Skip to main content
NCBI home page
Human Brain Mapping logoLink to Human Brain Mapping
. 2023 Mar 27;44(8):3433–3445. doi: 10.1002/hbm.26290

Identifying individuals with attention‐deficit/hyperactivity disorder based on multisite resting‐state functional magnetic resonance imaging: A radiomics analysis

Guanlu Liu 1, Weizhao Lu 1,2, Jianfeng Qiu 1,✉, Liting Shi 2,✉
PMCID: PMC10171499  PMID: 36971664

Abstract

Attention‐deficit/hyperactivity disorder (ADHD) is one of the most common neurodevelopmental disorders, characterized by symptoms of age‐inappropriate inattention, hyperactivity, and impulsivity. Apart from behavioral symptoms investigated by psychiatric methods, there is no standard biological test to diagnose ADHD. This study aimed to explore whether the radiomics features based on resting‐state functional magnetic resonance (rs‐fMRI) have more discriminative power for the diagnosis of ADHD. The rs‐fMRI of 187 subjects with ADHD and 187 healthy controls were collected from 5 sites of ADHD‐200 Consortium. A total of four preprocessed rs‐fMRI images including regional homogeneity (ReHo), amplitude of low‐frequency fluctuation (ALFF), voxel‐mirrored homotopic connectivity (VMHC) and network degree centrality (DC) were used in this study. From each of the four images, we extracted 93 radiomics features within each of 116 automated anatomical labeling brain areas, resulting in a total of 43,152 features for each subject. After dimension reduction and feature selection, 19 radiomics features were retained (5 from ALFF, 9 from ReHo, 3 from VMHC and 2 from DC). By training and optimizing a support vector machine model using the retained features of training dataset, we achieved the accuracy of 76.3% and 77.0% (areas under curve = 0.811 and 0.797) in the training and testing datasets, respectively. Our findings demonstrate that radiomics can be a novel strategy for fully utilizing rs‐fMRI information to distinguish ADHD from healthy controls. The rs‐fMRI‐based radiomics features have the potential to be neuroimaging biomarkers for ADHD.

Keywords: attention‐deficit/hyperactivity disorder, radiomics, resting‐state functional magnetic resonance imaging, support vector machine

Radiomics features extracted from rs‐fMRI metrics can be used for distinguishing ADHD patients from HC with better classification performance. There was a significant correlation between radiomics features and clinical measures, thus radiomics may be potential neuroimaging biomarkers of ADHD.

graphic file with name HBM-44-3433-g002.jpg

1. INTRODUCTION

Attention‐deficit/hyperactivity disorder (ADHD) affects nearly 5% of school‐age children and 2–4% of adolescents worldwide (Polanczyk & Jensen, 2008). ADHD is one of the most common neurodevelopmental disorders, characterized by symptoms of age‐inappropriate inattention, hyperactivity, and impulsivity (Sharma & Couture, 2014). Although the symptoms of children with ADHD decrease with age, they do not disappear completely; the impairing symptoms continue to adulthood in two‐thirds of children without appropriate treatment (Leung & Hon, 2016). In addition, ADHD has an adverse effect on social, cognitive, educational, and emotional functions (Shaw et al., 2014; Willcutt, 2012), which contributes to poor school or job performance, significantly impacted quality of life, and long‐term burden on affected families (Nigg, 2013). Thus, the early and accurate diagnosis of ADHD is crucial.

Unfortunately, it is a complicated pipeline to identify ADHD individuals, which has a sensitivity of 70–90% based on the Diagnostic and Statistical Manual of Mental Disorders, Fifth edition (DSM‐5), and as a result false‐positive diagnoses are involved (Weiler et al., 2000). However, there is no standard biological test to diagnose ADHD apart from behavioral symptoms investigated by psychiatric methods. Recently, there have been an increasing number of studies to explore possible neurobiological mechanisms in patients with ADHD using imaging‐based methods (Arbabshirani et al., 2017; Zhou et al., 2018). Specifically, as a non‐invasive and high‐resolution imaging technique, functional magnetic resonance imaging (fMRI) may show brain abnormalities of ADHD patients and provide helpful objective information to clinical psychiatric evaluators for diagnosing ADHD (Bos et al., 2017; de Lacy et al., 2018; Liu et al., 2010). Several fMRI imaging‐based diagnostic models have been established for ADHD using machine learning (Jie et al., 2016; Miao et al., 2019; Riaz et al., 2018, 2020; Shao et al., 2018; Wang et al., 2018), which have received much attention as a promising approach to differentiating psychiatric patients from healthy controls (Huys et al., 2016; Savage, 2017). As a data‐driven method, the combination of machine learning and neuroimaging can help to seek potential biomarkers of ADHD.

In ADHD, the classification performance of most fMRI‐based diagnostic models is still not high enough (Miao et al., 2019; Riaz et al., 2018, 2020; Shao et al., 2018; Wang et al., 2018). One possible explanation for these results is that the information content of fMRI is not being fully utilized due to the limitations of statistical analysis. A robust adjunct analysis technique is needed to characterize fMRI more effectively and reliably. Such analysis should be able to capture minor differences and fluctuations of the signal intensity occurring in brain functional activity. Radiomics, as an emerging medical image analysis framework, has the potential to address this issue through mining high‐throughput imaging features with strong discriminative power (Gillies et al., 2016; Kumar et al., 2012). Recently, radiomics has been used in clinical oncology to extract imaging features within tumor tissue, which can quantitatively characterize tumor spatial heterogeneity that cannot be visually assessed (Conti et al., 2021; Kumar et al., 2012; Mayerhoefer et al., 2020). Radiomics features extracted from medical images reflect the texture information of tumors that may correlate to changes of tumor microenvironment. This indicates that radiomics represents sophisticated image information and has potential to characterize pathological diseases. Similarly, brain fMRI images can be quantitatively analyzed using a radiomics framework by segmenting the preprocessed images into a standard brain atlas and extracting characteristic parameters from each region of interest (ROI) for further analysis.

To the best of our knowledge, this is the first study to introduce radiomics into rs‐fMRI of ADHD. We aimed to extract radiomics features from four preprocessed resting‐state fMRI (rs‐fMRI) metrics, including amplitude of low‐frequency fluctuation (ALFF) (Zou et al., 2008), regional homogeneity (ReHo) (Zang et al., 2004), voxel‐mirrored homotopic connectivity (VMHC) (Zuo et al., 2010), and network degree centrality (DC) (Buckner et al., 2009), for the quantitative analysis of ADHD. These fMRI metrics are widely studied in the researches of ADHD and other mental diseases and have contributed interesting findings (Kelly et al., 2011; Kim et al., 2018; Shang et al., 2021; Wang et al., 2020; Yoo et al., 2018). We hypothesized that the rs‐fMRI‐based radiomics features have better discrimination ability than traditional mean signal features and have the potential to be neuroimaging biomarkers of ADHD.

2. MATERIALS AND METHODS

2.1. Datasets and participant selection

The rs‐fMRI and phenotype data were downloaded from ADHD‐200 consortium (http://fcon1000.projects.nitrc.org/indi/adhd200/index.html). All enrolled participants met the following criteria: (1) had no necessary information missing or abnormal parameters; (2) were right‐handed; (3) had no obvious brain injury; (4) healthy controls had no history of neurological diseases and other mental disorders; (5) had no excessive head movements (>3.0 mm of translation or degrees of rotation in any direction); (6) rs‐fMRI passed the quality control process. A case–control strategy was applied to select matched samples in each dataset (Schwarz et al., 2019). We used the propensity score matching to select subsets of ADHD and healthy control (HC), 1:1 matched on age and gender. Matching was performed separately for each site. There are increasing evidences that intelligence quotient (IQ) should be treated as a part of ADHD phenotype rather than a nuisance variance or included as a covariate in analyses (de Zeeuw et al., 2012; Dennis et al., 2009; Wilson et al., 2011). Therefore, we neither took IQ as a covariate nor involved it as a factor in propensity score matching. Finally, the 5 well‐matched subsets of 20 ADHD and 20 HC in the Kennedy Krieger Institute (KKI) dataset, 82 ADHD and 82 HC in the New York University Medical Center (NYU) dataset, 44 ADHD and 44 HC in the Peking University (PU) 1 dataset, 22 ADHD and 22 HC in the PU 2, and 19 ADHD and 19 HC in the PU 3 dataset were identified. Participants’ demographic information is included in Table 1. A brief description of data scanning parameters is provided in Appendix S1. The flowchart of this study is presented in Figure 1.

TABLE 1.

Descriptive statistics for the ADHD and HC groups across the five sites.

KKI NYU PU 1 PU 2 PU 3
HC (n = 20) ADHD (n = 20) p HC (n = 82) ADHD (n = 82) p HC (n = 44) ADHD (n = 44) p HC (n = 22) ADHD (n = 22) p HC (n = 19) ADHD (n = 19) p
Age 9.9 ± 1.3 10.2 ± 1.7 .532 12.1 ± 3.0 11.8 ± 2.8 .536 11.3 ± 2.0 11.2 ± 2.1 .852 11.6 ± 1.8 12.2 ± 1.8 .314 13.3 ± 1.0 13.3 ± 1.4 .954
Gender (F/M) 10/10 8/12 .751 42/40 31/51 .116 17/27 11/33 .252 0/22 0/22 — 0/19 0/19 —
FIQ 111.4 ± 13.5 105.6 ± 15.3 .212 111.7 ± 14.1 106 ± 13.6 .024 118.2 ± 14.0 103.4 ± 13.6 <.001 121.6 ± 14.0 110.0 ± 12.8 .007 112.7 ± 13.2 102.7 ± 10.4 .014
ADHD index 45.9 ± 4.7 75.5 ± 9.7 <.001 45.5 ± 6.4 72.3 ± 9.6 <.001 30.6 ± 7.1 48.0 ± 6.3 <.001 28.1 ± 4.3 52.9 ± 9.4 <.001 28.3 ± 7.3 50.3 ± 9.4 <.001
Inattentive 45.9 ± 4.7 74.8 ± 10.1 <.001 45.8 ± 6.4 71.8 ± 9.8 <.001 16.3 ± 4.3 27.1 ± 4.4 <.001 15.0 ± 2.2 29.4 ± 3.0 <.001 15.6 ± 5.4 27.7 ± 4.4 <.001
Hyper/impulsive 46.5 ± 4.7 74.8 ± 10.2 <.001 46.5 ± 5.3 66.9 ± 12.8 <.001 14.3 ± 3.8 21.1 ± 6.2 <.001 13.1 ± 3.9 23.4 ± 6.9 <.001 12.7 ± 2.2 22.6 ± 6.6 <.001
Mean FD (mm) 0.07 ± 0.05 0.08 ± 0.03 .385 0.07 ± 0.03 0.08 ± 0.04 .039 0.07 ± 0.03 0.08 ± 0.03 .008 0.07 ± 0.03 0.08 ± 0.03 .113 0.09 ± 0.04 0.10 ± 0.05 .422
Open in a new tab

Note: The p values were calculated using a two‐sample t‐test except for the gender. The number of FIQ, ADHD index, inattentive, and hyper/impulsive available are described in the Table S1.

Abbreviations: ADHD, attention‐deficit/hyperactivity disorder; F, female; FD, framewise displacement; FIQ, full‐scale intelligence quotient; HC, healthy control; KKI, Kennedy Krieger Institute; M, male; n, number; NYU, New York University Medical Center; PU, Peking University.

FIGURE 1.

FIGURE 1

Open in a new tab

Flowchart of this study. (1) Radiomics features were extracted from each preprocessed rs‐fMRI image; (2) feature dimension reduction were performed using statistical analysis; (3) the SFS algorithm was used to select discriminative features; (4) a SVM model was built to classify ADHD subjects and HC. AAL, anatomical automatic labeling; ADHD, attention‐deficit/hyperactivity disorder; ALFF, amplitude of low frequency fluctuations; DC, network degree centrality; HC, healthy control; ReHo, regional homogeneity; rs‐fMRI, resting‐state functional magnetic resonance imaging; SFS, sequential forward selection; SVM, support vector machine; VMHC, voxel‐mirrored homotopic connectivity.

2.2. Imaging data preprocessing

The preprocessed imaging data were obtained from the R‐fMRI Maps Project (http://rfmri.org/maps). The image preprocessing pipeline was conducted using Data Processing Assistant for Resting‐State fMRI (http://rfmri.org/DPARSF), which is based on Statistical Parametric Mapping (SPM, http://www.fil.ion.ucl.ac.uk/spm/). The main steps applied to the rs‐fMRI data were as follows: (1) slice timing was corrected to ensure the stability of the blood oxygenation level‐dependent (BOLD) signal; (2) head motion was corrected by realigning to the reference scan; (3) covariates such as linear drift, white matter signal, and cerebrospinal fluid were removed without global signal regression; (4) the functional images were spatially normalized to the standard MNI space and were resampled to the voxel size of 3 × 3 × 3 mm3; (5) the data were filtered to 0.01–0.1 Hz to attenuate respiration and other high‐frequency physiological noises (except for ALFF calculation). The four preprocessed rs‐fMRI metrics including ALFF, ReHo, VMHC, and DC were used in this study. ALFF is defined as the mean value of the Fourier transform amplitude of each voxel time series in a specific frequency range. ReHo is defined as the Kendall's W between the time series of a given voxel and its nearest voxel time series. VMHC is defined as the Pearson correlation coefficient of BOLD signal time series of specific voxels and voxels at the same position in the contralateral hemisphere. DC is the number or sum of weights of significant connections for a voxel. Here, the functional connectivity (FC) of each voxel and the whole brain voxel level is calculated, and then the weighted sum of positive connection exceeding the threshold (threshold was set as Pearson correlation coefficient r > 0.25) was taken as the DC map. There were obvious outliers in some individual ReHo, ALFF, DC or VMHC map, defined as voxel values that lie beyond 1.5 times the interquartile range (IQR) measured from the first and the third quartile and these voxel values were replaced with the second quartile. Then, the individual ReHo, ALFF, and DC map of each voxel was further divided by the mean voxel value of whole brain for standardization. Meanwhile, VMHC was normalized to a symmetric template. Finally, we smoothed (FWHM = 6 mm) the map of each metric (except VMHC, it has been smoothed in processing) to improve the signal‐to‐noise ratio.

2.3. Feature extraction

The radiomics features were calculated using the PyRadiomics (https://pyradiomics.readthedocs.io/en/latest) package based on Python 3.7 (https://www.python.org/) in PyCharm (https://www.jetbrains.com/pycharm/ ). From each of four original (no filter applied) ReHo, ALFF, VMHC, and DC images, we extracted 93 radiomics features within each of 116 brain areas segmented by the anatomical automatic labeling (AAL) atlas (Tzourio‐Mazoyer et al., 2002), resulting in a total of 43,152 radiomics features. The 93 radiomics features included 18 first‐order features, 24 gray‐level cooccurrence matrix (GLCM) features, 16 gray‐level run‐length matrix (GLRLM) features, 16 gray‐level size zone matrix (GLSZM) features, 14 gray‐level dependence matrix (GLDM) features, and 5 neighborhood gray‐tone difference matrix (NGTDM) features. The PyRadiomics settings that control the process of feature extraction were specified including bin width (set to 25), interpolator (‘sitkBSpline’) and resampled pixel spacing (1 × 1 × 1 mm3). We calculated each radiomics features to compare them between ADHD subjects and HCs using the two‐sample two‐sided t‐test or two‐sided Mann–Whitney U test. For the radiomics features, p values less than .05 were considered to be significantly different in this study.

The mean signal features were calculated using the ROI Signal Extractor module in Data Processing & Analysis of Brain Imaging (DPABI, http://rfmri.org/dpabi) based on Matlab. The mean signal feature was defined as the mean value of all voxel values in a given brain region based on AAL atlas. Finally, we extracted 116 mean signal features within each rs‐fMRI metrics, resulting in a total of 464 mean signal features. We compared the mean signal features between ADHD subjects and HCs using the two‐sample two‐sided t‐test or two‐sided Mann–Whitney U test. As we hypothesized that radiomics features can better quantify imaging characteristics than the mean signal features within brain region, loose evidence of statistical significance (p < .1) was used in the mean signal‐based analysis (Ganesh & Cave, 2018).

2.4. Feature selection and classification

All patients were randomly allocated into the training and testing datasets (n = 300 and 74) using a stratified sampling method, in order to maintain the ADHD: HC ratio at 1:1 and the proportion of site was equal in each dataset. Descriptive statistics for the ADHD and HC groups in training and testing dataset are shown in Table S1. We standardized each feature independently on the samples in the training dataset, removing the mean and scaling to the unit variance, and then stored the mean and standard deviation so that it could later be applied to the testing dataset. All steps of feature dimension reduction, feature selection and model training were just produced in the training dataset.

Since the dimension of the features was extremely high and that may cause a dimensionality disaster easily, we reduced the redundancy (irrelevant and duplicate information) in vast radiomics features to speed the machine learning algorithm and improve the classification accuracy. We first tested whether each feature conforms to the normal distribution, and the two‐sample two‐sided t‐test was performed on those normally distributed features to calculate the differences between the ADHD group and the HC group. On the contrary, the two‐sided Mann–Whitney U test was used to test the features of non‐normal distribution. The features that were different between the two groups were retained (p < .1). Then, we removed the features with a strong pairwise correlation by Spearman's rank correlation coefficient. If two features had a high correlation (correlation coefficient > 0.8), we looked at the significance of each feature and retained the features with lower p value. We used the sequential forward selection (SFS) algorithm with 10‐fold cross‐validation to choose the most important features for classification only using data in the training dataset. The SFS algorithm first used all features to train the support vector machine (SVM) model and evaluated the classification loss in the 10‐fold cross‐validation. Each feature was sequentially added from the feature set to determine whether the classification loss decreased. If the loss was decreased, the features were added to the final feature set. The above steps were carried out in the features obtained from the four rs‐fMRI metrics separately, with the selected features in the training dataset merged to train and optimize a SVM model by a radial basis function (RBF) kernel. The coefficients which represent importance of 19 selected radiomics features were output through a linear kernel function. The radiomics features and mean signal features followed the same feature reduction and selection process. SVM is one of the most popular models used in neuroimaging‐based classification and can provide exact coefficient values for features, helping to target the heterogeneity of different brain regions. The hyperparameters of the SVM model were tuned and optimized in the training dataset by a 10‐fold cross‐validation and the trained model was determined the whole training dataset. Finally, the trained SVM model was used to classify the labels of subjects in the testing dataset.

In addition, we compared different feature selection and model building algorithms. Description and results are shown in Appendix S1.

All analyses above were carried out using Matlab 2019b.

2.5. Validation analysis

A receiver operating characteristic (ROC) curve illustrated the classification ability of the trained model, while an area under curve (AUC) was simultaneously calculated. The permutation test with 1000 permutations was used to assess whether the model was statistically significant (p < .05) (Liu et al., 2015).

In further analysis, the Spearman correlation analysis was performed to calculate the relationships between the final selected radiomics features and clinical measures including ADHD index, inattentive, and hyper/impulsive in ADHD patients. The effects of ADHD scale versions were removed from these clinical measures by a generalized linear model (GLM). In this procedure, subjects with missing measures were excluded (Table S2).

In particular, the effects of head motion should be carefully considered in fMRI research. At first, we discarded those subjects with head motion (>3.0 mm of translation or degrees of rotation in any direction), because the functional signals might be corrupted with severe head motions. Then, we compared the mean framewise displacement (FD) between the ADHD and HC groups by a two‐sample t‐tests. A SVM model was trained only using motion‐related parameters of mean FD, number of FD > 0.2 and percent of FD > 0.2.

3. RESULTS

3.1. Statistical analysis

In Table 1, we provide the complete demographic and clinical characteristics for subjects participated in classification process of this study. Across the five sites, there were no significant differences in age and gender between ADHD patients and HC after propensity score matching. The ADHD index, inattentive, and hyper/impulsive were significantly different between ADHD patients and HC within five sites (all p < .001 based on a two‐sample two‐sided t‐test). The site NYU, PU 1, PU 2, and PU 3 had significant difference in FIQ between the two groups, while the site KKI did not. In this procedure, subjects with missing measures were excluded. The number of ADHD and HC subjects with available clinical measures across five sites are shown in Table S2.

In terms of the mean signal of 116 AAL regions, there were 20 brain regions showed a trend to be statistically different (p < .1) between ADHD subjects and HCs in at least three rs‐fMRI metrics at the same time (Table S3 and Figure 2a). For the radiomics features, there were eight brain regions showed significant differences (p < .05) between ADHD subjects and HCs in four rs‐fMRI metrics at the same time (Tables S4–S7 and Figure 2b). There was a spatial concordance between the results of conventional analysis (the mean signal of brain region) and the results of radiomics analysis. In addition, radiomics further explored more brain regions with significant differences.

FIGURE 2.

FIGURE 2

Open in a new tab

The significant brain regions in the conventional analysis or the radiomics. In (a, b), the colors represent brain regions simultaneously selected at least three types of preprocessed images across different radiomics feature types. In (c), the brain regions anchored by the finally selected radiomics features used to construct the SVM model were shown. ALFF, amplitude of low frequency fluctuations; DC, network degree centrality; ReHo, regional homogeneity; SVM, support vector machine; VMHC, voxel‐mirrored homotopic connectivity.

3.2. Feature dimensionality reduction and selection

After the dimensionality reduction based on two samples t‐test, Mann–Whitney U test and Spearman correlation, 43,152 radiomics features (93 × 116 × 4) from the four preprocessed rs‐fMRI images were reduced to 3321 potential variables. Among them, 19 radiomics features were selected for SVM model building, including 5 features from ALFF, 9 features from ReHo, 3 features from VMHC, and 2 features from DC (Figure 2c). We listed the final selected radiomics features and corresponding brain regions indexed in the AAL template and the coefficients which represent importance of 19 selected radiomics features were shown in Table S8. Figure 3 shows boxplots of the mean signal features and radiomics features of representative brain regions. Radiomics can discriminate ADHD subjects from HCs with stronger evidence of statistical differences than the mean signal features.

FIGURE 3.

FIGURE 3

Open in a new tab

Boxplots of the mean signal (a, c, e, g) features and the radiomics features (b, d, f, h) of representative brain regions. ADHD, attention‐deficit/hyperactivity disorder; ALFF, amplitude of low frequency fluctuations; DC, network degree centrality; glszm, gray level size zone matrix; HC, healthy control; ReHo, regional homogeneity; VMHC, voxel‐mirrored homotopic connectivity.

3.3. Classification performance

The accuracies of SVM model built using the mean signal features were 60.1% (AUC = 0.636) in the training dataset and 59.5% (AUC = 0.618) in the testing dataset (Figure 4). The SVM model, which was trained using the selected 19 radiomics features of ALFF, ReHo, VMHC, and DC, achieved an accuracy of 76.3% (AUC = 0.811) in the training dataset and an accuracy of 77.0% (p = .007, DeLong's test; AUC = 0.797; permutation test p < .001) in the testing dataset (Figure 4). In addition, Table 2 summarizes the detailed performance for classifying ADHD patients and HC. Figure 5 shows the confusion matrices of SVM models that were built using radiomics features or the mean signal features in the training and testing dataset.

FIGURE 4.

FIGURE 4

Open in a new tab

Receiver operating characteristic curves of the support vector machine models in the training dataset (a) and the testing dataset (b). AUC, area under curve.

TABLE 2.

The classification results of ADHD patients and HCs in the training and testing datasets.

Training dataset Testing dataset
Accuracy AUC Sensitivity Specificity Accuracy AUC Sensitivity Specificity
Mean signals 60.1% 0.636 57.3% 64.0% 59.5% 0.618 64.9% 54.1%
Radiomics 76.3% 0.811 78.7% 74.0% 77.0% 0.797 83.8% 70.3%
Open in a new tab

Abbreviations: ADHD, attention‐deficit/hyperactivity disorder; AUC, area under curve; HC, healthy control.

FIGURE 5.

FIGURE 5

Open in a new tab

Confusion matrices of support vector machine models that were built using the mean signal features and the radiomics features in the training dataset (a, c) and the testing dataset (b, d). Each column of the confusion matrix represents the instances in an actual class and their percentages in all datasets, while each row represents the instances in a predicted class and their percentages. The numbers in the right column and the lower row are the accuracy and loss for each row and each column, respectively. The number on the lower right is the total accuracy and loss. ADHD, attention‐deficit/hyperactivity disorder; HC, healthy control.

3.4. Validation analysis

In the Spearman correlation analysis, we found that the two final selected radiomics features were significantly associated with ADHD index, inattentive, and hyper/impulsive at the same time (Figure 6, r: −0.181 to 0.468; p: <.0001–.016). The remaining radiomics features significantly associated with the above ADHD measures are shown in Figure S1.

FIGURE 6.

FIGURE 6

Open in a new tab

The final selected radiomics features that were significantly associated with ADHD index (a, d), inattentive (b, e), and hyper/impulsive (c, f) at the same time. ADHD, attention‐deficit/hyperactivity disorder; ALFF, amplitude of low frequency fluctuations; DC, network degree centrality.

The mean FD between the ADHD and HC groups did not significantly differ at all sites except in the NYU and PU 1 sites (Table 1). However, the classification accuracy of the SVM model built by motion‐related parameters of our training dataset was only 58.1% (p = .006, DeLong's test; AUC = 0.584) in the testing dataset.

4. DISCUSSION

This study presented a novel framework for quantitative diagnosis of ADHD based on radiomics features extracted from preprocessed rs‐fMRI images, which obtained good performance in classifying ADHD patients and HC. We found that the model classification performance based on radiomics features was significantly higher than the mean signal features. Furthermore, some selected radiomics features showed significant associations with the clinical measures of ADHD (ADHD index, inattentive, and hyper/impulsive). Our results indicated that the radiomics method can effectively capture fMRI information and has the potential to find neuroimaging biomarkers for ADHD.

The aim to diagnose ADHD according to neuroimaging data has long been pursued. The largest attempt to classify ADHD using neuroimaging data comes from the ADHD‐200 Global Competition, which collected rs‐fMRI and T1‐weighted MRI data of 285 ADHD patients and 491 HC (Consortium, 2012). The best result achieved high specificity of 94%, but poor sensitivity of 21%, the overall accuracy was only 61%. One possible explanation of poor results was that the information of MRI data was not fully used. A previous study used a variety of imaging measures based on both structural MRI and rs‐fMRI to distinguish ADHD patients and showed a great classification performance (Qureshi et al., 2017). Compared with their study, our model was established using only a single imaging modality thus was more simple, effective, and more conducive to future clinical application. In addition, we collected more data from several different sites, making our proposed model more robust and classification performance more reliable. In our study, a power adjunct analysis framework based on radiomics was established to improve the classification accuracy and reliability of model. By incorporating radiomics features, our classifier achieved the best accuracy of 77.0% (sensitivity: 83.8%; specificity: 70.3%) for discriminating individuals with ADHD from HC in the testing dataset, somewhat better than results of those studies using the ADHD‐200 database.

Previously, almost all studies based on rs‐fMRI to identify ADHD were conducted through brain functional connection of different brain areas (Chen et al., 2020; Riaz et al., 2020; Sun et al., 2020; Tang et al., 2021). Our study innovatively used high‐throughput and powerful discriminative radiomics features on rs‐fMRI to diagnose ADHD. The great discriminative power of radiomics enables it to capture minor brain activity signals thus may be used to further explore the brain abnormal function mechanism of ADHD patients in future research. Additionally, the radiomics features were extracted based on independent brain regions and features of different brain regions were combined to establish machine learning models. This can help to explore neuroimaging markers based on not only the association of brain areas but also features of independent brain regions. Furthermore, our results were higher than most of previous studies (Chen et al., 2020; Riaz et al., 2020; Sun et al., 2020; Tang et al., 2021). Our results showed that more image information can be obtained from independent brain regions and used for objective and reliable diagnosis of ADHD based on radiomics, which to some extent filled the research gap of identifying ADHD patients based on independent brain region on rs‐fMRI.

In this study, we discovered that the radiomics features obtained in rs‐fMRI could capture the changes of brain activity within ADHD patients. A meta‐analysis showed consistent deficits in the left inferior prefrontal, parietal, and cerebellar regions of individuals with ADHD (Hart et al., 2012). Another meta‐analysis of task‐related fMRI studies reported hypo‐activation in the left putamen, inferior frontal gyrus, temporal pole, and right caudate of ADHD patients (Cortese et al., 2016). A previous study reported that frontal and cerebellar regions showed top discriminative power in classifying ADHD and HCs (Cheng et al., 2012). Another study used ReHo to discriminate between ADHD and HC, and found that the most discriminative brain regions included the prefrontal cortex, anterior cingulate cortex, and cerebellum (Zhu et al., 2008). In addition, a plethora of functional imaging investigations reveal that frontostriatal connectivity is highly related to ADHD symptom patterns. Overall, frontal, and cerebellar regions appear to be the leading candidates for exploring disorder and progression biomarkers of ADHD. It is worth noting that our results also emphasized the two brain regions for the diagnosis of ADHD. However, some researchers caution that multiple areas and pathways are likely to be implicated in ADHD‐related behaviors, far beyond the frontostriatal model per se (Castellanos & Proal, 2012). This point was confirmed by the well‐known triple‐network dysfunction model of pathophysiology associated with multiple psychiatric disorders (Menon, 2011). A meta‐analysis showed that the ADHD was associated with hypoconnectivity within the default mode network (DMN) and there was imbalanced connectivity between the frontoparietal network and regions of the DMN and ventral attention network (Gao et al., 2019). Beyond this model, dysconnectivity was also found between these three networks and two other functional networks: the somatosensory network and the affective network (Gao et al., 2019). In this study, the radiomics features we obtained come from a wide range of brain regions, which are well unified with the brain regions contained in each brain network. Therefore, the radiomics may be potential to reflect the dysfunction of brain networks.

Notably, we found that a considerable part of the selected radiomics features in our study came from the cerebellum. There is cumulative evidence suggesting that the cerebellum is not only important in motor learning and coordination but also plays a role in cognition and emotion. Meanwhile, there is an increasing appreciation of its role in neurodevelopmental disorders such as ADHD. In ADHD, smaller cerebellar volumes were reported in quantitative study of brain morphometry (Stoodley, 2014). ADHD symptom severity has been shown to correlate with the degree of reduction in the posterior vermis (Ivanov et al., 2014) and overall cerebellar volume (Castellanos et al., 2002). A meta‐analysis found consistent gray matter reductions in ADHD bilaterally in lobule IX (Stoodley, 2014). Further, differences in both structural (such as in the middle cerebellar peduncles [Bechtel et al., 2009]) and functional cerebellar connectivity (Tomasi & Volkow, 2012) have been reported in ADHD. Preliminary associations have been found between the adolescent course of ADHD and earlier measures of neuroanatomy (specifically, dimensions of the anterior cingulate cortex and superior cerebellar vermis) and functional connectivity (coupling between activity in the medial and dorsolateral prefrontal cortex) (Mackie et al., 2007; Shaw et al., 2006; Whitfield‐Gabrieli et al., 2020). The concept of the cerebellar connectome may link the cerebellar development with human behavior, disease status, and the design of better treatment strategies.

The present study has several limitations. The classification for subtypes of ADHD has not been considered in this study due to the large difference in the proportion of each ADHD subtype in the ADHD‐200 database. At the same time, only right‐handed subjects were included in this study. This limited the generalization of these results to left‐handed subjects with ADHD. Despite the robust classification accuracy was attained across sites for distinguishing the ADHD and HC groups, further work with multi‐center imaging datasets, narrower age ranges, and more standardized protocols for diagnosis and medication is needed to assess the reproducibility of our findings and develop clinically useful biomarkers. There are much more rs‐fMRI metrics except ALFF, ReHo, VMHC, and DC are available in current research field. Future studies will enroll other rs‐fMRI metrics to see whether they can improve the classification performance. Finally, possible pathological rationales of rs‐fMRI‐based radiomics have not been investigated in this study. Future studies should be designed to further explore the possible mechanisms behind the radiomics‐quantified brain activity alterations in ADHD before its clinical implementation.

5. CONCLUSION

This study found that radiomics features extracted from rs‐fMRI can be used for identifying ADHD patients from HC with better classification performance than mean signal features based on multisite dataset, providing a potential adjunctive approach to clinical diagnostic systems. Radiomics features may reflect information about the pathology of ADHD, and with additional research, the quantitative method of radiomics could potentially be used to develop robust multisite neuroimaging biomarkers for the ADHD.

AUTHOR CONTRIBUTIONS

Liting Shi and Jianfeng Qiu conceived of the presented idea and supervised the project. Guanlu Liu, Liting Shi, and Weizhao Lu designed the computational model. Guanlu Liu downloaded the sample, analyzed the data, and wrote the manuscript. All authors provided critical feedback and helped shape the analysis and manuscript.

FUNDING INFORMATION

This study was supported by Taishan Scholars Program of Shandong Province [No. ts201712065], Academic promotion programme of Shandong First Medical University [2019QL009], and Science and technology funding from Jinan [No. 2020GXRC018]. Data were provided by the ADHD‐200 Consortium.

CONFLICT OF INTEREST STATEMENT

The authors declare no potential conflicts of interest.

Supporting information

Appendix S1: Supporting information

Click here for additional data file. (800.5KB, docx)

ACKNOWLEDGMENTS

We thank LetPub (www.letpub.com) for its linguistic assistance during the preparation of this manuscript.

Liu, G. , Lu, W. , Qiu, J. , & Shi, L. (2023). Identifying individuals with attention‐deficit/hyperactivity disorder based on multisite resting‐state functional magnetic resonance imaging: A radiomics analysis. Human Brain Mapping, 44(8), 3433–3445. 10.1002/hbm.26290

Contributor Information

Jianfeng Qiu, Email: jfqiu100@gmail.com.

Liting Shi, Email: ltshi@foxmail.com.

DATA AVAILABILITY STATEMENT

Data were provided by the ADHD‐200 Consortium in the international neuroimaging data‐sharing initiative datasets (http://fcon_1000.projects.nitrc.org/indi/adhd200/index.html).

REFERENCES

  1. Arbabshirani, M. R. , Plis, S. , Sui, J. , & Calhoun, V. D. (2017). Single subject prediction of brain disorders in neuroimaging: Promises and pitfalls. NeuroImage, 145(Pt B), 137–165. 10.1016/j.neuroimage.2016.02.079 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Bechtel, N. , Kobel, M. , Penner, I. K. , Klarhofer, M. , Scheffler, K. , Opwis, K. , & Weber, P. (2009). Decreased fractional anisotropy in the middle cerebellar peduncle in children with epilepsy and/or attention deficit/hyperactivity disorder: A preliminary study. Epilepsy & Behavior, 15(3), 294–298. 10.1016/j.yebeh.2009.04.005 [DOI] [PubMed] [Google Scholar]
  3. Bos, D. J. , Oranje, B. , Achterberg, M. , Vlaskamp, C. , Ambrosino, S. , de Reus, M. A. , van den Heuvel, M. P. , Rombouts, S. A. R. B. , & Durston, S. (2017). Structural and functional connectivity in children and adolescents with and without attention deficit/hyperactivity disorder. Journal of Child Psychology and Psychiatry, 58(7), 810–818. 10.1111/jcpp.12712 [DOI] [PubMed] [Google Scholar]
  4. Buckner, R. L. , Sepulcre, J. , Talukdar, T. , Krienen, F. M. , Liu, H. , Hedden, T. , Andrews‐Hanna, J. R. , Sperling, R. A. , & Johnson, K. A. (2009). Cortical hubs revealed by intrinsic functional connectivity: Mapping, assessment of stability, and relation to Alzheimer's disease. The Journal of Neuroscience, 29(6), 1860–1873. 10.1523/JNEUROSCI.5062-08.2009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Castellanos, F. X. , Lee, P. P. , Sharp, W. , Jeffries, N. O. , Greenstein, D. K. , Clasen, L. S. , Blumentha, J. D. , James, R. S. , Ebens, C. L. , Walter, J. M. , Zijdenbos, A. , Evans, A. C. , Giedd, J. N. , & Rapoport, J. L. (2002). Developmental trajectories of brain volume abnormalities in children and adolescents with attention‐deficit/hyperactivity disorder. JAMA, 288(14), 1740–1748. 10.1001/jama.288.14.1740 [DOI] [PubMed] [Google Scholar]
  6. Castellanos, F. X. , & Proal, E. (2012). Large‐scale brain systems in ADHD: Beyond the prefrontal‐striatal model. Trends in Cognitive Sciences, 16(1), 17–26. 10.1016/j.tics.2011.11.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Chen, Y. , Tang, Y. , Wang, C. , Liu, X. , Zhao, L. , & Wang, Z. (2020). ADHD classification by dual subspace learning using resting‐state functional connectivity. Artificial Intelligence in Medicine, 103, 101786. 10.1016/j.artmed.2019.101786 [DOI] [PubMed] [Google Scholar]
  8. Cheng, W. , Ji, X. , Zhang, J. , & Feng, J. (2012). Individual classification of ADHD patients by integrating multiscale neuroimaging markers and advanced pattern recognition techniques. Frontiers in Systems Neuroscience, 6, 58. 10.3389/fnsys.2012.00058 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Consortium, H. D. (2012). The ADHD‐200 Consortium: A model to advance the translational potential of neuroimaging in clinical neuroscience. Frontiers in Systems Neuroscience, 6, 62. 10.3389/fnsys.2012.00062 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Conti, A. , Duggento, A. , Indovina, I. , Guerrisi, M. , & Toschi, N. (2021). Radiomics in breast cancer classification and prediction. Seminars in Cancer Biology, 72, 238–250. 10.1016/j.semcancer.2020.04.002 [DOI] [PubMed] [Google Scholar]
  11. Cortese, S. , Castellanos, F. X. , Eickhoff, C. R. , D'Acunto, G. , Masi, G. , Fox, P. T. , Laird, A. R. , & Eickhoff, S. B. (2016). Functional decoding and meta‐analytic connectivity modeling in adult attention‐deficit/hyperactivity disorder. Biological Psychiatry, 80(12), 896–904. 10.1016/j.biopsych.2016.06.014 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. de Lacy, N. , Kodish, I. , Rachakonda, S. , & Calhoun, V. D. (2018). Novel in silico multivariate mapping of intrinsic and anticorrelated connectivity to neurocognitive functional maps supports the maturational hypothesis of ADHD. Human Brain Mapping, 39(8), 3449–3467. 10.1002/hbm.24187 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. de Zeeuw, P. , Schnack, H. G. , van Belle, J. , Weusten, J. , van Dijk, S. , Langen, M. , Brouwer, R. M. , van Engeland, H. , & Durston, S. (2012). Differential brain development with low and high IQ in attention‐deficit/hyperactivity disorder. PLoS One, 7(4), e35770. 10.1371/journal.pone.0035770 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Dennis, M. , Francis, D. J. , Cirino, P. T. , Schachar, R. , Barnes, M. A. , & Fletcher, J. M. (2009). Why IQ is not a covariate in cognitive studies of neurodevelopmental disorders. Journal of the International Neuropsychological Society, 15(3), 331–343. 10.1017/S1355617709090481 [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Ganesh, S. , & Cave, V. (2018). P‐values, p‐values everywhere! New Zealand Veterinary Journal, 66(2), 55–56. 10.1080/00480169.2018.1415604 [DOI] [PubMed] [Google Scholar]
  16. Gao, Y. , Shuai, D. , Bu, X. , Hu, X. , Tang, S. , Zhang, L. , Li, H. , Hu, X. , Lu, L. , Gong, Q. , & Huang, X. (2019). Impairments of large‐scale functional networks in attention‐deficit/hyperactivity disorder: A meta‐analysis of resting‐state functional connectivity. Psychological Medicine, 49(15), 2475–2485. 10.1017/S003329171900237X [DOI] [PubMed] [Google Scholar]
  17. Gillies, R. J. , Kinahan, P. E. , & Hricak, H. (2016). Radiomics: Images are more than pictures, they are data. Radiology, 278(2), 563–577. 10.1148/radiol.2015151169 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Hart, H. , Radua, J. , Mataix‐Cols, D. , & Rubia, K. (2012). Meta‐analysis of fMRI studies of timing in attention‐deficit hyperactivity disorder (ADHD). Neuroscience and Biobehavioral Reviews, 36(10), 2248–2256. 10.1016/j.neubiorev.2012.08.003 [DOI] [PubMed] [Google Scholar]
  19. Huys, Q. J. , Maia, T. V. , & Frank, M. J. (2016). Computational psychiatry as a bridge from neuroscience to clinical applications. Nature Neuroscience, 19(3), 404–413. 10.1038/nn.4238 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Ivanov, I. , Murrough, J. W. , Bansal, R. , Hao, X. , & Peterson, B. S. (2014). Cerebellar morphology and the effects of stimulant medications in youths with attention deficit‐hyperactivity disorder. Neuropsychopharmacology, 39(3), 718–726. 10.1038/npp.2013.257 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Jie, B. , Wee, C. Y. , Shen, D. , & Zhang, D. (2016). Hyper‐connectivity of functional networks for brain disease diagnosis. Medical Image Analysis, 32, 84–100. 10.1016/j.media.2016.03.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Kelly, C. , Zuo, X. N. , Gotimer, K. , Cox, C. L. , Lynch, L. , Brock, D. , Imperati, D. , Garavan, H. , Rotrosen, J. , Castellanos, F. X. , & Milham, M. P. (2011). Reduced interhemispheric resting state functional connectivity in cocaine addiction. Biological Psychiatry, 69(7), 684–692. 10.1016/j.biopsych.2010.11.022 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Kim, J. I. , Yoo, J. H. , Kim, D. , Jeong, B. , & Kim, B. N. (2018). The effects of GRIN2B and DRD4 gene variants on local functional connectivity in attention‐deficit/hyperactivity disorder. Brain Imaging and Behavior, 12(1), 247–257. 10.1007/s11682-017-9690-2 [DOI] [PubMed] [Google Scholar]
  24. Kumar, V. , Gu, Y. , Basu, S. , Berglund, A. , Eschrich, S. A. , Schabath, M. B. , Forster, K. , Aerts, H. J. W. L. , Dekker, A. , Fenstermacher, D. , Goldgof, D. B. , Hall, L. O. , Lambin, P. , Balagurunathan, Y. , Gatenby, R. A. , & Gillies, R. J. (2012). Radiomics: The process and the challenges. Magnetic Resonance Imaging, 30(9), 1234–1248. 10.1016/j.mri.2012.06.010 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Leung, A. K. , & Hon, K. L. (2016). Attention‐deficit/hyperactivity disorder. Advances in Pediatrics, 63(1), 255–280. 10.1016/j.yapd.2016.04.017 [DOI] [PubMed] [Google Scholar]
  26. Liu, D. , Yan, C. , Ren, J. , Yao, L. , Kiviniemi, V. J. , & Zang, Y. (2010). Using coherence to measure regional homogeneity of resting‐state FMRI signal. Frontiers in Systems Neuroscience, 4, 24. 10.3389/fnsys.2010.00024 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Liu, F. , Guo, W. , Fouche, J. P. , Wang, Y. , Wang, W. , Ding, J. , Zeng, L. , Qiu, C. , Gong, Q. , Zhang, W. , & Chen, H. (2015). Multivariate classification of social anxiety disorder using whole brain functional connectivity. Brain Structure & Function, 220(1), 101–115. 10.1007/s00429-013-0641-4 [DOI] [PubMed] [Google Scholar]
  28. Mackie, S. , Shaw, P. , Lenroot, R. , Pierson, R. , Greenstein, D. K. , Nugent, T. F., 3rd , Sharp, W. S. , Giedd, J. N. , & Rapoport, J. L. (2007). Cerebellar development and clinical outcome in attention deficit hyperactivity disorder. The American Journal of Psychiatry, 164(4), 647–655. 10.1176/ajp.2007.164.4.647 [DOI] [PubMed] [Google Scholar]
  29. Mayerhoefer, M. E. , Materka, A. , Langs, G. , Haggstrom, I. , Szczypinski, P. , Gibbs, P. , & Cook, G. (2020). Introduction to radiomics. Journal of Nuclear Medicine, 61(4), 488–495. 10.2967/jnumed.118.222893 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Menon, V. (2011). Large‐scale brain networks and psychopathology: A unifying triple network model. Trends in Cognitive Sciences, 15(10), 483–506. 10.1016/j.tics.2011.08.003 [DOI] [PubMed] [Google Scholar]
  31. Miao, B. , Zhang, L. L. , Guan, J. L. , Meng, Q. F. , & Zhang, Y. L. (2019). Classification of ADHD individuals and neurotypicals using reliable RELIEF: A resting‐state study. IEEE Access, 7, 62163–62171. 10.1109/Access.2019.2915988 [DOI] [Google Scholar]
  32. Nigg, J. T. (2013). Attention‐deficit/hyperactivity disorder and adverse health outcomes. Clinical Psychology Review, 33(2), 215–228. 10.1016/j.cpr.2012.11.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Polanczyk, G. , & Jensen, P. (2008). Epidemiologic considerations in attention deficit hyperactivity disorder: A review and update. Child and Adolescent Psychiatric Clinics of North America, 17(2), 245–260, vii. 10.1016/j.chc.2007.11.006 [DOI] [PubMed] [Google Scholar]
  34. Qureshi, M. N. I. , Oh, J. , Min, B. , Jo, H. J. , & Lee, B. (2017). Multi‐modal, multi‐measure, and multi‐class discrimination of ADHD with hierarchical feature extraction and extreme learning machine using structural and functional brain MRI. Frontiers in Human Neuroscience, 11, 157. 10.3389/fnhum.2017.00157 [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Riaz, A. , Asad, M. , Alonso, E. , & Slabaugh, G. (2018). Fusion of fMRI and non‐imaging data for ADHD classification. Computerized Medical Imaging and Graphics, 65, 115–128. 10.1016/j.compmedimag.2017.10.002 [DOI] [PubMed] [Google Scholar]
  36. Riaz, A. , Asad, M. , Alonso, E. , & Slabaugh, G. (2020). DeepFMRI: End‐to‐end deep learning for functional connectivity and classification of ADHD using fMRI. Journal of Neuroscience Methods, 335, 108506. 10.1016/j.jneumeth.2019.108506 [DOI] [PubMed] [Google Scholar]
  37. Savage, N. (2017). Machine learning: Calculating disease. Nature, 550(7676), S115–S117. 10.1038/550S115a [DOI] [PubMed] [Google Scholar]
  38. Schwarz, E. , Doan, N. T. , Pergola, G. , Westlye, L. T. , Kaufmann, T. , Wolfers, T. , Brecheisen, R. , Quarto, T. , Ing, A. J. , Di Carlo, P. , Gurholt, T. P. , Harms, R. L. , Noirhomme, Q. , Moberget, T. , Agartz, I. , Andreassen, O. A. , Bellani, M. , Bertolino, A. , Blasi, G. , … Karolinska Schizophrenia Project (KaSP) Consortium . (2019). Reproducible grey matter patterns index a multivariate, global alteration of brain structure in schizophrenia and bipolar disorder. Translational Psychiatry, 9(1), 12. 10.1038/s41398-018-0225-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Shang, C. Y. , Lin, H. Y. , & Gau, S. S. (2021). The norepinephrine transporter gene modulates intrinsic brain activity, visual memory, and visual attention in children with attention‐deficit/hyperactivity disorder. Molecular Psychiatry, 26(8), 4026–4035. 10.1038/s41380-019-0545-7 [DOI] [PubMed] [Google Scholar]
  40. Shao, L. , Xu, Y. , & Fu, D. (2018). Classification of ADHD with bi‐objective optimization. Journal of Biomedical Informatics, 84, 164–170. 10.1016/j.jbi.2018.07.011 [DOI] [PubMed] [Google Scholar]
  41. Sharma, A. , & Couture, J. (2014). A review of the pathophysiology, etiology, and treatment of attention‐deficit hyperactivity disorder (ADHD). The Annals of Pharmacotherapy, 48(2), 209–225. 10.1177/1060028013510699 [DOI] [PubMed] [Google Scholar]
  42. Shaw, P. , Lerch, J. , Greenstein, D. , Sharp, W. , Clasen, L. , Evans, A. , Giedd, J. , Castellanos, F. X. , & Rapoport, J. (2006). Longitudinal mapping of cortical thickness and clinical outcome in children and adolescents with attention‐deficit/hyperactivity disorder. Archives of General Psychiatry, 63(5), 540–549. 10.1001/archpsyc.63.5.540 [DOI] [PubMed] [Google Scholar]
  43. Shaw, P. , Stringaris, A. , Nigg, J. , & Leibenluft, E. (2014). Emotion dysregulation in attention deficit hyperactivity disorder. The American Journal of Psychiatry, 171(3), 276–293. 10.1176/appi.ajp.2013.13070966 [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Stoodley, C. J. (2014). Distinct regions of the cerebellum show gray matter decreases in autism, ADHD, and developmental dyslexia. Frontiers in Systems Neuroscience, 8, 92. 10.3389/fnsys.2014.00092 [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Sun, Y. , Zhao, L. , Lan, Z. , Jia, X. Z. , & Xue, S. W. (2020). Differentiating boys with ADHD from those with typical development based on whole‐brain functional connections using a machine learning approach. Neuropsychiatric Disease and Treatment, 16, 691–702. 10.2147/NDT.S239013 [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Tang, Y. , Wang, C. , Chen, Y. , Sun, N. , Jiang, A. , & Wang, Z. (2021). Identifying ADHD individuals from resting‐state functional connectivity using subspace clustering and binary hypothesis testing. Journal of Attention Disorders, 25(5), 736–748. 10.1177/1087054719837749 [DOI] [PubMed] [Google Scholar]
  47. Tomasi, D. , & Volkow, N. D. (2012). Abnormal functional connectivity in children with attention‐deficit/hyperactivity disorder. Biological Psychiatry, 71(5), 443–450. 10.1016/j.biopsych.2011.11.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Tzourio‐Mazoyer, N. , Landeau, B. , Papathanassiou, D. , Crivello, F. , Etard, O. , Delcroix, N. , Mazoyer, B. , & Joliot, M. (2002). Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single‐subject brain. NeuroImage, 15(1), 273–289. 10.1006/nimg.2001.0978 [DOI] [PubMed] [Google Scholar]
  49. Wang, X. H. , Jiao, Y. , & Li, L. (2018). Identifying individuals with attention deficit hyperactivity disorder based on temporal variability of dynamic functional connectivity. Scientific Reports, 8(1), 11789. 10.1038/s41598-018-30308-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Wang, Y. , Sun, K. , Liu, Z. , Chen, G. , Jia, Y. , Zhong, S. , Pan, J. , Huang, L. , & Tian, J. (2020). Classification of unmedicated bipolar disorder using whole‐brain functional activity and connectivity: A radiomics analysis. Cerebral Cortex, 30(3), 1117–1128. 10.1093/cercor/bhz152 [DOI] [PubMed] [Google Scholar]
  51. Weiler, M. D. , Bellinger, D. K. , Simmons, E. K. , Rappaport, L. K. , Urion, D. K. , Mitchell, W. J. , Bassett, N. , Burke, P. J. , Marmor, J. , & Waber, D. (2000). Reliability and validity of a DSM‐IV based ADHD screener. Child Neuropsychology, 6(1), 3–23. 10.1076/0929-7049(200003)6:1;1-B;FT003 [DOI] [PubMed] [Google Scholar]
  52. Whitfield‐Gabrieli, S. , Wendelken, C. , Nieto‐Castanon, A. , Bailey, S. K. , Anteraper, S. A. , Lee, Y. J. , Chai, X.‐Q. , Hirshfeld‐Becker, D. R. , Biederman, J. , Cutting, L. E. , & Bunge, S. A. (2020). Association of intrinsic brain architecture with changes in attentional and mood symptoms during development. JAMA Psychiatry, 77(4), 378–386. 10.1001/jamapsychiatry.2019.4208 [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Willcutt, E. G. (2012). The prevalence of DSM‐IV attention‐deficit/hyperactivity disorder: A meta‐analytic review. Neurotherapeutics, 9(3), 490–499. 10.1007/s13311-012-0135-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Wilson, V. B. , Mitchell, S. H. , Musser, E. D. , Schmitt, C. F. , & Nigg, J. T. (2011). Delay discounting of reward in ADHD: Application in young children. Journal of Child Psychology and Psychiatry, 52(3), 256–264. 10.1111/j.1469-7610.2010.02347.x [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Yoo, J. H. , Kim, D. , Choi, J. , & Jeong, B. (2018). Treatment effect of methylphenidate on intrinsic functional brain network in medication‐naive ADHD children: A multivariate analysis. Brain Imaging and Behavior, 12(2), 518–531. 10.1007/s11682-017-9713-z [DOI] [PubMed] [Google Scholar]
  56. Zang, Y. , Jiang, T. , Lu, Y. , He, Y. , & Tian, L. (2004). Regional homogeneity approach to fMRI data analysis. NeuroImage, 22(1), 394–400. 10.1016/j.neuroimage.2003.12.030 [DOI] [PubMed] [Google Scholar]
  57. Zhou, X. , Reynolds, C. R. , Zhu, J. , Kamphaus, R. W. , & Zhang, O. (2018). Evidence‐based assessment of ADHD diagnosis in children and adolescents. Applied Neuropsychology: Child, 7(2), 150–156. 10.1080/21622965.2017.1284661 [DOI] [PubMed] [Google Scholar]
  58. Zhu, C. Z. , Zang, Y. F. , Cao, Q. J. , Yan, C. G. , He, Y. , Jiang, T. Z. , Sui, M.‐Q. , & Wang, Y. F. (2008). Fisher discriminative analysis of resting‐state brain function for attention‐deficit/hyperactivity disorder. NeuroImage, 40(1), 110–120. 10.1016/j.neuroimage.2007.11.029 [DOI] [PubMed] [Google Scholar]
  59. Zou, Q. H. , Zhu, C. Z. , Yang, Y. , Zuo, X. N. , Long, X. Y. , Cao, Q. J. , Wang, Y.‐F. , & Zang, Y. F. (2008). An improved approach to detection of amplitude of low‐frequency fluctuation (ALFF) for resting‐state fMRI: Fractional ALFF. Journal of Neuroscience Methods, 172(1), 137–141. 10.1016/j.jneumeth.2008.04.012 [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Zuo, X. N. , Kelly, C. , Di Martino, A. , Mennes, M. , Margulies, D. S. , Bangaru, S. , Grzadzinski, R. , Evans, A. C. , Zang, Y.‐F. , Castellanos, F. X. , & Milham, M. P. (2010). Growing together and growing apart: Regional and sex differences in the lifespan developmental trajectories of functional homotopy. The Journal of Neuroscience, 30(45), 15034–15043. 10.1523/JNEUROSCI.2612-10.2010 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

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

Supplementary Materials

Appendix S1: Supporting information

Click here for additional data file. (800.5KB, docx)

Data Availability Statement

Data were provided by the ADHD‐200 Consortium in the international neuroimaging data‐sharing initiative datasets (http://fcon_1000.projects.nitrc.org/indi/adhd200/index.html).

Articles from Human Brain Mapping are provided here courtesy of Wiley

RESOURCES