience. 2020;12:125. [DOI] [PMC free article] [PubMed] [Google Scholar]
60.Li L, Ji B, Zhao T, Cui X, Chen J, Wang Z. The structural changes of gray matter in Parkinson disease patients with mild cognitive impairments. PLoS One. 2022;17(7):e0269787. doi: 10.1371/journal.pone.0269787 [DOI] [PMC free article] [PubMed] [Google Scholar]
61.van de Mortel LA, Thomas RM, van Wingen GA. Grey matter loss at different stages of cognitive decline: a role for the thalamus in developing Alzheimer’s disease. Journal of Alzheimer’s Disease. 2021;83(2):705–20. doi: 10.3233/JAD-201123 [DOI] [PMC free article] [PubMed] [Google Scholar]
62.Fu T, Klietz M, Nösel P, Wegner F, Schrader C, Höglinger GU, et al. Brain Morphological Alterations Are Detected in Early-Stage Parkinson’s Disease with MRI Morphometry. J Neuroimaging. 2020;30(6):786–92. doi: 10.1111/jon.12769 [DOI] [PubMed] [Google Scholar]
63.Lee SH, Kim SS, Tae WS, Lee SY, Choi JW, Koh SB, et al. Regional volume analysis of the Parkinson disease brain in early disease stage: gray matter, white matter, striatum, and thalamus. AJNR Am J Neuroradiol. 2011;32(4):682–7. doi: 10.3174/ajnr.A2372 [DOI] [PMC free article] [PubMed] [Google Scholar]
64.Tavel P, Nikolai T. Amygdala and emotionality in Parkinson’s disease: An integrative review of the neuropsychological evidence. Neuroendocrinology Letters. 2018;39(2). [PubMed] [Google Scholar]
65.Lupton M, Strike L, Hansell N, Wen W, Mather K, Armstrong N. The effect of increased genetic risk for Alzheimer’s disease on hippocampal and amygdala volume. Neurobiology of Aging. 2016;40(1). [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Miller M, Younes L, Ratnanather J, Brown T, Trinh H, Lee D. Amygdalar atrophy in symptomatic Alzheimer’s disease based on diffeomorphometry: the BIOCARD cohort. Neurobiology of Aging. 2015;36:S3–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
67.Convit A, de Asis J, de Leon MJ, Tarshish CY, De Santi S, Rusinek H. Atrophy of the medial occipitotemporal, inferior, and middle temporal gyri in non-demented elderly predict decline to Alzheimer’s disease. Neurobiol Aging. 2000;21(1):19–26. doi: 10.1016/s0197-4580(99)00107-4 [DOI] [PubMed] [Google Scholar]
68.Busatto G, Garrido G, Almeida O, Castro C, Camargo C, Cid C. A voxel-based morphometry study of temporal lobe gray matter reductions in Alzheimer’s disease. Neurobiology of Aging. 2003;24(2):221–31. doi: 10.1016/S0197-4580(02)00267-0 [DOI] [PubMed] [Google Scholar]
69.Guo X, Wang Z, Li K, Li Z, Qi Z, Jin Z, et al. Voxel-based assessment of gray and white matter volumes in Alzheimer’s disease. Neurosci Lett. 2010;468(2):146–50. doi: 10.1016/j.neulet.2009.10.086 [DOI] [PMC free article] [PubMed] [Google Scholar]
70.Hafkemeijer A, Möller C, Dopper EGP, Jiskoot LC, van den Berg-Huysmans AA, van Swieten JC, et al. Differences in structural covariance brain networks between behavioral variant frontotemporal dementia and Alzheimer’s disease. Hum Brain Mapp. 2016;37(3):978–88. doi: 10.1002/hbm.23081 [DOI] [PMC free article] [PubMed] [Google Scholar]
71.Irish M, Piguet O, Hodges JR, Hornberger M. Common and unique gray matter correlates of episodic memory dysfunction in frontotemporal dementia and Alzheimer’s disease. Hum Brain Mapp. 2014;35(4):1422–35. doi: 10.1002/hbm.22263 [DOI] [PMC free article] [PubMed] [Google Scholar]
72.Filoteo JV, Reed JD, Litvan I, Harrington DL. Volumetric correlates of cognitive functioning in nondemented patients with Parkinson’s disease. Mov Disord. 2014;29(3):360–7. doi: 10.1002/mds.25633 [DOI] [PMC free article] [PubMed] [Google Scholar]
73.Kobayakawa M, Tsuruya N, Kawamura M. Decision-making performance in Parkinson’s disease correlates with lateral orbitofrontal volume. Journal of the neurological sciences. 2017;372(1):232–8. [DOI] [PubMed] [Google Scholar]
74.Cerasa A, Salsone M, Nigro S, Chiriaco C, Donzuso G, Bosco D. Cortical volume and folding abnormalities in Parkinson’s disease patients with pathological gambling. Parkinsonism & related disorders. 2014;20(11):1209–14. [DOI] [PubMed] [Google Scholar]
75.Owen AM, Doyon D. The cognitive neuropsychology of Parkinson’s disease: a functional neuroimaging perspective. Advances in Neurology. 1999;80(1):49–56. [PubMed] [Google Scholar]
76.Burton EJ, McKeith IG, Burn DJ, Williams ED, O’Brien JT. Cerebral atrophy in Parkinson’s disease with and without dementia: a comparison with Alzheimer’s disease, dementia with Lewy bodies and controls. Brain. 2004;127(Pt 4):791–800. doi: 10.1093/brain/awh088 [DOI] [PubMed] [Google Scholar]
77.Beyer M, Janvin C, Larsen J, Aarsland D. A magnetic resonance imaging study of patients with Parkinson’s disease with mild cognitive impairment and dementia using voxel-based morphometry. Journal of Neurology, Neurosurgery & Psychiatry. 2007;78(3):254–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
78.Nagano-Saito A, Washimi Y, Arahata Y, Kachi T, Lerch J, Evans A. Cerebral atrophy and its relation to cognitive impairment in Parkinson disease. Neurology. 2005;64(2):224–9. [DOI] [PubMed] [Google Scholar]
79.Zhao H, Wang R, Qi D, Xie J, Cao J, Liao WH. Wearable gait monitoring for diagnosis of neurodegenerative diseases. Measurement. 2022. Oct 1;202:111839. [Google Scholar]
80.Littlejohns T, Holliday J, Gibson L, Garratt S, Oesingmann N, Alfaro-Almagro F, et al. The UK Biobank imaging enhancement of 100,000 participants: rationale, data collection, management and future directions. Nature Communications. 2020;11(1):2624. doi: 10.1038/s41467-020-16445-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
81.Moore A, Bell M. XGBoost, A Novel Explainable AI Technique, in the Prediction of Myocardial Infarction: A UK Biobank Cohort Study. Clin Med Insights Cardiol. 2022;16:11795468221133611. doi: 10.1177/11795468221133611 [DOI] [PMC free article] [PubMed] [Google Scholar]
82.Stokanović S, Đukić D, Miljković N. Robustness of XGBoost Algorithm to Missing Features for Binary Classification of Medical Data. Proceedings of the 23rd International Symposium INFOTEH-JAHORINA (INFOTEH). 2024:1–6. [Google Scholar]
83.Ma B, Meng F, Yan G, Yan H, Chai B, Song F. Diagnostic classification of cancers using extreme gradient boosting algorithm and multi-omics data. Comput Biol Med. 2020;121:103761. doi: 10.1016/j.compbiomed.2020.103761 [DOI] [PubMed] [Google Scholar]
84.Price J, Yamazaki T, Fujihara K, Sone H. XGBoost: interpretable machine learning approach in medicine. In2022 5th World Symposium on Communication Engineering (WSCE) 2022. Sep 16 (pp. 109-113). IEEE. [Google Scholar]

PLOS Digit Health. 2025 Apr 25;4(4):e0000795. doi: 10.1371/journal.pdig.0000795.r001

Author response to Decision Letter 0

30 Apr 2024

PLOS Digit Health. doi: 10.1371/journal.pdig.0000795.r002

Decision Letter 0

Martin G Frasch, Md Mehedi Hassan

2 Aug 2024

PDIG-D-24-00170

Multi-modal Machine Learning Approach for Early Detection of Neurodegenerative Diseases Leveraging Brain MRI and Wearable Sensor Data

PLOS Digital Health

Dear Dr. Vardhanabhuti,

Thank you for submitting your manuscript to PLOS Digital Health. After careful consideration, we feel that it has merit but does not fully meet PLOS Digital Health's publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript within 60 days Oct 01 2024 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at digitalhealth@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pdig/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

* A rebuttal letter that responds to each point raised by the editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

* A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

* An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

We look forward to receiving your revised manuscript.

Kind regards,

Md. Mehedi Hassan

Academic Editor

PLOS Digital Health

Additional Editor Comments (if provided):

Authors need to focus on paper presentations, ensuring the outcomes are clearly rewritten to highlight the proper findings from this study.

I recommend addressing all reviewers' comments thoroughly and resubmitting the paper to the journal.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Does this manuscript meet PLOS Digital Health’s publication criteria ? Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe methodologically and ethically rigorous research with conclusions that are appropriately drawn based on the data presented.

Reviewer #1: No

Reviewer #2: Yes

Reviewer #3: Partly

Reviewer #4: No

--------------------

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

Reviewer #4: No

--------------------

3. Have the authors made all data underlying the findings in their manuscript fully available (please refer to the Data Availability Statement at the start of the manuscript PDF file)?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception. The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: No

Reviewer #4: No

--------------------

4. Is the manuscript presented in an intelligible fashion and written in standard English?<br/><br/>PLOS Digital Health does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: No

Reviewer #2: Yes

Reviewer #3: No

Reviewer #4: No

--------------------

5. Review Comments to the Author<br/><br/>Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The paper titled "Multi-modal Machine Learning Approach for Early Detection of Neurodegenerative Diseases Leveraging Brain MRI and Wearable Sensor Data" represents a Multi-modal Machine Learning Approach. However, several areas require further development to enhance its scientific impact:

1. Improve the abstract by clearly outlining the specific aims, methods, key results, and conclusions. Mention specific neurodegenerative diseases being targeted.

2. Provide more detailed background on the burden of neurodegenerative diseases, including current statistics on incidence and prevalence.

3. Include a separate Literature Review section to include recent advances in MRI biomarkers and wearable sensor technology for neurodegenerative diseases.

4. Clearly articulate the research gap your study is addressing and how it advances current knowledge.

5. Explicitly state the hypothesis being tested in the study.

6. Specify which MRI brain parameters were selected and why. Include details on how these parameters are linked to neurodegenerative diseases.

7. In the Classification Performance section, explain the calculation of polygenic risk scores (PRS) and their relevance to neurodegenerative diseases.

8. Describe the lifestyle information collected and its potential impact on neurodegenerative disease progression.

9. Detail the data preprocessing steps for both MRI and accelerometry data to ensure reproducibility.

10. Justify the choice of XGBoost as the machine learning algorithm and explain any hyperparameter tuning performed. Implement at least five more machine learning algorithms and compare the performance.

11. Add a Performance Matrices section to discuss the evaluation metrics used (AUC, accuracy, etc.) and why they were chosen.

12. In the Feature Importance section provide a more detailed explanation of the feature importance analysis and its implications. Such as visualizing the performance with and without selecting important features.

13. In the Statistical Analysis section include visual graphs to present statistical analyses conducted to validate the findings, such as significance testing and confidence intervals.

14. Add a section Comparison with Existing Methods to compare your multi-modal model with other existing models in the literature to highlight its advantages.

15. Discuss the limitations of your study, including any potential biases or confounding factors.

16. Provide more concrete suggestions for future prospective studies, including specific types of neurodegenerative diseases to be investigated.

17. Include more figures and tables to increase reader understandability. Ensure all figures and tables are clear, and well-labeled.

18. The paper lacks a well-structured outline, making it difficult for readers to follow the flow of information. It needs clear, distinct sections for the Introduction, Literature Review, Methods, Dataset, Data Preprocessing, Performance Metrics, Result Analysis, Comparison with Existing Work, Discussion, and Conclusion. Each section should be clearly defined and logically organized to enhance readability and comprehension.

By addressing these comments, the paper can be significantly improved in terms of scientific precision, clarity, and overall impact on the field of neurodegenerative disease research.

Reviewer #2: The primary goal of this study is to investigate the combination of accelerometry data from wearable devices and MRI-based brain parameters in predicting the incidence of neurodegenerative diseases. However, the authors should concentrate on the following issues before resubmission:

(i) The authors should show the contributions in point-by-point basis.

(ii) The authors should add a literature review section as well.

(iii) The authors should elaborate the technical method applied here in the study.

(iv) The authors should also show the results of more analysis to come to the conclusion.

Reviewer #3: Suggestions for Improvement formatting

Consolidate Related Questions:

Group similar questions together to avoid redundancy.

Use Numbering or Bullet Points:

Organize questions with clear numbering or bullet points.

Clear Headers or Titles:

Provide headers for different sections of questions to enhance readability.

Structured Responses:

Respond to each question in a structured format, corresponding to the numbering or bullet points used for questions.

Seek Clarification if Needed:

Request clarification on unclear or repetitive questions to streamline communication.

Novelty

What is the primary focus of the study regarding neurodegenerative diseases?

How many participants were involved in the study cohort, and where were they sourced from?

What types of data were collected from the participants?

How was the incidence of neurodegenerative diseases defined and recorded during the study?

Which machine learning algorithm was used, and what was its purpose in the study?

What was the highest achieved AUC value, and which model configuration achieved it?

According to the feature importance analysis, what were the top contributors to predicting neurodegenerative disease incidence?

Major concern

Detailed Questions Raised:

Evidence of Improvement:

What specific empirical evidence or studies demonstrate a clear improvement in predictive accuracy when combining accelerometry data and MRI-based brain parameters compared to using either method alone?

How do these studies control for potential confounding variables and biases that may influence the observed outcomes?

Limitations of MRI Parameters:

Given the heterogeneous nature of neurodegenerative diseases, what are the acknowledged limitations of MRI-based parameters in accurately capturing disease progression and early pathological changes?

How do these limitations impact the reliability and generalizability of using MRI data as a biomarker for predicting disease incidence?

Accuracy and Reliability of Accelerometry Data:

What methodologies or validation studies support the accuracy and reliability of accelerometry data from wearable devices in capturing subtle movement patterns and motor abnormalities associated with neurodegenerative diseases?

How do factors such as device variability, data processing algorithms, and participant compliance affect the consistency and interpretability of accelerometry data?

Comparison with Alternative Predictors:

In comparative terms, how does the predictive performance of combining accelerometry and MRI data stack up against alternative predictors like genetic markers, lifestyle factors, or other advanced imaging techniques (e.g., PET scans, fMRI)?

What are the strengths and weaknesses of each predictor in terms of sensitivity, specificity, and clinical applicability?

Generalizability Across Neurodegenerative Diseases:

To what extent can findings from studies primarily focused on Parkinson's Disease, as highlighted by Rastegari et al. and Schalkamp et al., be extrapolated to other neurodegenerative diseases such as Alzheimer's Disease and various types of dementia?

What disease-specific factors might influence the utility and predictive power of combined accelerometry and MRI data across different neurodegenerative conditions?

Longitudinal Study Design:

What are the key methodological considerations and challenges associated with using longitudinal data, such as that derived from large cohorts like the UK Biobank, to investigate the predictive potential of integrated accelerometry and MRI data?

How do study designs account for participant attrition, data heterogeneity over time, and changes in disease status to ensure robustness and reliability of findings?

Clinical Utility and Implementation:

Beyond research settings, how might the integration of accelerometry data with MRI parameters influence clinical decision-making and patient management in terms of early detection, treatment planning, and monitoring of neurodegenerative diseases?

What are the potential barriers and facilitators to translating research findings into practical clinical applications, and how can these be addressed to optimize patient outcomes?

These detailed questions aim to critically assess the validity, reliability, and practical implications of the hypothesis, prompting a deeper exploration of the methodological, clinical, and translational aspects surrounding the integration of accelerometry and MRI data in neurodegenerative disease prediction.

How do these studies control for potential confounding variables and biases that may influence the observed outcomes?

Ethical Concern Raised:

The demographic characteristics of the study cohort raise ethical concerns regarding potential biases and implications for the generalizability of findings.

Detailed Ethical Questions Raised:

Age Discrepancy Impact: How might the significant age difference between the neurodegenerative disease group and the healthy group influence the study's conclusions regarding disease prediction and risk factors?

Gender Disparity: What ethical considerations should be taken into account regarding the higher proportion of men in the neurodegenerative disease group compared to the healthy group? How might this gender disparity affect the interpretation of study outcomes?

Representation and Generalizability: To what extent do the demographic characteristics of the study cohort (e.g., age distribution, gender ratio) affect the generalizability of findings to broader populations, particularly those with different demographic profiles?

Potential Bias in Lifestyle Factors: How might the observed differences in lifestyle factors such as smoking and drinking habits between the disease and healthy groups introduce biases into the study outcomes? What measures were taken to mitigate these biases?

Ethical Recruitment and Consent: What ethical considerations were addressed during participant recruitment and consent processes to ensure transparency, fairness, and voluntary participation, especially given the sensitive nature of neurodegenerative disease research?

Equity in Access to Healthcare: Considering the study's findings may impact healthcare policies and resource allocation, what ethical obligations exist to ensure equitable access to early detection and intervention strategies identified through research?

Privacy and Confidentiality: How were participant privacy and confidentiality protected throughout the study, particularly concerning the sensitive health data collected and analyzed?

The results from Table 2 starkly expose the pivotal importance of brain MRI data in achieving high predictive accuracy, rendering the exclusion of such data in Model 2 a catastrophic misstep that significantly compromised the model's performance.

Questions Raised:

Justification for Exclusions: What rationale guided the decision to exclude brain MRI data in Model 2, despite its demonstrated pivotal role in enhancing predictive accuracy compared to other modalities?

Impact of Lifestyle Information: How do the findings from Model 4, which omitted lifestyle information, underscore the critical role of non-genetic factors in neurodegenerative disease prediction, despite the moderate decrease in AUC compared to Model 1?

Ethical Implications: What ethical considerations arise from the decision to exclude potentially vital information sources (such as brain MRI data) from predictive models, given their substantial impact on diagnostic accuracy and patient outcomes?

Future Research Directions: In light of these findings, what future research avenues could explore integrating additional modalities or refining existing models to optimize predictive accuracy while maintaining ethical standards and practical feasibility?

Clinical Translation: How might the findings influence clinical decision-making regarding the adoption of multi-modal predictive models for neurodegenerative diseases, particularly in terms of resource allocation and patient management strategies?

Bias and Interpretation: What steps were taken to mitigate potential biases in interpreting the contribution of each modality to predictive performance, considering the complex interplay between MRI data, genetic predisposition (PRS), lifestyle factors, and accelerometry data?

Transparency and Reporting: How were the results communicated transparently to stakeholders, including patients, healthcare providers, and policymakers, to ensure informed decision-making regarding the adoption of predictive models in clinical practice?

Results

How might focusing on specific anatomical brain regions as predictive biomarkers overlook the multifaceted nature of neurodegenerative diseases, including genetic, environmental, and socio-economic factors?

What ethical considerations arise from attributing predictive value to certain brain regions, particularly in terms of potential stigmatization and psychosocial impact on individuals with or at risk of neurodegenerative diseases?

To what extent do the identified brain regions as predictive biomarkers represent the demographic diversity of the study cohort, and how might this impact the reliability and applicability of predictive models across different populations?

How were participants informed about the potential implications of identifying specific brain regions as predictive biomarkers, and what measures were taken to provide adequate counseling and support?

What ethical guidelines should be implemented to ensure responsible communication and interpretation of research findings related to specific brain regions as biomarkers for neurodegenerative diseases?

Reviewer #4: 1. In the abstract It is recommended to include major contributions, pitfalls of the work, and a summary of results in the paper. Not properly mentioned.

2. The introduction to the paper is poorly written. There are no challenges found in the existing works, and there is no mention of the need for this work in the current scenarios. The authors do not provide any motivation behind this article. It is recommended to provide motivation through an illustrative example for better understanding. Considering the most recent work in the literature and specifying the limitations is advised. It is recommended that the summary includes the limitations of the existing works and which limitations are addressed by the authors.

3. After introduction section, directly results mentioned. Why? Their is not any flow in the manuscript.

4. The Results are incomplete. Mentioned in statistical analysis devised 4 models? Where are those in brief? No Results further.

Lack of Novelty, Not Recommended.

--------------------

6. PLOS authors have the option to publish the peer review history of their article (what does this mean? ). If published, this will include your full peer review and any attached files.

Do you want your identity to be public for this peer review? If you choose “no”, your identity will remain anonymous but your review may still be made public.

For information about this choice, including consent withdrawal, please see our Privacy Policy .

Reviewer #1: No

Reviewer #2: No

Reviewer #3: Yes: Anurag Sinha

Reviewer #4: No

--------------------

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

Attachment

Submitted filename: PDIG-D-24-00170.pdf

pdig.0000795.s007.pdf^{(119.2KB, pdf)}

PLOS Digit Health. 2025 Apr 25;4(4):e0000795. doi: 10.1371/journal.pdig.0000795.r003

Author response to Decision Letter 1

21 Sep 2024

Attachment

Submitted filename: Response to Reviewers_updated.docx

pdig.0000795.s008.docx^{(39.9KB, docx)}

PLOS Digit Health. doi: 10.1371/journal.pdig.0000795.r004

Decision Letter 1

Martin G Frasch, Md Mehedi Hassan

14 Nov 2024

PDIG-D-24-00170R1Multi-modal Machine Learning Approach for Early Detection of Neurodegenerative Diseases Leveraging Brain MRI and Wearable Sensor DataPLOS Digital HealthDear Dr. Vardhanabhuti,

Please submit your revised manuscript within 60 days Jan 13 2025 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at digitalhealth@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pdig/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

* A rebuttal letter that responds to each point raised by the editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers '. This file does not need to include responses to any formatting updates and technical items listed in the 'Journal Requirements' section below.

* A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes '.

* An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript '.

If you would like to make changes to your financial disclosure, competing interests statement, or data availability statement, please make these updates within the submission form at the time of resubmission. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

We look forward to receiving your revised manuscript.

Kind regards,

Md. Mehedi Hassan

Academic Editor

PLOS Digital Health

Leo Anthony Celi

Editor-in-ChiefPLOS Digital Healthorcid.org/0000-0001-6712-6626

Additional Editor Comments (if provided):

The authors are invited to revise their manuscript in accordance with the reviewers' comments and to resubmit after thoroughly addressing all feedback. Specifically, the authors are requested to provide a clear and comprehensive description of the training dataset.

[Note: HTML markup is below. Please do not edit.]

Reviewers' Comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

**********

2. Does this manuscript meet PLOS Digital Health’s publication criteria ? Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe methodologically and ethically rigorous research with conclusions that are appropriately drawn based on the data presented.

Reviewer #1: No

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available (please refer to the Data Availability Statement at the start of the manuscript PDF file)?

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS Digital Health does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: No

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: After careful review, there is still significant room for improvement in the manuscript titled "Multi-modal Machine Learning Approach for Early Detection of Neurodegenerative Diseases Leveraging Brain MRI and Wearable Sensor Data." Addressing the following areas could enhance the paper's scientific impact:

1. The abstract still lacks specificity in outlining the aims, methods, key results, and conclusions, particularly regarding the neurodegenerative diseases targeted.

2. The introduction does not provide sufficient background on the burden of neurodegenerative diseases, including critical statistics on incidence and prevalence.

3. A dedicated Literature Review section is necessary to contextualize the study within recent advances in MRI biomarkers and wearable sensor technology.

4. The research gap and hypothesis are not clearly articulated, which weakens the study's contribution to the field.

5. The selected MRI parameters and their relevance are described in insufficient detail, and polygenic risk scores and data preprocessing steps are inadequately explained.

6. The comparison with additional machine learning algorithms lacks depth and justification.

7. The manuscript lacks a conclusive section that summarizes the findings and their implications, which is essential for any scientific paper.

8. The overall manuscript structure remains unclear, making it challenging to follow the flow of information.

Addressing these points could significantly enhance the clarity and scientific impact of the manuscript.

Reviewer #2: Thanks to the authors for resolving the raised issues.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean? ). If published, this will include your full peer review and any attached files.

Do you want your identity to be public for this peer review? If you choose “no”, your identity will remain anonymous but your review may still be made public.

For information about this choice, including consent withdrawal, please see our Privacy Policy .

Reviewer #1: Yes

Reviewer #2: No

**********

Figure resubmission:

Reproducibility:

To enhance the reproducibility of your results, we recommend that authors of applicable studies deposit laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Attachment

Submitted filename: PDIG-D-24-00170_R1.pdf

pdig.0000795.s009.pdf^{(110.5KB, pdf)}

PLOS Digit Health. 2025 Apr 25;4(4):e0000795. doi: 10.1371/journal.pdig.0000795.r005

Author response to Decision Letter 2

31 Dec 2024

Attachment

Submitted filename: Response to Reviewers_updated R2.docx

pdig.0000795.s010.docx^{(21.3KB, docx)}

PLOS Digit Health. doi: 10.1371/journal.pdig.0000795.r006

Decision Letter 2

Martin G Frasch

20 Feb 2025

Multi-modal Machine Learning Approach for Early Detection of Neurodegenerative Diseases Leveraging Brain MRI and Wearable Sensor Data

PDIG-D-24-00170R2

Dear Dr. Vardhanabhuti,

We are pleased to inform you that your manuscript 'Multi-modal Machine Learning Approach for Early Detection of Neurodegenerative Diseases Leveraging Brain MRI and Wearable Sensor Data' has been provisionally accepted for publication in PLOS Digital Health.

Before your manuscript can be formally accepted you will need to complete some formatting changes, which you will receive in a follow-up email from a member of our team.

Please note that your manuscript will not be scheduled for publication until you have made the required changes, so a swift response is appreciated.

IMPORTANT: The editorial review process is now complete. PLOS will only permit corrections to spelling, formatting or significant scientific errors from this point onwards. Requests for major changes, or any which affect the scientific understanding of your work, will cause delays to the publication date of your manuscript.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they'll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact digitalhealth@plos.org.

Thank you again for supporting Open Access publishing; we are looking forward to publishing your work in PLOS Digital Health.

Best regards,

Martin G Frasch

Section Editor

PLOS Digital Health

***********************************************************

Additional Editor Comments (if provided):

Reviewer Comments (if any, and for reference):

Reviewer's Responses to Questions

Comments to the Author

Reviewer #1: All comments have been addressed

**********

Reviewer #1: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available (please refer to the Data Availability Statement at the start of the manuscript PDF file)?

Reviewer #1: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

Reviewer #1: Yes

**********

6. Review Comments to the Author

Reviewer #1: The authors have thoroughly addressed all comments, providing clear and well-justified responses, and I recommend the paper for acceptance.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean? ). If published, this will include your full peer review and any attached files.

Do you want your identity to be public for this peer review? If you choose “no”, your identity will remain anonymous but your review may still be made public.

For information about this choice, including consent withdrawal, please see our Privacy Policy .

Reviewer #1: Yes: Farhana Yasmin

**********

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

S1 Table. T1 Brain MRI features and the related field IDs in UK Biobank.

(DOCX)

pdig.0000795.s001.docx^{(21.8KB, docx)}

S2 Table. T2 Brain MRI features and the related field IDs in UK Biobank.

(DOCX)

pdig.0000795.s002.docx^{(14.7KB, docx)}

S3 Table. RPS scores and the related field IDs in UK Biobank.

(DOCX)

pdig.0000795.s003.docx^{(17.7KB, docx)}

S4 Table. Accelerometry average data and the related field IDs in UK Biobank.

(DOCX)

pdig.0000795.s004.docx^{(16.8KB, docx)}

S5 Table. Accelerometry-derived data and the related field IDs in UK Biobank.

(DOCX)

pdig.0000795.s005.docx^{(15.8KB, docx)}

S6 Table. Comparison with Common Models.

(DOCX)

pdig.0000795.s006.docx^{(14.7KB, docx)}

Attachment

Submitted filename: PDIG-D-24-00170.pdf

pdig.0000795.s007.pdf^{(119.2KB, pdf)}

Attachment

Submitted filename: Response to Reviewers_updated.docx

pdig.0000795.s008.docx^{(39.9KB, docx)}

Attachment

Submitted filename: PDIG-D-24-00170_R1.pdf

pdig.0000795.s009.pdf^{(110.5KB, pdf)}

Attachment

Submitted filename: Response to Reviewers_updated R2.docx

pdig.0000795.s010.docx^{(21.3KB, docx)}

Data Availability Statement

The data used for this work is from the UK Biobank Study, which can be accessed at https://www.ukbiobank.ac.uk/.

[pdig.0000795.ref001] 1.Feigin VL, Vos T, Nichols E, Owolabi MO, Carroll WM, Dichgans M, et al. The global burden of neurological disorders: translating evidence into policy. Lancet Neurol. 2020;19(3):255–65. doi: 10.1016/S1474-4422(19)30411-9 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref002] 2.Van Schependom J, D’haeseleer M. Advances in Neurodegenerative Diseases. J Clin Med. 2023;12(5):1709. doi: 10.3390/jcm12051709 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref003] 3.Zahra W, Rai SN, Birla H, Singh SS, Dilnashin H, Rathore AS, et al. The global economic impact of neurodegenerative diseases: Opportunities and challenges. Bioeconomy for sustainable development. 2020:333–45. [Google Scholar]

[pdig.0000795.ref004] 4.Nichols E, Steinmetz J, Vollset S, Fukutaki K, Chalek J, Abd-Allah F. Estimation of the global prevalence of dementia in 2019 and forecasted prevalence in 2050: an analysis for the Global Burden of Disease Study 2019. The Lancet Public Health. 2022;7(2):e105–25. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref005] 5.Ding C, Wu Y, Chen X, Chen Y, Wu Z, Lin Z. Global, regional, and national burden and attributable risk factors of neurological disorders: The Global Burden of Disease study 1990–2019. Frontiers in Public Health. 2022;10:952161. doi: 10.3389/fpubh.2022.952161 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref006] 6.Ehrenberg AJ, Khatun A, Coomans E, Betts MJ, Capraro F, Thijssen EH, et al. Relevance of biomarkers across different neurodegenerative diseases. Alzheimers Res Ther. 2020;12(1):56. doi: 10.1186/s13195-020-00601-w [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref007] 7.Young PNE, Estarellas M, Coomans E, Srikrishna M, Beaumont H, Maass A, et al. Imaging biomarkers in neurodegeneration: current and future practices. Alzheimers Res Ther. 2020;12(1):49. doi: 10.1186/s13195-020-00612-7 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref008] 8.Henley SMD, Bates GP, Tabrizi SJ. Biomarkers for neurodegenerative diseases. Curr Opin Neurol. 2005;18(6):698–705. doi: 10.1097/01.wco.0000186842.51129.cb [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref009] 9.Koikkalainen J, Rhodius-Meester H, Tolonen A, Barkhof F, Tijms B, Lemstra A. Differential diagnosis of neurodegenerative diseases using structural MRI data. NeuroImage: Clinical. 2016;11:435–49. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref010] 10.Agosta F, Galantucci S, Filippi M. Advanced magnetic resonance imaging of neurodegenerative diseases. Neurol Sci. 2017;38(1):41–51. doi: 10.1007/s10072-016-2764-x [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref011] 11.Hansson O. Biomarkers for neurodegenerative diseases. Nat Med. 2021;27(6):954–63. doi: 10.1038/s41591-021-01382-x [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref012] 12.Frisoni GB, Jack CR Jr, Bocchetta M, Bauer C, Frederiksen KS, Liu Y, et al. The EADC-ADNI Harmonized Protocol for manual hippocampal segmentation on magnetic resonance: evidence of validity. Alzheimers Dement. 2015;11(2):111–25. doi: 10.1016/j.jalz.2014.05.1756 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref013] 13.Risacher SL, Shen L, West JD, Kim S, McDonald BC, Beckett LA, et al. Longitudinal MRI atrophy biomarkers: relationship to conversion in the ADNI cohort. Neurobiol Aging. 2010;31(8):1401–18. doi: 10.1016/j.neurobiolaging.2010.04.029 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref014] 14.Whitwell J, Shiung M, Przybelski S, Weigand S, Knopman D, Boeve B. MRI patterns of atrophy associated with progression to AD in amnestic mild cognitive impairment. Neurology. 2008;70(7):512–20. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref015] 15.Wei H, Kong M, Zhang C, Guan L, Ba M, for Alzheimer’s Disease Neuroimaging Initiative*. The structural MRI markers and cognitive decline in prodromal Alzheimer’s disease: a 2-year longitudinal study. Quant Imaging Med Surg. 2018;8(10):1004–19. doi: 10.21037/qims.2018.10.08 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref016] 16.Risacher S, Anderson W, Charil A, Castelluccio P, Shcherbinin S, Saykin A. Alzheimer disease brain atrophy subtypes are associated with cognition and rate of decline. Neurology. 2017;89(21):2176–86. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref017] 17.Rascovsky K, Hodges JR, Knopman D, Mendez MF, Kramer JH, Neuhaus J, et al. Sensitivity of revised diagnostic criteria for the behavioural variant of frontotemporal dementia. Brain. 2011;134(Pt 9):2456–77. doi: 10.1093/brain/awr179 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref018] 18.Hanganu A, Bedetti C, Degroot C, Mejia-Constain B, Lafontaine A-L, Soland V, et al. Mild cognitive impairment is linked with faster rate of cortical thinning in patients with Parkinson’s disease longitudinally. Brain. 2014;137(Pt 4):1120–9. doi: 10.1093/brain/awu036 [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref019] 19.Mitchell T, Lehéricy S, Chiu SY, Strafella AP, Stoessl AJ, Vaillancourt DE. Emerging Neuroimaging Biomarkers Across Disease Stage in Parkinson Disease: A Review. JAMA Neurol. 2021;78(10):1262–72. doi: 10.1001/jamaneurol.2021.1312 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref020] 20.Zeighami Y, Ulla M, Iturria-Medina Y, Dadar M, Zhang Y, Larcher K. Network structure of brain atrophy in de novo Parkinson’s disease. eLife. 2015;4:e08440. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref021] 21.Velayudhan L, Proitsi P, Westman E, Muehlboeck J, Mecocci P, Vellas B, et al. Entorhinal cortex thickness predicts cognitive decline in Alzheimer’s disease. Journal of Alzheimer’s Disease. 2013;33(3):755–66. [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref022] 22.Du A-T, Schuff N, Kramer JH, Rosen HJ, Gorno-Tempini ML, Rankin K, et al. Different regional patterns of cortical thinning in Alzheimer’s disease and frontotemporal dementia. Brain. 2007;130(Pt 4):1159–66. doi: 10.1093/brain/awm016 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref023] 23.McAleese KE, Alafuzoff I, Charidimou A, De Reuck J, Grinberg LT, Hainsworth AH, et al. Post-mortem assessment in vascular dementia: advances and aspirations. BMC Med. 2016;14(1):129. doi: 10.1186/s12916-016-0676-5 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref024] 24.Abraham HMA, Wolfson L, Moscufo N, Guttmann CRG, Kaplan RF, White WB. Cardiovascular risk factors and small vessel disease of the brain: Blood pressure, white matter lesions, and functional decline in older persons. J Cereb Blood Flow Metab. 2016;36(1):132–42. doi: 10.1038/jcbfm.2015.121 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref025] 25.Richard E, Gouw AA, Scheltens P, van Gool WA. Vascular care in patients with Alzheimer disease with cerebrovascular lesions slows progression of white matter lesions on MRI: the evaluation of vascular care in Alzheimer’s disease (EVA) study. Stroke. 2010;41(3):554–6. doi: 10.1161/STROKEAHA.109.571281 [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref026] 26.Prins ND, Scheltens P. White matter hyperintensities, cognitive impairment and dementia: an update. Nat Rev Neurol. 2015;11(3):157–65. doi: 10.1038/nrneurol.2015.10 [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref027] 27.Gaubert M, Lange C, Garnier-Crussard A, Köbe T, Bougacha S, Gonneaud J, et al. Topographic patterns of white matter hyperintensities are associated with multimodal neuroimaging biomarkers of Alzheimer’s disease. Alzheimers Res Ther. 2021;13(1):29. doi: 10.1186/s13195-020-00759-3 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref028] 28.Lee S, Viqar F, Zimmerman ME, Narkhede A, Tosto G, Benzinger TLS, et al. White matter hyperintensities are a core feature of Alzheimer’s disease: Evidence from the dominantly inherited Alzheimer network. Ann Neurol. 2016;79(6):929–39. doi: 10.1002/ana.24647 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref029] 29.Yoshita M, Fletcher E, Harvey D, Ortega M, Martinez O, Mungas DM, et al. Extent and distribution of white matter hyperintensities in normal aging, MCI, and AD. Neurology. 2006;67(12):2192–8. doi: 10.1212/01.wnl.0000249119.95747.1f [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref030] 30.Damulina A, Pirpamer L, Seiler S, Benke T, Dal-Bianco P, Ransmayr G. White matter hyperintensities in Alzheimer’s disease: a lesion probability mapping study. Journal of Alzheimer’s Disease. 2019;68(2):789–96. [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref031] 31.Garnier-Crussard A, Cotton F, Krolak-Salmon P, Chételat G. White matter hyperintensities in Alzheimer’s disease: Beyond vascular contribution. Alzheimers Dement. 2023;19(8):3738–48. doi: 10.1002/alz.13057 [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref032] 32.Linortner P, McDaniel C, Shahid M, Levine TF, Tian L, Cholerton B, et al. White Matter Hyperintensities Related to Parkinson’s Disease Executive Function. Mov Disord Clin Pract. 2020;7(6):629–38. doi: 10.1002/mdc3.12956 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref033] 33.Zou X, Dong Z, Chen X, Yu Q, Yin H, Yi L. White matter hyperintensities burden in the frontal regions is positively correlated to the freezing of gait in Parkinson’s disease. Frontiers in Aging Neuroscience. 2023;15:1156648. doi: 10.3389/fnagi.2023.1156648 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref034] 34.Ozzoude M, Varriano B, Beaton D, Ramirez J, Adamo S, Holmes MF, et al. White matter hyperintensities and smaller cortical thickness are associated with neuropsychiatric symptoms in neurodegenerative and cerebrovascular diseases. Alzheimers Res Ther. 2023;15(1):114. doi: 10.1186/s13195-023-01257-y [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref035] 35.Chouliaras L, O’Brien JT. The use of neuroimaging techniques in the early and differential diagnosis of dementia. Mol Psychiatry. 2023;28(10):4084–97. doi: 10.1038/s41380-023-02215-8 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref036] 36.Beauchet O, Annweiler C, Dubost V, Allali G, Kressig RW, Bridenbaugh S, et al. Stops walking when talking: a predictor of falls in older adults?. Eur J Neurol. 2009;16(7):786–95. doi: 10.1111/j.1468-1331.2009.02612.x [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref037] 37.Cicirelli G, Impedovo D, Dentamaro V, Marani R, Pirlo G, D’Orazio TR. Human gait analysis in neurodegenerative diseases: a review. IEEE Journal of Biomedical and Health Informatics. 2021;26(1):229–42. [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref038] 38.Morris M, Iansek R, Matyas T, Summers J. The pathogenesis of gait hypokinesia in Parkinson’s disease. Brain. 1994;117(5):1169–81. [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref039] 39.del Pozo Cruz B, Ahmadi M, Naismith S, Stamatakis E. Association of daily step count and intensity with incident dementia in 78 430 adults living in the UK. JAMA Neurology. 2022;79(10):1059–63. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref040] 40.Raichlen D, Aslan D, Sayre M, Bharadwaj P, Ally M, Maltagliati S. Sedentary behavior and incident dementia among older adults. JAMA. 2023;330(10):934–40. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref041] 41.Del Din S, Godfrey A, Mazzà C, Lord S, Rochester L. Free‐living monitoring of Parkinson’s disease: Lessons from the field. Movement Disorders. 2016;31(9):1293–313. [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref042] 42.Rastegari E, Ali H, Marmelat V. Detection of Parkinson’s Disease Using Wrist Accelerometer Data and Passive Monitoring. Sensors (Basel). 2022;22(23):9122. doi: 10.3390/s22239122 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref043] 43.Williamson J, Telfer B, Mullany R, Friedl K. Detecting Parkinson’s disease from wrist-worn accelerometry in the UK Biobank. Sensors. 2021;21(6):2047. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref044] 44.Schalkamp A-K, Peall KJ, Harrison NA, Sandor C. Wearable movement-tracking data identify Parkinson’s disease years before clinical diagnosis. Nat Med. 2023;29(8):2048–56. doi: 10.1038/s41591-023-02440-2 [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref045] 45.Winer JR, Lok R, Weed L, He Z, Poston KL, Mormino EC, et al. Impaired 24-h activity patterns are associated with an increased risk of Alzheimer’s disease, Parkinson’s disease, and cognitive decline. Alzheimers Res Ther. 2024;16(1):35. doi: 10.1186/s13195-024-01411-0 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref046] 46.De Flores R, La Joie R, Chételat G. Structural imaging of hippocampal subfields in healthy aging and Alzheimer’s disease. Neuroscience. 2015;309:29–50. [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref047] 47.Xu R, Hu X, Jiang X, Zhang Y, Wang J, Zeng X. Longitudinal volume changes of hippocampal subfields and cognitive decline in Parkinson’s disease. Quant Imaging Med Surg. 2020;10(1):220–32. doi: 10.21037/qims.2019.10.17 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref048] 48.Calabresi P, Castrioto A, Di Filippo M, Picconi B. New experimental and clinical links between the hippocampus and the dopaminergic system in Parkinson’s disease. Lancet Neurol. 2013;12(8):811–21. doi: 10.1016/S1474-4422(13)70118-2 [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref049] 49.Galton C, Patterson K, Graham K, Lambon-Ralph M, Williams G, Antoun N. Differing patterns of temporal atrophy in Alzheimer’s disease and semantic dementia. Neurology. 2001;57(2):216–25. [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref050] 50.Chapleau M, Aldebert J, Montembeault M, Brambati S. Atrophy in Alzheimer’s disease and semantic dementia: an ALE meta-analysis of voxel-based morphometry studies. Journal of Alzheimer’s disease. 2016;54(3):941–55. [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref051] 51.Lee HM, Kwon K-Y, Kim M-J, Jang J-W, Suh S-I, Koh S-B, et al. Subcortical grey matter changes in untreated, early stage Parkinson’s disease without dementia. Parkinsonism Relat Disord. 2014;20(6):622–6. doi: 10.1016/j.parkreldis.2014.03.009 [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref052] 52.Pini L, Pievani M, Bocchetta M, Altomare D, Bosco P, Cavedo E. Brain atrophy in Alzheimer’s disease and aging. Ageing research reviews. 2016;30:25–48. [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref053] 53.Slachevsky A, Forno G, Barraza P, Mioshi E, Delgado C, Lillo P. Mapping the neuroanatomy of functional decline in Alzheimer’s disease from basic to advanced activities of daily living. Journal of Neurology. 2019;266(6):1310–22. [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref054] 54.Lee P, Chou K, Chung C, Lai T, Zhou J, Wang P. Posterior cingulate cortex network predicts Alzheimer’s disease progression. Frontiers in Aging Neuroscience. 2020;12:608667. doi: 10.3389/fnagi.2020.608667 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref055] 55.Leech R, Sharp DJ. The role of the posterior cingulate cortex in cognition and disease. Brain. 2014;137(Pt 1):12–32. doi: 10.1093/brain/awt162 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref056] 56.Jones BF, Barnes J, Uylings HBM, Fox NC, Frost C, Witter MP, et al. Differential regional atrophy of the cingulate gyrus in Alzheimer disease: a volumetric MRI study. Cereb Cortex. 2006;16(12):1701–8. doi: 10.1093/cercor/bhj105 [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref057] 57.de Schipper L, van der Grond J, Marinus J, Henselmans J, van Hilten J. Loss of integrity and atrophy in cingulate structural covariance networks in Parkinson’s disease. NeuroImage: Clinical. 2017;15:587–93. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref058] 58.Zhornitsky S, Chaudhary S, Le T, Chen Y, Zhang S, Potvin S. Cognitive dysfunction and cerebral volumetric deficits in individuals with Alzheimer’s disease, alcohol use disorder, and dual diagnosis. Psychiatry Research: Neuroimaging. 2021;317:111380. doi: 10.1016/j.pscychresns.2021.111380 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref059] 59.Liu C, Jiang J, Zhou H, Zhang H, Wang M, Jiang J. Brain functional and structural signatures in Parkinson’s disease. Frontiers in Aging Neuroscience. 2020;12:125. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref060] 60.Li L, Ji B, Zhao T, Cui X, Chen J, Wang Z. The structural changes of gray matter in Parkinson disease patients with mild cognitive impairments. PLoS One. 2022;17(7):e0269787. doi: 10.1371/journal.pone.0269787 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref061] 61.van de Mortel LA, Thomas RM, van Wingen GA. Grey matter loss at different stages of cognitive decline: a role for the thalamus in developing Alzheimer’s disease. Journal of Alzheimer’s Disease. 2021;83(2):705–20. doi: 10.3233/JAD-201123 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref062] 62.Fu T, Klietz M, Nösel P, Wegner F, Schrader C, Höglinger GU, et al. Brain Morphological Alterations Are Detected in Early-Stage Parkinson’s Disease with MRI Morphometry. J Neuroimaging. 2020;30(6):786–92. doi: 10.1111/jon.12769 [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref063] 63.Lee SH, Kim SS, Tae WS, Lee SY, Choi JW, Koh SB, et al. Regional volume analysis of the Parkinson disease brain in early disease stage: gray matter, white matter, striatum, and thalamus. AJNR Am J Neuroradiol. 2011;32(4):682–7. doi: 10.3174/ajnr.A2372 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref064] 64.Tavel P, Nikolai T. Amygdala and emotionality in Parkinson’s disease: An integrative review of the neuropsychological evidence. Neuroendocrinology Letters. 2018;39(2). [PubMed] [Google Scholar]

[pdig.0000795.ref065] 65.Lupton M, Strike L, Hansell N, Wen W, Mather K, Armstrong N. The effect of increased genetic risk for Alzheimer’s disease on hippocampal and amygdala volume. Neurobiology of Aging. 2016;40(1). [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref066] 66.Miller M, Younes L, Ratnanather J, Brown T, Trinh H, Lee D. Amygdalar atrophy in symptomatic Alzheimer’s disease based on diffeomorphometry: the BIOCARD cohort. Neurobiology of Aging. 2015;36:S3–10. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref067] 67.Convit A, de Asis J, de Leon MJ, Tarshish CY, De Santi S, Rusinek H. Atrophy of the medial occipitotemporal, inferior, and middle temporal gyri in non-demented elderly predict decline to Alzheimer’s disease. Neurobiol Aging. 2000;21(1):19–26. doi: 10.1016/s0197-4580(99)00107-4 [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref068] 68.Busatto G, Garrido G, Almeida O, Castro C, Camargo C, Cid C. A voxel-based morphometry study of temporal lobe gray matter reductions in Alzheimer’s disease. Neurobiology of Aging. 2003;24(2):221–31. doi: 10.1016/S0197-4580(02)00267-0 [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref069] 69.Guo X, Wang Z, Li K, Li Z, Qi Z, Jin Z, et al. Voxel-based assessment of gray and white matter volumes in Alzheimer’s disease. Neurosci Lett. 2010;468(2):146–50. doi: 10.1016/j.neulet.2009.10.086 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref070] 70.Hafkemeijer A, Möller C, Dopper EGP, Jiskoot LC, van den Berg-Huysmans AA, van Swieten JC, et al. Differences in structural covariance brain networks between behavioral variant frontotemporal dementia and Alzheimer’s disease. Hum Brain Mapp. 2016;37(3):978–88. doi: 10.1002/hbm.23081 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref071] 71.Irish M, Piguet O, Hodges JR, Hornberger M. Common and unique gray matter correlates of episodic memory dysfunction in frontotemporal dementia and Alzheimer’s disease. Hum Brain Mapp. 2014;35(4):1422–35. doi: 10.1002/hbm.22263 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref072] 72.Filoteo JV, Reed JD, Litvan I, Harrington DL. Volumetric correlates of cognitive functioning in nondemented patients with Parkinson’s disease. Mov Disord. 2014;29(3):360–7. doi: 10.1002/mds.25633 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref073] 73.Kobayakawa M, Tsuruya N, Kawamura M. Decision-making performance in Parkinson’s disease correlates with lateral orbitofrontal volume. Journal of the neurological sciences. 2017;372(1):232–8. [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref074] 74.Cerasa A, Salsone M, Nigro S, Chiriaco C, Donzuso G, Bosco D. Cortical volume and folding abnormalities in Parkinson’s disease patients with pathological gambling. Parkinsonism & related disorders. 2014;20(11):1209–14. [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref075] 75.Owen AM, Doyon D. The cognitive neuropsychology of Parkinson’s disease: a functional neuroimaging perspective. Advances in Neurology. 1999;80(1):49–56. [PubMed] [Google Scholar]

[pdig.0000795.ref076] 76.Burton EJ, McKeith IG, Burn DJ, Williams ED, O’Brien JT. Cerebral atrophy in Parkinson’s disease with and without dementia: a comparison with Alzheimer’s disease, dementia with Lewy bodies and controls. Brain. 2004;127(Pt 4):791–800. doi: 10.1093/brain/awh088 [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref077] 77.Beyer M, Janvin C, Larsen J, Aarsland D. A magnetic resonance imaging study of patients with Parkinson’s disease with mild cognitive impairment and dementia using voxel-based morphometry. Journal of Neurology, Neurosurgery & Psychiatry. 2007;78(3):254–9. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref078] 78.Nagano-Saito A, Washimi Y, Arahata Y, Kachi T, Lerch J, Evans A. Cerebral atrophy and its relation to cognitive impairment in Parkinson disease. Neurology. 2005;64(2):224–9. [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref079] 79.Zhao H, Wang R, Qi D, Xie J, Cao J, Liao WH. Wearable gait monitoring for diagnosis of neurodegenerative diseases. Measurement. 2022. Oct 1;202:111839. [Google Scholar]

[pdig.0000795.ref080] 80.Littlejohns T, Holliday J, Gibson L, Garratt S, Oesingmann N, Alfaro-Almagro F, et al. The UK Biobank imaging enhancement of 100,000 participants: rationale, data collection, management and future directions. Nature Communications. 2020;11(1):2624. doi: 10.1038/s41467-020-16445-0 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref081] 81.Moore A, Bell M. XGBoost, A Novel Explainable AI Technique, in the Prediction of Myocardial Infarction: A UK Biobank Cohort Study. Clin Med Insights Cardiol. 2022;16:11795468221133611. doi: 10.1177/11795468221133611 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pdig.0000795.ref082] 82.Stokanović S, Đukić D, Miljković N. Robustness of XGBoost Algorithm to Missing Features for Binary Classification of Medical Data. Proceedings of the 23rd International Symposium INFOTEH-JAHORINA (INFOTEH). 2024:1–6. [Google Scholar]

[pdig.0000795.ref083] 83.Ma B, Meng F, Yan G, Yan H, Chai B, Song F. Diagnostic classification of cancers using extreme gradient boosting algorithm and multi-omics data. Comput Biol Med. 2020;121:103761. doi: 10.1016/j.compbiomed.2020.103761 [DOI] [PubMed] [Google Scholar]

[pdig.0000795.ref084] 84.Price J, Yamazaki T, Fujihara K, Sone H. XGBoost: interpretable machine learning approach in medicine. In2022 5th World Symposium on Communication Engineering (WSCE) 2022. Sep 16 (pp. 109-113). IEEE. [Google Scholar]

PERMALINK

Multi-modal machine learning approach for early detection of neurodegenerative diseases leveraging brain MRI and wearable sensor data

Andrew Li

Jie Lian

Varut Vardhanabhuti

Roles

Abstract

Author summary

Introduction

Structural MRI brain parameters as imaging biomarkers for neurogenerative diseases

Accelerometry data from wearable devices as a diagnostic and prognostic tool for neurodegenerative diseases

Study objective

Results

Demographics of study cohort

Table 1. Participants demographic data with lifestyle factors.

Model performance

Table 2. The comparison of model performances in terms of AUROC.

Feature importance

Fig 1. Top 20 important features in predicting neurodegenerative disease as identified by Model 1 (Abbreviations: GMV.

Discussion

Methods

Ethical approval

Participants and study design

Fig 2. (A) Flowchart of Patient Inclusion Process; (B) Study omics including MRI, accelerometry data, genetic data and lifestyle information.

Brain MRI pre-processing

Polygenic risk score data

Accelerometry data

Lifestyle data

Input modalities

Classification performance

Statistical analysis

Supporting information

Data Availability

Funding Statement

References

Author response to Decision Letter 0

Decision Letter 0

Martin G Frasch

Md Mehedi Hassan

Roles

Author response to Decision Letter 1

Decision Letter 1

Martin G Frasch

Md Mehedi Hassan

Roles

Author response to Decision Letter 2

Decision Letter 2

Martin G Frasch

Roles

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases