Abstract
Background
Glioma irradiation often unavoidably damages the brain volume and affects cognition. This study aims to evaluate the relationship of remote cognitive assessments in determining cognitive impairment of irradiated glioma patients in relation to the quality of life and MRI changes.
Methods
Thirty patients (16–76 aged) with two imaging (pre- and post-RT) and completed cognitive assessments were recruited. Cerebellum, right and left temporal lobes, corpus callosum, amygdala and spinal cord were delineated and their dosimetry parameters were collected. Cognitive assessments were given post-RT via telephone (Telephone Interview Cognitive Status (TICS), Telephone Montreal Cognitive Assessment (T-MoCA), Telephone Mini Addenbrooke’s Cognitive Examination (Tele-MACE)). Regression models and deep neural network (DNN) were used to evaluate the relationship between brain volume, cognition and treatment dose in patients.
Results
Cognitive assessments were highly inter-correlated (r > 0.9) and impairment was shown between pre- and post-RT findings. Brain volume atrophy was shown post-RT, and cognitive impairments were correlated with radiotherapy-associated volume atrophy and dose-dependent in the left temporal lobe, corpus callosum, cerebellum and amygdala. DNN showed a good area under the curve for cognitive prediction; TICS (0.952), T-MoCA (0.909) and Tele-MACE (0.822).
Conclusions
Cognition can be evaluated remotely in which radiotherapy-related brain injury is dose-dependent and volume-dependent. Prediction models can assist in the early identification of patients at risk for neurocognitive decline following RT for glioma, thus facilitating potential treatment interventions.
Supplementary Information
The online version contains supplementary material available at 10.1007/s11060-023-04303-9.
Keywords: Cognition, Glioma, Remote assessment, Deep learning, Radiotherapy
Introduction
Neurological complications have been reported in glioma patients following radiotherapy [1]. However, radiation-induced brain injury may be the mediator for cognitive impairment in glioma patients after intensity modulated radiation therapy (IMRT) especially functions associated with structures close to the irradiated target [2]. Innovative radiation techniques such as limited fraction dose, stereotactic radiotherapy and hippocampus sparing during irradiation may result in fewer cognitive problems in patients with brain tumours [3].
Severe psychological and physiological stress is caused by long radiotherapy periods, high dosage and radiotherapy side effects [4]. Despite improved survival and toxicity outcomes in the era of IMRT, survivors display long-term quality of life (QoL) decline associated with cognition as well as depression, fatigue, sleep disturbance and pain in high grade glioma patients [5]. Even with technical sophistication and treatment advancement, long-term deficits in cognitive QoL, that is site-specific, years after radiotherapy raises concern.
Traditionally, cognitive impairment and QoL assessments require in-person evaluation. However, this method may not be feasible during pandemics. Therefore, validated telephone-based cognitive screen is required such as the Telephone Interview for Cognitive Status (TICS) [6] and Telephone Montreal Cognitive Assessment (T-MoCA) [7] and Telephone- Mini Addenbrooke’s Cognitive Examination (Tele-MACE) [8]. Simple adaptation of the cognitive assessments for telephone used is comparable in results given that both the sensitivity and specificity in the detection of mild cognitive impairment (MCI) remain reasonable when visual items are removed from the scale [7–9]. European Organization for Research and Treatment of Cancer (EORTC) QLQ-BN20 is a QoL assessment specific to brain neoplasms. The module evaluates the tumour effects and treatment symptoms, functions and health-related QoL of brain tumour patients [10]. This cancer subtype-specific QOL questionnaire is often used as a supplement to the QLQ-C30 core general questionnaire.
In observing radiation-induced anatomical differences, quantitative volumetric MRI measurements were used. Several studies focused on volumetric changes in the whole brain (WB), white matter (WM) and grey matter (GM) [11], while some studies evaluated volume changes in specific brain regions such as the hippocampus [12], amygdala [12], cerebellum [13], corpus callosum [14] and subregions of the cerebral cortex [15, 16]. From the studies, volume decrease was always shown in the brain tissue as a response to radiation which was linked to cognitive function [16].
In this study, we aimed to assess cognitive functions and QoL using remote or teleconsultation assessments in glioma patients following radiotherapy and evaluate their correlates to anatomical volume change and dosimetry parameters. Additionally, we built models to predict the development of cognitive impairment following glioma radiotherapy which may be a basis for future clinical applications.
Materials and methods
Subjects
A total of 30 healthy controls (HC) and 30 gliomas (17 glioma, 6 glioblastoma multiforme, 4 astrocytoma and 3 anaplastic astrocytoma) patients treated with radiotherapy (RT) and had both MRI scans (pre- (before the initiation of RT) and post-RT (6 months to 1 year)) were identified from the National Cancer Institute Radiotherapy and Oncology Department database with staging from T1N0M0 to T4N2M0. Patients received treatment from 2015 to 2021. Patients are planned for radiotherapy using three-dimensional conformal, IMRT or Tomotherapy with the dose prescription of 54 Gy (Gy) in 30 fractions, 59.4 Gy in 33 fractions or 60 Gy in 30 fractions. The inclusion criteria were newly diagnosed glioma patients aged between 16 and 76 years, able to understand and communicate in Malay or English. Patients were excluded include those with presence of metastasis or abnormalities in brain, medical issues, without RT or flow-up scan, claustrophobic and contraindicated in imaging due to implants or prosthetics.
Remote neurocognitive tests and quality of life (QoL) assessments
Neurocognitive tests
The TICS is a brief, standardised test of cognitive functioning with eleven individual item scores. The total score provides a measure of global cognitive functioning. A score of ≤ 31 as the optimal cut-off score to separate subjects with MCI from normal cognition [17]. While T-MoCA [18] assesses several cognitive domains; memory, attention, language, abstraction, delayed recall and orientation for the detection of MCI. The assessment contains 22-point test and a score of 18 and higher generally considered normal. The Tele-MACE evaluates three main cognitive domains (orientation, memory, and language) and it is an adaptation of MACE with the omission of visuospatial domain. The test denominators were reduced to 25 with a score of 19 and higher considered no MCI. The time frame between the imaging and neurocognitive tests varied between 2.8 (1–7) years. As late-delayed phase occurs at more than 6 months post-RT is often related to cognitive impairment, thus the condition suits the study design and aim. All the tests were given in the raw form once by a single caller to avoid variability and analyse randomly by the neuropsychologist. Analyses of the data were done blindly with assessments done prior image delineation and dose-volume extraction.
Quality of life (QoL)
The EORTC QLQ-BN20 has 20 items with four multi-item scales (future uncertainty, visual disorders, motor dysfunction and communication deficit) and seven single items (headache. seizures, drowsiness, hair loss, itchy skin, weakness of the legs and bladder control) [10]. It is used in conjunction with QLQ-C30. The validated Malay and English version of the QLQ-C30 and QLQ-BN20 were used. All the tests and assessments were responded by patients without assistance. Clear and precise instructions and explanations were given to patients before conducting questionnaire.
Data collection and region of interest dosimetry
Clinical factors were retrieved from the database, including age, sex, education and histological grade. Staging was done according to the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM). The selected region of interest (ROI) is based on its proximity to tumour location, organs at risk (OARs), institution standard protocols and previous studies [19, 20]. The gross tumor volume (GTV), clinical target volume (CTV) and planning target volume (PTV) were delineated using ESTRO-ACROP guideline and RTOG guideline by the clinician. Besides, ROI were also delineated (ESTRO-ACROP guideline) [21] to collect precise dose-related parameters, including the hippocampus, left and right temporal lobes (TLs), cerebellum, caudate nucleus (CN), corpus callosum (CC), amygdala, thalamus, optic chiasm (OC), spinal cord (SC) and brain stem (BS) by a single trained radiation therapist to avoid variability. Then, dose-volume histogram (DVH) factors were extracted, including the ROI volumes, mean/maximum/minimum dose (Dmean/ Dmax/ Dmin), D10 (dose to 10% of volume), D10-D60 in D10 increments, aV10 (absolute volume receiving more than 10 Gy (Gy)), aV10-aV60 in aV10 increments. These were done using using Monaco 5.1 and TomoHD 5.1.1.6 treatment planning systems. For dosimetric parameters, each evaluated structure was assumed to be independent. The extracted information was evaluated as predictive measures in developing the predictive model.
Image acquisition
Volumetric MRI scans are done using Siemens Magnetom Verio 3 tesla (3T). A total of 2 scans (pre and post-RT) were conducted on each patient. Patients laid supine on the MRI couch and an 8-channel RF head coil was used. The standardised protocol includes pre- and post-contrast a 3-dimensional (3D) volumetric T1-weighted multi-echo magnetisation-prepared rapid-acquisition gradient echo (MPRAGE) and 3D T2-weighted fluid-attenuation inversion recovery (FLAIR) images. MR images parameters applied in the study were TR = 1900 ms, TE = 2.52ms, T1 = 900ms, flip angle = 9°, voxel size = 1.0 × 1.0 × 1.0 mm3, with no interslice gap. During scanning, subjects were instructed to remain calm and keep their eyes closed. Scans took approximately 45 min.
Statistical analysis
Independent t-tests and Chi-square tests were used respectively to detect demographic differences between glioma patients and healthy controls at baseline. Paired t-test and Wilcoxon signed-rank tests were used to determine the volumetric changes of the evaluated regions following radiotherapy by investigating the pre- and post-MRI scans. P < 0.05 was considered statistically significant. The relationship between the cognitive assessments and QoL was shown by Pearson correlation tests. The multiple linear regression (MLR) analysis was done to determine the relationship between brain volume changes and DVH factors to cognitive results with demographic data as the contributing factors using the backward stepwise method.
MLP deep neural network
Dose-related features were determined and a multilayer perceptron (MLP) network was performed with two hidden layers using hyperbolic tangent as the activation function and softmax in the output layer for rescaling of dependent variables in predictors associated with the incidence of cognitive decline. The dataset of structures was randomly divided into a train set (70%) and a test set (30%) using hold-out cross validation by SPSS. Analysis of specificity, sensitivity, area under the curve (AUC), Variable Importance Index and the proportion of accurately classified groups for the MLP deep neural network (DNN) were generated. The statistical measure yielded by the receiver operator characteristic (ROC) curve that tests the prediction’s diagnostic efficacy was shown by AUC. Whereas, the Normalised Importance Index (NII) for each independent variable indicates the weight of each independent variable in predicting the cognitive groups (Decline and Healthy groups) of each neurocognitive test and ranges between 0.00% and 100%. Higher indices show that the variable has a stronger influence in predicting or classifying the cognitive groups.
Results
Demographic tests
Table 1 summarises and compares the demographic information of the participants. There were no significant differences between the HC and glioma patients in terms of age, gender and education.
Table 1.
Descriptive Data
| Mean ± SD | |||
|---|---|---|---|
| Variable | RT + Glioma (n = 30) | HC (n = 30) | p-value |
| Gender, M/F | 8/22 | 12/18 | 0.312~ |
| Age, years | 48.93 ± 16.57 (16–76) | 38.73 ± 6.46 (26–60) | 0.487* |
| Education, level | 10.20 ± 2.188 | 12.27 ± 1.288 | 0.788~ |
|
Histological grade Grade I/II Grade III/IV |
2.5 ± 1.14 8/6 9/7 |
||
| TICS | 29 ± 11.89 | 44.93 ± 4.35 | 0.001* |
| T-MoCA | 12.4 ± 5.28 | 18.9 ± 1.49 | 0.001* |
| Tele-MACE | 13.5 ± 4.85 | 20.77 ± 2.1 | 0.001* |
| QoL | 41.17 ± 12.87 | 26.87 ± 6.68 | 0.001* |
~Chi-square
* Independent t-test
Abbreviations: RT – radiotherapy, HC – Healthy control, TICS – Telephone Interview Cognitive Status, T-MoCA – Telephone Montreal Cognitive Assessment, Tele-MACE - Telephone Mini Addenbrooke’s Cognitive Examination, QoL – Quality of life, SD – standard deviation
Remote assessments
For all the remote tests, there were significant differences (p < 0.001) in patients to HC [Table 1]. All the neurocognitive tests by glioma patients were also significantly and positively correlated to one another but negatively correlated to QoL [Supplementary 1].
Cerebral volume
After treatment, cerebral volumes in patients decreased significantly in all the selected regions [Table 2], with significant differences between pre- and post-RT volumes. Changes were most prominent in both left (difference (post-pre), Δ = -2.96 ± 0.93) and right (Δ = -2.21 ± 1.1) TLs, given their large area and proximity to the irradiated volume.
Table 2.
Cerebral volume differences pre- and post-radiotherapy of glioma
| Pre-RT (mean ± SD) |
Post-RT (mean ± SD) |
Difference, Δ | p-value | |
|---|---|---|---|---|
| Hippocampus | 3.34 ± 1.35 | 3.2 ± 1.23 | -0.14 ± 0.12 | < 0.001~ |
| Right Temporal Lobe | 56.71 ± 20.8 | 54.5 ± 19.7 | -2.21 ± 1.1 | 0.006* |
| Left Temporal Lobe | 59.07 ± 19.96 | 56.11 ± 19.03 | -2.96 ± 0.93 | < 0.001~ |
| Cerebellum | 22.23 ± 10.4 | 21.33 ± 9.75 | -0.9 ± 0.65 | < 0.001~ |
| Corpus callosum | 3.08 ± 1.26 | 2.95 ± 1.13 | -0.13 ± 0.13 | < 0.001~ |
| Amygdala | 1.69 ± 0.56 | 1.61 ± 0.53 | -0.08 ± 0.03 | < 0.001* |
| Caudate nucleus | 2.12 ± 1.06 | 2.03 ± 0.99 | -0.09 ± 0.07 | < 0.001~ |
| Thalamus | 5.87 ± 1.5 | 5.61 ± 1.37 | -0.26 ± 0.13 | < 0.001~ |
| Spinal cord | 13.1 ± 8.22 | 12.84 ± 8.06 | -0.26 ± 0.16 | < 0.001* |
| Brain stem | 26.66 ± 5.04 | 25.86 ± 4.89 | -0.8 ± 0.15 | < 0.001* |
| Optic chiasm | 0.45 ± 0.32 | 0.43 ± 0.31 | -0.02 ± 0.01 | < 0.001~ |
*Paired t-test
~Wilcoxon-signed rank
Correlation between volume changes and neurocognitive tests
Volume changes in BS was significantly (p < 0.05) correlated to T-MoCA (r = 0.436), TICS (r = 0.403) and Tele-MACE (r = 0.49), while SC was significantly (p < 0.05) correlated to Tele-MACE (r = 0.365) and T-MoCA (r = 0.362). No significant correlation between cerebral volumes and neurocognitive assessments was shown in other ROI regions.
Dose and volume-related cognition-response in irradiated patients
Multiple linear regression was fitted to explain cognitive score based on brain region volumes post-treatment and DVH factor values. Initially, the assumptions of regression modelling including normality, lack of multicollinearity, and homoscedasticity were tested. For TICS, the overall model explains 86.3% (R2 = 0.863) variation of the cognitive score, and is significantly useful in explaining the TICS score, F (12,17) = 8.936, p < 0.001. Significant changes were observed in the post-treatment volume of the hippocampus (t(17) = 2.3, p = 0.034), the left TL (t(17) = 4.565, p < 0.001) and CC (t(17) = 2.603, p = 0.019), treatment dose in the BS Dmin (t(17) = -4.051, p = 0.001), the OC Dmin (t(17) = -4.325, p < 0.001) the hippocampus Dmax (t(17) = -2.826, p = 0.012), the right TL Dmax and Dmean (t(17) = -3.885, p = 0.001; t(17) = -2.885, p = 0.01), with increase in volume and decrease in treatment dose, increases the cognitive score. Besides, significant changes was found in education level (t(17) = 6.872, p < 0.001), with one unit increase, the TICS score increases by 12.317.
A different set of regression modelling was shown in T-MoCA with the overall model of 86.3% (R2 = 0.863, F (11,18) = 10.274, p < 0.001). Significant changes were observed in the post-treatment volume of the hippocampus (t(18) = 2.371, p = 0.029), the left TL (t(18) = 4.778, p < 0.001), the CC (t(18) = 3.449, p = 0.003) and the SC (t(18)= -2.675, p = 0.015), treatment dose in the right TL Dmean and Dmax (t(18)= -3.379, p = 0.003; t(18)=-2.867, p = 0.01) and the Dmin of the OC (t(18)= -4.189, p = 0.001), the BS (t(18)= -4.206, p = 0.001) and the SC (t(18) = 2.481, p = 0.023) and education level (t(18) = 3.553, p = 0.002).
The overall model of Tele-MACE explains 89.2% (R2 = 0.892) of the cognitive score and is statistically significant (F (12,17) = 11.731, p < 0.001). Significant changes were also shown in post-treatment volume; hippocampus (t(17) = 2.418, p = 0.027), left TL (t(17) = 5.001, p < 0.001), CC (t(17) = 3.011, p = 0.008), SC (t(17) = -2.711, p = 0.015), treatment dose Dmin; BS (t(17) =-4.211, p < 0.001) and OC (t(17) = -4.415, p < 0.001), Dmax; right TL (t(17) = -3.813, p = 0.001), Dmean; right TL (t(17) =-2.798, p = 0.012) and education level (t(17) = 4.569, p < 0.001).
Deep neural network (DNN)
The Neural Networks of Tele-MACE and T-MoCA comprises 2 factors, 18 and 20 covariates in the input layer, 9 neurons in the first hidden layer and 7 neurons in the second hidden layer, and two output levels (decline and normal cognitive scores). While in TICS, there were 23 covariates in the input layer, 10 neurons in the first hidden layer and 8 neurons in the second hidden layer. The input variables comprised of stage, education level, post-treatment volumes and DVH factors with cognitive-related dosimetric predictors that MLR selected.
Supplementary 2 (a-c) gives the weight indices of the input and output variables for the networks. Unlike the β coefficients of the regression models, the DNN weight statistics have intra-variable variation. An example, the weights of hippocampus_P (post-volume) across nine neurons in the hidden layer, notated as H(1:1–9), is 0.488, 0.191, 0.182, -0.433, -0.265, -0.546, -0.086, 0.330 and − 0.267 in T-MoCA network indicating a certain degree of nonlinearity in the data. The relatively high intra-variable variance was also noted in other variables across Tele-MACE and TICS networks, indicating high nonlinearity between variables and cognitive classification. Similarly, the weights of connection between the hidden and output layers have a relatively large range, indicating high nonlinearity. Bias statistics for hidden and output layers are also presented in Supplementary 2. The presence of bias helps the network learn the data’s underlying patterns more efficiently, and in all the networks, the bias coefficients exhibit some degree of variation. Overall, all the networks had high accuracy, evidenced by the percentage of incorrect classifications, sensitivity statistics and area under the ROC with curve area between 0.8 and 0.95 which is considered good to excellent [22] in the testing and training stages [Table 3].
Table 3.
Classification Accuracy, Specificity and Sensitivity Estimated by the DNN Model
| Percentage of Incorrect Predictions | Specificity | Sensitivity | AUC -ROC | ||
|---|---|---|---|---|---|
| T-MoCA | Testing | 15.0% | 82.4% | 85% | 0.909 |
| Training | 10.0% | 88.9% | 90% | 0.909 | |
| Tele-MACE | Testing | 18.8% | 76.9% | 81.3% | 0.822 |
| Training | 7.1% | 92.3% | 92.9% | 0.822 | |
| TICS | Testing | 14.3% | 88.9% | 83.3% | 0.952 |
| Training | 11.1% | 89% | 88.9% | 0.952 |
Table 4 presents the ten most important independent variables for each network which influenced cognitive grouping to varying degrees. Hippocampus post-volume had the highest influence, as indicated by its NII (100) in both T-MoCA and TICS network, while in Tele-MACE it’s the second highest (82.5) after right TL_D60 (100). From Table 4, education level also played a major importance in the prediction output for both T-MoCA (63.5%) and TICS (79.1%) but not in the Tele-MACE (36.8%) as emphasis was given more on grading (58.8%). Similar predictor variables were shown in all the networks, including the hippocampus post-volume, left TL post-volume, education level, right TL_D30 and right TL_Dmean, which suggest the potential and importance of those variables in predicting cognitive group of each neurocognitive test. The NII of the remainder of the input variables gradually decreases but never reaches zero suggesting each variable made a significant contribution to the test group cognitive. Overall, the DNN modelling yielded different Importance Indices with high sensitivity and specificity.
Table 4.
The DNN-Estimated Importance of Independent Variables in Classifying Cognitive Group in Tele-MACE, T-MoCA and TICS
| T-MoCA | Importance | Tele-MACE | Importance | TICS | Importance | |
|---|---|---|---|---|---|---|
| 1 | Hippocampus post-volume | 100 | Right temporal lobe_D60 | 100 | Hippocampus post-volume | 100 |
| 2 | Left temporal lobe post-volume | 79.9 | Hippocampus post-volume | 82.5 | Right temporal lobe_Dmax | 89.6 |
| 3 | Amygdala_V60 | 72.2 | Spinal cord post-volume | 67.4 | Caudate nucleus_V10 | 85.6 |
| 4 | Education level | 63.5 | Grade | 58.8 | Right temporal lobe_Dmean | 82.1 |
| 5 | Right temporal lobe_D30 | 62.4 | Optic chiasm_Dmin | 56.8 | Education level | 79.1 |
| 6 | Hippocampus_V40 | 61.0 | Caudate nuclues_V10 | 46.7 | Hippocampus_D40 | 76.7 |
| 7 | Left temporal lobe_V60 | 59.6 | Amygdala_V40 | 43.6 | Left temporal lobe post-volume | 75.5 |
| 8 | Corpus callosum post-volume | 54.5 | Right temporal lobe_Dmean | 43.3 | Amygdala_V60 | 73.9 |
| 9 | Right temporal lobe_Dmean | 53.8 | Corpus callosum_V60 | 42.1 | Brain stem_D10 | 72.2 |
| 10 | Caudate nucleus_D60 | 52.4 | Right temporal lobe_D30 | 39.1 | Cerebellum_V10 | 67.3 |
Discussion
Comparing neurocognitive data between irradiated glioma patients to HC revealed that cognitive function in irradiated glioma have a worse cognitive function than the HC as other factors such as gender, age, and education were matched-sample. After RT, a significant decrease in brain region volumes relative to pre-RT was shown in the hippocampus, right and left TLs, CN, amygdala CC, thalamus, SC, BS and OC. Significant correlations were mainly in the BS and SC as shown by the lower cognition scores. However, after RT, patients also showed cognitive alterations due to radiation at the ROI. The findings indicate that radiation could influence human cognition. This may be the underlying reason that cognitive impairments emerge and subsequently worsen in patients after RT. Besides that, a moderate negative correlation between the QoL and neurocognitive outcomes was shown in our study, suggesting that alleviating the cancer burden outweighs the effect of diminishing cognition.
Nevertheless, we could not exclude the initial presence of a tumour as a causal effect on the cognitive impairment in glioma patients as tumour growth, regardless of type, size and location may significantly alter functional connectivity [23]. According to Reijneveld et al. [24], patients with low grade-glioma (WHO grade II) had already suffered from cognitive loss compared to matched healthy controls prior to treatment. Besides, cognitive dysfunction is also correlated with the volume of brain regions. Based on Clouston et al. [25], brain volume is associated to the manifestation of cognitive deficits rather than with the number of metastases in patients with brain metastases. It is also related to the extent of the impairment [26] where cognitive deterioration has a predictive ability on patients’ survival [27]. Radiating the hippocampus increases apoptosis in the subgranular zone of the dentate gyrus, decreasing the proliferation and differentiation of neuronal stem cells [23]. These changes promote reduced hippocampal neurogenesis, which has also been linked with deficits in hippocampal-dependent spatial learning and memory [28]. Furthermore, hippocampal atrophy is correlated with the mean hippocampus dose, with dose greater than 40 Gy, it significantly decreases the volume by nearly 6% a year [29]. Doses at ≥ 7.2 Gy to bilateral hippocampi also resulted in worse cognitive function as measured by the Wechsler Memory test [30].
According to Yang et al. [19], a decrease in gross tumour volume (GTV) which, therefore, reduces the irradiated volume of the surrounding regions, would decrease the radiation dose delivered to OARs and healthy brain tissue during RT. Besides, dynamic microstructural changes in the CC, TLs and hippocampus were observed in patients following cranial radiotherapy, which is associated with brain morphological changes [31] as shown in our study with the decreased ROIs volume following RT. Furthermore, changes in the central nervous system could be observed within hours, days, or weeks of irradiation and may synergize over time to form long term structural abnormalities, resulting in permanent cognitive impairment [28].
Radiation-induced cognitive impairment is reported to occur up to 50% in brain tumour patients after 6 months post-RT [32]. This is marked by the decrease in memory function, attention and problem-solving ability [28] that is significant in patients at 2 years post whole brain irradiation with severity continued to rise over time [32]. RT-treated patients are also at risk of developing subacute toxicity (1 to 6 months after RT completion), which hampers verbal memory, attention and fine motor dexterity [33]. Additionally, radiation could induce late brain toxicity such as radiation necrosis where the incidence increases with higher radiation dose, fraction size and the administration of chemotherapy [34]. Nevertheless, cerebral injury is the most frequent neurotoxic effect in cranial irradiation compared to focal necrosis as radiation-induced cognitive decline occurs at doses lower than radionecrosis [35]. Though cognitive dysfunction affects several aspects of QoL, such as independence, overall well-being, social functioning and employment [36], it is more pronounced in younger than in older age groups [37]. This is because younger patients more often suffer from depression, hopelessness and suicidal tendencies [37], and these symptoms affect overall QoL such as insomnia, fatigue and reduced daytime activities [38]. Nevertheless, psychological symptoms such as emotional distress, depression and anxiety may also link to executive dysfunction [39]. This suggests the importance of assessment and intervention of depression symptoms in cancer patients as it could lead to a better outcome in improving patient’s QoL.
The COVID-19 pandemic (movement and in-person constraints) was the main reason the authors conducted neuropsychological assessments remotely, thus the absence of standard assessments. This is the first-ever study that uses remote neurocognitive assessments to monitor glioma patients’ cognitive functions and correlating them with morphological changes and radiation dose. The correlations obtained were similar to conventional neurocognitive assessments in post-RT patients [40]. A study done by Cook et al. [41] showed that TICS modified telephone interviews was able to independently identify amnestic MCI with an overall 86% correct classification. Besides, TICS could reflect on general cognitive ability, mainly detecting impairments, but subtle decrements in cognitive functioning may be limited [6]. High sensitivity and specificity were also achieved in both dementia and MCI diagnosis (all > 0.8) [7] with high negative predictive values for both diagnoses (> 0.85) [8], thus, suggesting the ability of remote assessments in providing desirable cognitive evaluation. Therefore, existing cognitive assessments may be adapted for telephone implementation.
In this study, two strategies were implemented in the build-up of the prediction models, MLR and MLP-DNN. Each strategy will provide 3 neurocognitive test endpoints (TICS, T-MoCA and Tele-MACE), thus, a total of 6 models were generated. From the study, the relatively high prediction power of the regression model in the present study is in agreement with some of the previous studies where the models showed an area under the curve of 0.78 [42] and R2 ≥ 0.78 [43] with education and treatment dose as the variables in model prediction. Acknowledging that both variables could have affected the analysis results, we presumed that the regression model would be less efficient when the variables were excluded. Therefore, it can be concluded that both education and treatment dose plays a significant effect on neurocognitive function impairments. Besides, DNN models also showed that education level and brain structure parameters primarily in the hippocampus, TLs, BS and SC were predictors for neurocognitive impairments. Although these variables help to predict these future outcomes, not all of these variables were independently related to the measured outcomes, such as age, grade and gender. It is possible that the intensity of the RT may not be influenced by age and gender, as shown in adult cancer survivors who received primarily chemotherapy indicating that age is not a moderating factor in cognitive performance [43]. Nevertheless, future studies with a larger sample size are required to evaluate how the association of age, grade and gender with cognitive performance may differ by treatment plan. The DNN analysis also showed that all input variables could substantially contribute to differentiating cognitive groups with a high degree of accuracy. In previous studies, the application of artificial neural networks had achieved significantly high precision over regression [44]. Besides, DNN is also beginning to be utilised more frequently in cognitive assessments in relation to radiotherapy, as it offers significant advantages over conventional approaches such as discriminant and regression analyses [45].
There are several limitations to our study which should be considered. This is a cross-sectional study without pre-treatment cognitive data. Thus, we were unable to assess changes in neurocognitive function and QoL over time. Due to the study time frame during the pandemics, thus, unwarranted assessments for the well-being of patients should be avoided. Besides, the small sample size could be a deterrent in concluding the study outcome, and establishing the DNN and the measured outcomes. Patients might have experienced other health issues between the treatment and study participation, which could impact their neurocognitive function and QoL. But given the nature of the disease, prior studies on neurocognitive function after intracranial irradiation [43] and significant statistical output, we could assume that the number of subjects is sufficient in providing satisfactory result with the given follow-up time frame. According to Olson et al. [46], deep neural nets can achieve superior classification accuracy and rapid experimentation on small, noisy data sets despite memorising the training data. Given the circumstances of the pandemic, a more flexible, streamlined approach is required, thus the application of telephone-based interviews. Additionally, the use of remote assessments could also be beneficial, as an interview can be done at the patient’s convenience without being pressured by their surroundings, thus, a better evaluation outcome. However, it might be affected by various issues, such as poor line connection and change of number.
As both MLP DNN and MLR showed promising capability in predicting cognitive dysfunction following glioma radiotherapy, coupled with the application of remote neurocognitive assessments, monitoring of patient’s cognitive status can be easily integrated into current workflow or primary care settings given its simplicity, cost-effectiveness and convenient screening tool. Thus, potential interventions can be provided to manage cognitive changes in patients following RT.
Conclusions
Our study showed that radiotherapy-related brain injury in glioma patients is dose-dependent and volume-dependent that suggest the possible of early bio-makers in cognitive decline. The application of prediction models can aid in distinguishing individual patients with the likelihood of suffering cognitive impairments following glioma RT.
Electronic supplementary material
Below is the link to the electronic supplementary material.
Author Contribution
All authors contributed to the study conception and design. Material preparation, literature search and data analysis were performed by Noor Shatirah Voon, Noorazrul Yahya and Hanani A. Manan. The first draft of the manuscript was written by Noor Shatirah Voon and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.
Funding
This work was funded by the Ministry of Higher Learning (Malaysia)- Fundamental Research Grant (FRGS/1/2021/SS03/UKM/02/1) and the National University of Malaysia under grants FF-2020-013 and GP-2021-K017963.
Data Availability
The data presented in this article are available in the main article or supplementary materials.
Declarations
Institutional review board statement
The study was conducted in accordance with the Declaration of Helsinki, and approved by the Institutional Review Board (or Ethics Committee) of Malaysia Ministry of Health (NMRR-19-2456-49608 (IIR), 14 July 2020).
Informed consent statement
Informed consent was obtained from all subjects involved in the study.
Conflict of interest
The authors declare no conflict of interest.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Herrlinger U, Schafer N, Fimmers R, Griesinger F, Rauch M, Kirchen H, et al. Early whole brain radiotherapy in primary CNS lymphoma: negative impact on quality of life in the randomised G-PCNSL-SG1 trial. J Cancer Res Clin Oncol. 2017;143:1815–1821. doi: 10.1007/s00432-017-2423-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Turnquist C, Harris BT, Harris CC. Radiation-induced brain injury: a current concepts and therapeutic strategies targeting neuroinflammation. Neuro-Oncology Adv. 2020;2(1):1–10. doi: 10.1093/noajnl/vdaa057. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Okoukoni C, McTyre ER, Peacock DNA, Peiffer AM, Strowd R, Cramer C, et al. Hippocampal dose volume histogram predicts Hopkins Verbal Learning Test scores after brain irradiation. Adv Radiat Oncol. 2017;2:624–629. doi: 10.1016/j.adro.2017.08.013. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Sutherland A, Bennett NC, Herst P. Psychological stress affects the severity of radiation-induced acute skin reactions in breast cancer patients. Eur J Cancer Care. 2017;26(1):e12737. doi: 10.1111/ecc.12737. [DOI] [PubMed] [Google Scholar]
- 5.Fox SW, Lyon D, Farace E. Symptom clusters in patients with high-grade glioma. J Nurs Sch. 2007;39:61–67. doi: 10.1111/j.1547-5069.2007.00144.x. [DOI] [PubMed] [Google Scholar]
- 6.Castanho TC, Amorin L, Zihl J, Palha JA, Sousa N, Santos NC. Telephone-based screening tools for mild cognitive impairment and dementia in aging studies: a review of validated instruments. Front Aging Neurosci. 2014 doi: 10.3389/fnagi.2014.00016. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Wittich W, Phillips N, Nasreddine ZS, Chertkow H. Sensitivity and specificity of the Montreal Cognitive Assessment modified for individuals who are visually impaired. J Visual Impairment Blindness. 2010;105(6):360–368. doi: 10.1177/0145482X1010400606. [DOI] [Google Scholar]
- 8.Larner AJ. Cognitive testing in the COVID-19 era: can existing screeners be adapted for telephone use? Neurodegener Dis Manag. 2021;11(1):77–82. doi: 10.2217/nmt-2020-0040. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Zietemann V, Kopczak A, Muller C, Wollenweber FA, Dichgans M. Validation of the telephone interview of Cognitive Status and Telephone Montreal Cognitive Assessment against detailed cognitive testing and clinical diagnosis of mild cognitive impairment after stroke. Stroke. 2017;48(11):2952–2957. doi: 10.1161/STROKEAHA.117.017519. [DOI] [PubMed] [Google Scholar]
- 10.Taphoorn MJ, Claassens L, Aaronson NK, Coens C, Mauer M, Osoba D, et al. An international validation study of the EORTC brain cancer module (EORTC QLQ-BN20) for assessing health-related quality of life and symptoms in brain cancer patients. Eur J Cancer. 2010;46:1033–1040. doi: 10.1016/j.ejca.2010.01.012. [DOI] [PubMed] [Google Scholar]
- 11.Gommlich A, Raschke F, Wahl H, Troost EGC. Retrospective assessment of MRI-based volumetric changes of normal tissues in glioma patients following radio(chemo)therapy. Clin Trasnsl Radiat Oncol. 2018;8:17–21. doi: 10.1016/j.ctro.2017.11.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Nagtegaal SHJ, David S, Philippens MEP, Snijers TJ, Leemans A, Verhoeff JJC. Dose-dependent volume loss in subcortical deep grey matter structures after cranial radiotherapy. Clin Transl Radiat Oncol. 2021;26:35–41. doi: 10.1016/j.ctro.2020.11.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Raschke F, Seidlitz A, Wesemann T, Lock S, Jentsch C, Platzek I, et al. Dose-dependent cerebellar atrophy in glioma patients after radio(chemo)therapy. Radiother Oncol. 2020;150:262–267. doi: 10.1016/j.radonc.2020.07.044. [DOI] [PubMed] [Google Scholar]
- 14.Palmer SL, Reddick WE, Glass JO, Gajjar A, Goloubeva O, Mulhern RK. Decline in corpus callosum volume among pediatric patients with medulloblastoma: longitudinal MR imaging study. Ajnr Am J Neuroradiol. 2002;23:1088–1094. [PMC free article] [PubMed] [Google Scholar]
- 15.Seibert TM, Karunamuni R, Kaifi S, Burkeen J, Connor M, Krishnan AP, et al. Cerebral cortex regions selectively vulnerable to radiation dose-dependent atrophy. Int J Radiat Oncol Biol Phys. 2017;97:910–918. doi: 10.1016/j.ijrobp.2017.01.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Cabeza R, Nyberg L. Imaging cognition II: an empirical review of 275 PET and fMRI studies. J Cogn Neurosci. 2000;12:1–47. doi: 10.1162/08989290051137585. [DOI] [PubMed] [Google Scholar]
- 17.Knopman DS, Roberts RO, Geda YE, Pankratz VS, Christanson TJH, Petersen RC, et al. Validation of the telephone interview for cognitive status-modified in subjects with normal cognition, mild cognitive impairment, or dementia. Neuroepidemiology. 2010;34:34–42. doi: 10.1159/000255464. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Nasreddine ZS, Phillips NA, Bédirian V, Charbonneau S, Whitehead V, Collin I, et al. The Montreal Cognitive Assessment, MoCA: a brief screening tool for mild cognitive impairment. J Am Geriatr Soc. 2005;53(4):695–699. doi: 10.1111/j.1532-5415.2005.53221.x. [DOI] [PubMed] [Google Scholar]
- 19.Yang Z, Zhang Z, Wang X, Hu YM, Zp L. Intensity-modulated radiotherapy for gliomas: dosimetric effects of changes in gross tumor volume on organs at risk and healthy brain tissue. Onco Targets and Therapy. 2016;9:3545–3554. doi: 10.2147/OTT.S100455. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Pagett CJH, Lilley J, Lindsay R, Short S, Murray L. Optimising tumour coverage and organ at risk sparing for hypofractionated re-irridation in glioblastoma. Phys Imaging Radiation Oncol. 2022;21:84–89. doi: 10.1016/j.phro.2022.02.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Niyazi M, Brada M, Chalmers AJ, Combs SE, Erridge SC, Fiorentino A, et al. ESTRO-ACROP guideline “target delineation of glioblastomas. Radiother Oncol. 2016;118(1):35–42. doi: 10.1016/j.radonc.2015.12.003. [DOI] [PubMed] [Google Scholar]
- 22.Nahm FS. Receiver operating characteristic curve: overview and practical use for clinicians. Korean J Anesthesiol. 2022;75(1):25–36. doi: 10.4097/kja.21209. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Manan HA, Franz EA, Yahya N. Functional connectivity changes in patientswith brain tumours - a systematic review on resting state-fMRI. Neurol Psychiatry Brain Res. 2020;36:73–82. doi: 10.1016/j.npbr.2020.03.003. [DOI] [Google Scholar]
- 24.Reijneveld JC, Sitskoorn MM, Klein M, Nuyen J, Taphoorn MJ. Cognitive status and quality of life in patients with suspected versus proven low-grade gliomas. Neurology. 2001;56(5):618–623. doi: 10.1212/WNL.56.5.618. [DOI] [PubMed] [Google Scholar]
- 25.Clouston PD, DeAngelis LM, Posner JB. The spectrum of neurological disease in patients with systemic cancer. Ann Neurol. 1992;31(3):268–273. doi: 10.1002/ana.410310307. [DOI] [PubMed] [Google Scholar]
- 26.Schagen SB, Klein M, Reijneveld JC, Brain E, Deprez S, Joly F, et al. Monitoring and optimizing cognitive function in cancer patients: present knowledge and future directions. EJC Supplements. 2014;12:29–40. doi: 10.1016/j.ejcsup.2014.03.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Armstrong TS, Wefel JS, Wang M, Gilbert MR, Won M, Bottomley A, et al. Net clinical benefit analysis of radiation therapy oncology group 0525: a phase III trial comparing conventional adjuvant temozolomide with dose-intensive temozolomide in patients with newly diagnosed glioblastoma. J Clin Oncol. 2013;31(32):4076–4084. doi: 10.1200/JCO.2013.49.6067. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Greene-Schloesser D, Robbins ME, Peiffer AM, Shaw EG, Wheeler KT, Chan MD (2012) Radition-induced brain injury: a review. Front. Oncol. ; 19 (2):73. doi: 10.3389/fonc.2012.00073. eCollection 2012 [DOI] [PMC free article] [PubMed]
- 29.Seibert TM, Karunamuni R, Bartsch H, Kaifi S, Krishnan AP, Dalia Y, et al. Radiation dose-dependent hippocampal atrophy detected with longitudinal volumetric magnetic resonance imaging. Int J Radiat Oncol Biol Phys. 2017;97(2):263–269. doi: 10.1016/j.ijrobp.2016.10.035. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Gondi V, Hermann BP, Mehta MP, Tome WA. Hippocampal dosimetry predicts neurocognitive function impairment after fractionated stereotactic radiotherapy for Benign or low-grade adult brain tumors. Int J Radiat Oncol Biol Phys. 2011;83:e487–e493. doi: 10.1016/j.ijrobp.2011.10.021. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Voon NS, Lau FN, Zakaria R, Md Rani SA, Ismail F, Manan HA. MRI-based brain structural changes following radiotherapy of nasopharyngeal carcinoma: a systematic review. Cancer Radiother. 2021;25(1):62–71. doi: 10.1016/j.canrad.2020.07.008. [DOI] [PubMed] [Google Scholar]
- 32.Cramer CK, Cummings TL, Andrews RN, Strowd R, Rapp SR, Shaw EG, et al. Treatment of radiation-induced cognitive decline in adult brain tumor patients. Curr Treat Options Oncol. 2019;20(5):42. doi: 10.1007/s11864-019-0641-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Bradshaw M, Wefel J (2014) The neuropsychology of oncology. In: Parsons M, Hammeke T (eds). Clinical Neuropsychology: A pocket handbook for assessment, Third edition. Washington DC: American Psychological Association; :313–337
- 34.Chen J, Dassarath M, Yin Z, Liu H, Yang K, Wu G. Radiation induced temporal lobe necrosis in patients with nasopharyngeal carcinoma: a view of new avenues in its management. Radiat Oncol. 2011;6:128. doi: 10.1186/1748-717X-6-128. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Yahya N, Manan HA. Diffusion tensor imaging indices to predict cognitive changes following adult radiotherapy. Eur J Cancer Care. 2021;30(1):e13329. doi: 10.1111/ecc.13329. [DOI] [PubMed] [Google Scholar]
- 36.Reid-Arndt SA, Yee A, Perry MC, Hsieh C. Cognitive and psychological factors associated with early post-treatment functional outcomes in breast cancer survivors. J Psychosoc Oncol. 2009;27(4):415–434. doi: 10.1080/07347330903183117. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Peters E, Schulz LM, Reuss-Borst M. Quality of life cancer- how the extent of impairment is influenced by patient characteristics. BMC Cancer. 2016;16:787. doi: 10.1186/s12885-016-2822-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Yeo S, Lee J, Kim K, Kim HJ, Cheong S. Depression rather than cancer-related fatigue or insomnia, decreased the quality of life of cancer patients. Cancer Res Treat. 2021;53(3):641–649. doi: 10.4143/crt.2020.1212. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Snyder HR. Major depressive disorder is associated with broad impairments on neuropsychological measures of executive function: a meta-analysis and review. Psycholl Bull. 2013;139(1):81–132. doi: 10.1037/a0028727. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Voon NS, Abdul Manan H, Yahya N. Cognitive decline following Radiotherapy of Head and Neck Cancer: systematic review and Meta-analysis of MRI correlates. Cancers (Basel) 2021;13(24):6191. doi: 10.3390/cancers13246191. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Cook SE, Marsiske M, McCoy KJM. The use of the modified telephone interview for cognitive status (TICS-M) in the detection of amnestic mild cognitive impairment. J Ceriatr Psychiatry Neurol. 2009;22(2):103–109. doi: 10.1177/0891988708328214. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Zamanipoor Najafabadi AH, van der Meer PB, Boele FW, Taphoorn MJB, Klein M, Peerdeman SM, et al. Determinants and predictors for the long-term disease burden of intracranial meningioma patients. J Neurooncol. 2021;151:201–210. doi: 10.1007/s11060-020-03650-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Gui C, Vannorsdall TD, Kleinberg LR, Assadi R, Moore JA, Hu C, et al. A prospective cohort study of neural progenitor cell-sparing Radiation Therapy Plus Temozolomide for newly diagnosed patients with Glioblastoma. Neurosurgery. 2020;87:E31–E40. doi: 10.1093/neuros/nyaa107. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Franco GD, Santurro M. Machine learning, artificial neural networks and social research. Qual Quant. 2021;55:1007–1025. doi: 10.1007/s11135-020-01037-y. [DOI] [Google Scholar]
- 45.Akbilgic O, Davis RL. The promise of machine learning: when will it be delivered. J Card Fail. 2019;25(6):484–485. doi: 10.1016/j.cardfail.2019.04.006. [DOI] [PubMed] [Google Scholar]
- 46.Olson M, Wyner A, Berk R (2018) Modern neural networks generalize on small data sets.NeurIPS.
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data presented in this article are available in the main article or supplementary materials.