PTSD Classification Based on Surface-Based Morphometry: Integrating SHAP Analysis for Key Feature Identification

PTSD Classification Based on Surface-Based Morphometry: Integrating SHAP Analysis for Key Feature Identification | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article PTSD Classification Based on Surface-Based Morphometry: Integrating SHAP Analysis for Key Feature Identification Yulong Jia, Beining Yang, Haotian Xin, Qunya Qi, Yu Wang, Chaoyang Huang, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5777371/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Post-Traumatic Stress Disorder (PTSD) is associated with neurobiological alterations, which can be examined using surface-based morphology (SBM). While machine learning (ML) approaches have shown potential in classifying PTSD based on SBM features, further exploration is needed to improve interpretability and clinical relevance. Objectives This study seeks to integrate ML-based classification of PTSD with SHAP analysis to identify important SBM features and their potential associations with PTSD symptomatology, providing insights into the structural changes underlying PTSD. Methods High-resolution T1-weighted MRI data from 101 participants (62 PTSD, 39 healthy controls) were analyzed using FreeSurfer’s SBM pipeline, extracting cortical thickness, surface area, and curvature features from the aparc.a2009s atlas. Several ML models, including Random Forest, SVM, and XGBoost, were trained and evaluated using ten-fold cross-validation. SHAP analysis was applied to determine feature importance, and correlation analyses were conducted to examine relationships between key features and PTSD symptom severity. Results Sixteen cortical regions were identified with significant structural differences in PTSD, including reduced cortical thickness in the left lingual gyrus and increased thickness in the bilateral central sulcus. The Random Forest model achieved the highest accuracy (91%) in PTSD classification. SHAP analysis highlighted the left lingual gyrus and parahippocampal gyrus as key features. Correlation analysis suggested potential links between these features and specific PTSD symptom clusters. Conclusion The integration of SBM and interpretable ML methods provides a promising approach for investigating structural brain changes in PTSD. While further validation is needed, these findings contribute to a better understanding of PTSD neurobiology and may support future research on diagnostic and therapeutic strategies. PTSD Neuroimaging SBM machine learning SHAP analysis Figures Figure 1 Figure 2 Introduction PTSD is a complex psychiatric disorder that emerges following traumatic events, often resulting in chronic symptoms such as intrusive recollections, hyperarousal, emotional detachment, and avoidance behaviors, all of which significantly impair the quality of life[ 1 ]. In recent years, neuroimaging has provided critical insights into the neurobiological underpinnings of PTSD, revealing that trauma impacts specific brain regions involved in emotion regulation, memory consolidation, and cognitive processing[ 2 ]. Among the various neuroimaging techniques, SBM has become increasingly valuable for understanding PTSD-related structural changes[ 3 ]. Unlike traditional volumetric analysis, SBM provides fine-grained assessments of the cortical surface, including measures such as cortical thickness, surface area, and curvature[ 4 ]. These features are particularly informative for identifying subtle cortical alterations that may be linked to PTSD symptomatology, offering a more nuanced view of how PTSD manifests at the neuroanatomical level[ 5 ]. Notably, SBM studies in PTSD have revealed significant structural changes in cortical regions that contribute to PTSD's characteristic symptoms. For instance, reductions in cortical thickness in the prefrontal cortex, a region critical for cognitive control and emotional regulation, are frequently associated with diminished ability to manage intrusive thoughts and emotional responses in PTSD patients[ 6 ]. Additionally, structural irregularities in the hippocampus, such as reduced surface area, have been linked to impairments in memory and stress response, while changes in occipital regions involved in visual processing may influence trauma-related hyperarousal and threat perception[ 7 ]. Despite these findings, a gap remains in directly correlating SBM-derived features with specific PTSD symptoms, as traditional group comparisons often overlook how individual cortical features relate to symptom severity or symptom clusters[ 8 , 9 ]. Addressing this gap could enhance our understanding of the neural basis of specific PTSD symptoms and improve the targeting of therapeutic interventions. Advances in ML further enable more sophisticated analysis of SBM features in PTSD, particularly for classification purposes[ 10 ]. By analyzing high-dimensional data, ML techniques can identify patterns within complex SBM datasets that traditional approaches might miss[ 11 ], facilitating the differentiation of PTSD from healthy controls or even other psychiatric conditions[ 12 ]. However, while ML models can achieve high classification accuracy, their lack of interpretability has limited their clinical utility[ 13 ]. In clinical contexts, understanding how and why certain cortical features contribute to PTSD classification is essential for establishing confidence in the model's predictions and for advancing patient-centered applications[ 14 ]. To address this limitation, recent ML developments have introduced interpretability techniques like SHAP (SHapley Additive exPlanations)[ 15 ], which can clarify the contribution of each SBM feature to the model’s predictions. SHAP values are particularly useful in neuroimaging research because they provide a transparent way to identify and rank the most important SBM features for PTSD classification, making it possible to link these features directly with PTSD symptoms or symptom clusters. In this study, we aim to bridge the gap between neuroimaging analysis and clinical interpretability by building an ML model that uses SBM features for PTSD classification and then employing SHAP analysis to interpret the model’s decision-making process. Specifically, our objectives are to construct a robust ML model that utilizes SBM-derived measures, such as cortical thickness and curvature, and to apply SHAP analysis to determine which of these features are most associated with PTSD, thereby linking structural brain changes to specific symptomatology. By doing so, we hope to enhance the clinical applicability of SBM research in PTSD by making the model’s features and their contributions to PTSD classification both transparent and interpretable. This approach has the potential to improve diagnostic precision in PTSD while also providing clinicians with a clearer understanding of the structural brain changes associated with specific PTSD symptoms, advancing both the research and clinical landscapes of PTSD neuroimaging. Materials and Methods Participant Selection This study utilized data from the Brain Aging in ADNI-DoD database and the ADNI database (adni.loni.usc.edu), focusing on individuals aged 60 to 80 years. Given the purpose of this study to investigate PTSD-related structural differences, we carefully selected participants based on predefined inclusion and exclusion criteria: (1) PTSD Group (n = 62) Participants were selected from the ADNI-DoD database, which was designed to study the effects of PTSD and traumatic brain injury (TBI) in an aging veteran population. PTSD status was determined using CAPS-IV total severity scores, with a threshold score of CAPS-IV ≥ 45 used to identify clinically significant PTSD symptoms, requiring at least 1 B symptom (re-experiencing), 3 C symptoms (avoidance), 2 D symptoms (hyperarousal), symptoms persisting over 1 month, and significant social or occupational impairment, as confirmed by clinical psychiatrists in the DOD-ADNI dataset. Only individuals without a history of neurodegenerative disease (e.g., Alzheimer’s disease, Parkinson’s disease) were included. Individuals with severe cognitive impairment (MoCA < 26, MMSE < 24) were excluded to minimize confounding effects of cognitive decline. To reduce potential confounding effects of TBI, individuals with a history of moderate-to-severe TBI were excluded. Mild TBI cases were reviewed, and sensitivity analyses were conducted to assess potential impacts. (2) Healthy Control Group (n = 39) Participants were selected from the ADNI database to match the PTSD group in terms of age and education level. Inclusion required no history of PTSD or significant traumatic exposure, confirmed through self-report and clinical evaluation. Participants with clinically significant cognitive impairment (MoCA < 26, MMSE < 24) were excluded. Individuals with major psychiatric disorders (e.g., major depressive disorder, schizophrenia) or neurological conditions were excluded. From the extensive ADNI and ADNI-DoD databases, finally we identified 101 subjects who met the predefined inclusion and exclusion criteria. This selection was based on a comprehensive search for individuals with high-quality T1-weighted MRI scans and complete clinical data that aligned with our study requirements. MRI data acquisition and Surface-based morphometry MRI data used in this study were derived from the existing Alzheimer's Disease Neuroimaging Initiative–Department of Defense (ADNI-DoD) dataset; thus, no new MRI data were collected. The ADNI-DoD database was properly cited, and raw imaging data from this dataset consisted of high-resolution T1-weighted MRI scans. MRI scans from the ADNI-DoD database were processed utilizing surface-based morphometry (SBM) analysis through FreeSurfer software[ 16 , 17 ]. Specifically, original T1-weighted images in DICOM format were first converted into Neuroimaging Informatics Technology Initiative (NIfTI) format, facilitating subsequent data preprocessing and analysis steps. Preprocessing of the structural MRI data included cortical reconstruction and segmentation using FreeSurfer’s standardized processing pipeline[ 18 ]. This process was automated with a batch script, ensuring consistent data handling across participants[ 19 ]. Cortical reconstruction and volumetric segmentation were performed automatically, producing cortical thickness, surface area, curvature, and volume data aligned to the standard fsaverage template. Each derived cortical metric was spatially smoothed using a Gaussian kernel with a full-width at half maximum (FWHM) of 10 mm to enhance comparability across participants. The smoothed data were saved for further analysis. Region of interest (ROI)-level metrics were extracted according to standardized anatomical terminology defined by the Destrieux Atlas[ 20 ]. Statistical analyses were conducted using general linear modeling(GLM) to examine structural differences between PTSD patients and healthy controls, accounting for covariates including age and gender. Corrections for multiple comparisons were rigorously performed using false discovery rate (FDR) and Monte Carlo simulation methods[ 21 ]. After statistical analysis, significant anatomical clusters were identified and summarized for further interpretation and reporting. Classification model selection To identify the optimal classification model, we utilized the scikit-learn library to review and test machine learning models based on significant differences in SBM features for MRI-based PTSD classification. The candidate models included Support Vector Machine (SVM) with linear and RBF kernels, Random Forest (RF), Logistic Regression (LR), Decision Tree, and Extreme Gradient Boosting (XGBoost)[ 22 ]. We split the dataset into training and test sets at a 9:1 ratio. Hyperparameter tuning for each classifier was conducted using embedded 10-fold cross-validation on the training set. To ensure the statistical validity of the classification performance, we ran 1,000 permutation tests, enabling us to evaluate the robustness of the model by comparing true results to performance under randomized label assignments. Finally, we assessed and compared the classifiers’ performance on the test dataset. To interpret the contribution of each feature in the classification of PTSD, we employed SHAP values. SHAP values provide a consistent measure of feature importance by quantifying the contribution of each feature to the model’s output on an individual basis[ 23 ]. We computed SHAP values for the best-performing model using the SHAP Python library, which integrates with the scikit-learn framework to explain model predictions. The SHAP analysis involved calculating the average impact of each feature across all predictions, highlighting the most influential brain regions for PTSD classification. By visualizing SHAP values, we identified features that significantly contributed to model decisions, providing insights into which cortical characteristics were most predictive of PTSD. This feature importance analysis was integral in understanding the model’s decision-making process and validating the clinical relevance of identified biomarkers. Correlation Analysis Between SHAP-Ranked Brain Regions and PTSD Symptoms To further explore the relationship between model-identified brain regions and PTSD symptom severity, we conducted a Spearman correlation analysis. Specifically, we compared all differential brain regions with CAPS total scores as well as B, C, and D symptom cluster scores from the CAPS assessment, using Spearman's rank correlation to capture potential monotonic relationships. Results Demographic and clinical characteristics A total of 62 participants with PTSD and 39 healthy controls (HC) were included in this study. No significant differences were found between the PTSD and HC groups in terms of age (PTSD: 70.2 ± 5.8 years; HC: 69.8 ± 5.6 years; p > 0.05). Additionally, no significant differences were identified between the PTSD and HC groups regarding cognitive assessments, including the MoCA scores (PTSD: 27.5 ± 1.5; HC: 27.6 ± 1.2; p = 0.261) and MMSE scores (PTSD: 28.0 ± 1.8; HC: 28.1 ± 1.5; p = 0.233). Similarly, no significant differences were observed in depressive symptoms as assessed by the Geriatric Depression Scale (GDS; PTSD: 1.85 ± 1.85; HC: 1.85 ± 1.80; p = 0.315). None of the participants reported significant cognitive impairment or depressive symptoms. Notably, there was no significant difference in the gender distribution or history of traumatic brain injury between groups ( p > 0.05)(Table 1 ). Abnormal brain structural changes in PTSD patients Significant structural differences were observed between the PTSD group and the control group across 16 brain regions. Decreased cortical thickness was found in the left Lingual Gyrus, Inferior Temporal Gyrus, Parahippocampal Gyrus, and the right Lingual Gyrus. Additionally, reductions in surface area were identified in the left Lingual Gyrus, as well as the right Calcarine Sulcus and Occipital Pole, while curvature decreases were detected in the left Anterior Occipital Sulcus. Volume reductions were prominent in the left and right Lingual Gyrus. In contrast, increased cortical thickness was observed in the bilateral Central Sulcus, left Intraparietal and Transverse Parietal Sulcus, and right Inferior Precentral Sulcus, along with volume increases in the bilateral Central Sulcus. These findings were statistically corrected at the cluster level using FDR correction ( p < 0.01).(Table 2 ). Table 1 Demographic and Clinical Characteristics of PTSD and HC Variables PTSD (n = 62) HC (n = 39) p -value Age (years, mean ± SD) 70.2 ± 5.8 69.8 ± 5.6 0.687 Education (years, mean ± SD) 15.4 ± 2.1 15.8 ± 2.0 0.756 MoCA Score (mean ± SD) 27.5 ± 1.5 27.6 ± 1.2 0.261 MMSE Score (mean ± SD) 28.0 ± 1.8 28.1 ± 1.5 0.233 GDS Score (mean ± SD) 1.85 ± 0.9 1.80 ± 0.7 0.315 TBI History (n, %) 0 (0%) 0 (0%) N/A CAPS-IV Total Score (mean ± SD) 58.7 ± 8.3 N/A N/A Table 2 Significant differential brain regions in surface-based morphometry Brain Region Coordinates Property T value P value Left Lingual Gyrus (-5.6, -68.1, 4.0) Area -6.183 0.001 Left Anterior Occipital Sulcus (-41.3, -64.7, 1.8) Curv -5.767 0.039 Left Lingual Gyrus (-12.8, -81.1, -6.9) Thickness -2.882 0.001 Left Inferior Temporal Gyrus (-54.7, -48.4, -9.7) Thickness -4.786 0.003 Left Parahippocampal Gyrus (-33.8, -11.0, -27.4) Thickness -3.939 0.009 Left Central Sulcus (-11.9, -31.5, 58.2) Thickness 4.419 0.011 Left Central Sulcus (-44.0, -12.5, 27.0) Thickness 4.138 0.017 Left Intraparietal and Transverse Parietal Sulcus (-22.7, -58.5, 33.1) Thickness 3.502 0.031 Left Lingual Gyrus (-13.4, -63.9, 1.1) Volume -5.811 0.001 Left Central Sulcus (-41.3, -20.1, 35.8) Volume 3.768 0.005 Right Inferior Precentral Sulcus (38.2, 9.1, 41.5) Thickness 3.573 0.001 Right Lingual Gyrus (12.4, -78.9, -4.9) Thickness -3.394 0.004 Right Central Sulcus (25.5, -26.6, 52.6) Thickness 4.015 0.028 Right Calcarine Sulcus (21.0, -56.4, 2.8) Area -3.272 0.001 Right Occipital Pole (15.0, -93.1, -4.1) Area -3.085 0.002 Right Lingual Gyrus (6.7, -66.7,5.5) Volume -4.625 0.001 Model Performance Comparison Six machine learning models (Random Forest, SVM (Linear), SVM (RBF), XGBoost, Logistic Regression, and Decision Tree) were evaluated using ten-fold cross-validation and tested on an independent dataset. The performance metrics, including AUC, accuracy, sensitivity, and specificity, for each model are summarized as follows: The Random Forest model achieved an average AUC of 0.90, accuracy of 0.79, sensitivity of 0.88, and specificity of 0.66 during cross-validation. On the test set, it maintained a high accuracy of 0.91 and an AUC of 0.75, with 100% sensitivity and 75% specificity.The SVM (Linear) model showed a slightly higher average AUC of 0.93, accuracy of 0.79, sensitivity of 0.91, and specificity of 0.59 in cross-validation. On the test set, it delivered an accuracy of 0.91, sensitivity of 1.00, an AUC of 0.79, and specificity of 0.75.The SVM (RBF) model demonstrated a high sensitivity of 0.98 in cross-validation, but its specificity was low (0.12), resulting in an average AUC of 0.92. On the test set, despite achieving perfect sensitivity (1.00), the model performed poorly in terms of specificity (0.00), with an accuracy of 0.64 and an AUC of 0.79. The XGBoost model performed with an average AUC of 0.86, accuracy of 0.74, sensitivity of 0.82, and specificity of 0.64 during cross-validation. On the test set, it showed an accuracy of 0.82, sensitivity of 1.00, and an AUC of 0.64, with a lower specificity of 0.50. The Logistic Regression model exhibited strong performance with an average AUC of 0.93, accuracy of 0.70, sensitivity of 0.95, and specificity of 0.33 during cross-validation. On the test set, it achieved an AUC of 0.86, accuracy of 0.91, sensitivity of 1.00, and specificity of 0.75. The Decision Tree model, despite its lower average AUC of 0.67, accuracy of 0.68, sensitivity of 0.71, and specificity of 0.62 in cross-validation, showed a better performance on the test set with an accuracy of 0.82, sensitivity of 1.00, AUC of 0.75, and specificity of 0.50(Table 2 , Fig. 1 ). Table 2 Comparison of Classification Performance Among Different Machine Learning Models for PTSD Diagnosis Model AUC (Training Set) Accuracy (Training Set) Sensitivity (Training Set) Specificity (Training Set) AUC (Test Set) Accuracy (Test Set) Sensitivity (Test Set) Specificity (Test Set) Random Forest 0.90 0.79 0.88 0.66 0.75 0.91 1.00 0.75 SVM (Linear) 0.93 0.79 0.91 0.59 0.79 0.91 1.00 0.75 SVM (RBF) 0.92 0.83 0.98 0.12 0.79 0.64 1.00 0.00 XGBoost 0.86 0.74 0.82 0.64 0.64 0.82 1.00 0.50 Logistic Regression 0.93 0.70 0.95 0.33 0.86 0.91 1.00 0.75 Decision Tree 0.67 0.68 0.71 0.62 0.75 0.82 1.00 0.50 SHAP Analysis of Feature Importance The SHAP value analysis revealed several key brain regions that contribute to the classification of PTSD versus the control group. The intraparietal sulcus and posterior transverse segment and the medial temporal hippocampal gyrus showed the highest positive contributions to PTSD classification, highlighting their significant role in the model. Additionally, regions associated with the central sulcus had a significant impact on the classification results in both positive and negative directions. Among the volume and thickness-related features, the central region volume, lingual gyrus volume, and inferior temporal gyrus thickness were also identified as important contributors. Other significant brain regions included the lingual gyrus, occipital pole, and anterior occipital cortex, whose structural changes play a crucial role in classification(Fig. 2 ). Association between differential SBM and clinical measures Correlation analysis revealed a significant association between the curvature of the anterior occipital sulcus and the C symptom cluster of CAPS. A Spearman correlation analysis indicated a p-value of 0.04, suggesting a moderate relationship between structural variations in this brain region and the severity of C symptoms in PTSD. Discussion In this study, we sought to advance the understanding of PTSD by integrating SBM with ML techniques to classify PTSD and identify critical brain features, enhancing interpretability through SHAP analysis. Our findings not only corroborate existing evidence of PTSD-related structural brain changes but also introduce a novel, interpretable framework for leveraging SBM-derived features in PTSD classification. By pinpointing key regions such as the occipital-temporal areas and central sulcus, and linking these to specific symptom clusters, our approach offers new insights into the neurobiological underpinnings of PTSD and its potential diagnostic and therapeutic applications. Our results align with a substantial body of PTSD neuroimaging research that has consistently identified structural alterations in regions involved in emotion regulation, memory, and sensory processing[ 6 , 24 , 25 ], Specifically, we observed reduced cortical thickness, surface area, and volume in the left and right occipital-temporal regions, including the lingual gyrus and parahippocampal gyrus, alongside curvature changes in occipital areas. These findings resonate with prior studies reporting cortical thinning in the occipital-temporal cortex and hippocampus, which are implicated in emotional dysregulation, memory impairments, and heightened threat perception—hallmark symptoms of PTSD[ 3 , 8 , 9 ]. For instance, the lingual gyrus, a region critical for visual processing, has been associated with trauma-related hyperarousal and intrusive visual recollections[ 7 ], while the parahippocampal gyrus’ role in contextual memory processing may underlie difficulties in distinguishing safe from threatening environments in PTSD patients[ 9 ]. The observed volume reductions in these areas further support the hypothesis of trauma-induced structural atrophy, potentially reflecting chronic stress effects on neuroplasticity and neuronal integrity[ 25 ]. These changes may disrupt the brain’s ability to modulate sensory inputs and emotional responses, contributing to the persistent hypervigilance and avoidance behaviors seen in PTSD. Our correlation analysis revealed a significant association between the curvature of the anterior occipital sulcus and the C symptom cluster (avoidance) of CAPS. While direct literature on the anterior occipital sulcus in PTSD is limited, this finding partially aligns with studies like Crombie et al.[ 26 ], which linked occipital region structural changes to specific PTSD symptom clusters, including avoidance. The anterior occipital sulcus, located near visual processing areas, may influence avoidance behaviors by altering sensory integration of environmental cues. A potential reason for this correlation could be that changes in curvature reflect disrupted visual-spatial processing, leading to heightened avoidance of trauma-related stimuli as patients struggle to contextualize or filter sensory input. This suggests a nuanced role for occipital regions in PTSD symptomatology, though further research is needed to validate this link and explore underlying mechanisms. Conversely, we found increased cortical thickness and volume in the bilateral central sulcus and precentral gyrus, regions associated with motor function and sensory integration. This observation is less commonly reported in PTSD literature but may suggest compensatory plasticity, where the brain adapts to trauma-related deficits in other areas. Similar compensatory mechanisms have been noted in other neuropsychiatric conditions, such as ADHD and stroke recovery, where increased thickness in specific regions is thought to offset functional impairments[ 27 , 28 ].In the context of PTSD, this could reflect an adaptive response to heightened motor reactivity (e.g., startle responses) or efforts to maintain cognitive control amidst emotional dysregulation. However, this finding warrants cautious interpretation, as it contrasts with the predominant narrative of cortical thinning in PTSD. It may indicate heterogeneity in PTSD neurobiology, potentially linked to differences in trauma type, duration, or individual resilience factors, as suggested by recent studies[ 7 ]. Future longitudinal research could clarify whether these increases represent a protective mechanism or a maladaptive response to chronic stress. The integration of ML in this context allowed us to classify PTSD patients based on their SBM-derived features with considerable accuracy. Among the various classifiers tested, the RF model demonstrated the highest performance, achieving an average AUC of 0.90 and an impressive sensitivity of 0.88 during cross-validation. The model's ability to generalize, as shown by its high test set accuracy of 0.91, highlights the power of using ML for PTSD classification. This supports the idea that SBM features, when paired with advanced ML techniques, can serve as valuable biomarkers for PTSD diagnosis[ 29 ]. However, it is also important to note the varying specificity across models. While RF showed a balanced trade-off, other models like the SVM (RBF) had very high sensitivity but low specificity, which limits their clinical utility due to the potential for false positives. These findings underscore the importance of evaluating both sensitivity and specificity in the context of clinical diagnosis[ 30 ], where the cost of false positives (e.g., misdiagnosing healthy individuals) can be significant. SHAP analysis provided additional interpretability to the machine learning model by quantifying the contribution of individual brain regions to the model's decision-making process[ 31 ]. The top-ranked brain regions, including the left and right occipital-temporal regions and the left central sulcus, showed significant contributions to PTSD classification. These regions, which had the highest SHAP values, were further correlated with PTSD symptom severity, specifically the B (intrusive recollections), C (avoidance), and D (hyperarousal) symptom clusters. The correlation analysis revealed that cortical thickness in the left occipital-temporal regions was most strongly associated with symptom severity, particularly in the avoidance and hyperarousal dimensions. This suggests that alterations in these regions may be closely tied to the core symptoms of PTSD, particularly those related to emotional regulation and threat detection. Our results highlight several important implications for both research and clinical practice. First, the combination of SBM with machine learning techniques provides a promising avenue for identifying neurobiological markers of PTSD that are not only accurate but also interpretable, paving the way for more personalized diagnostic and therapeutic strategies. The use of SHAP analysis in particular offers a transparent means of identifying which features drive classification decisions, allowing for more targeted interventions based on specific brain alterations. However, We acknowledge several limitations of the current study. Firstly, the dataset utilized in this analysis was derived from the existing ADNI and ADNI-DoD databases, which, while rigorous, inherently present certain constraints, including age range restrictions (participants were aged between 60–80 years), and limited gender diversity (predominantly male veteran population). These demographic factors restrict the generalizability of our findings to broader populations. Secondly, although careful participant screening was performed to minimize confounding factors such as cognitive impairment and TBI, residual confounding may persist due to subtle cognitive differences or unreported mild TBI cases. While we explicitly excluded individuals with moderate-to-severe TBI and significant cognitive impairment, residual confounding from unreported mild TBI or subclinical cognitive decline remains possible.Additionally, although our analysis employed rigorous k-fold cross-validation methods and controlled for nuisance variables such as age and gender using GLM, scanner site differences and other subtle unmeasured factors inherent in multisite data could still influence the observed structural differences. Lastly, the identified structural differences between PTSD and control groups may not be exclusively specific to PTSD pathology, as they could be influenced by other factors inherent to the veteran cohort, such as comorbid psychiatric conditions or subthreshold clinical symptoms that were not explicitly measured or controlled. Given these limitations, our findings should be considered exploratory, emphasizing the need for replication and further validation in larger, more diverse, and rigorously controlled longitudinal studies. Given these limitations, our findings should be regarded as exploratory. The methodological approach employed in this study, specifically integrating SHAP analysis with SBM-derived structural features and applying machine learning techniques with rigorous cross-validation, provides a novel framework for identifying clinically relevant cortical regions in PTSD. Nevertheless, further methodological exploration is warranted. Future longitudinal research utilizing multimodal neuroimaging, enhanced covariate control, and larger, more diverse populations is essential for validating our results, clarifying causative pathways, and ultimately improving the clinical interpretability and utility of machine learning-based classification approaches in PTSD research. Declarations Funding This work was supported by the National Defense Science and Technology Special Zone Innovation Project (Contract Number:21-163-00-TS-018-007-01). Competing Interests The authors declare that they have no conflicts of interest in this manuscript. Ethics Approval This study was approved by the Ethics Committee of Xuanwu hospital, Capital Medical University. Consent to Participate Informed consent was obtained from all individual participants included in the study. Consent to Publish All participants have consented to the publication of this manuscript. Author Contributions Nan Chen were responsible for the experimental and methodological design; Yulong Jia was responsible for writing the article and processing the experimental data; Benin Yang, Haotian Xin, Qunya Qi, Yu Wang for data processing; Chaoyang Huang and Jie Lu were responsible for proofreading the article. Clinical trial number Not Applicable. Availability of Data and Materials The ADNI and ADNI-DOD data used in this study were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and the ADNI Department of Defense (ADNI-DOD) database, respectively. These datasets are shared without embargo through the LONI Image and Data Archive (IDA) and are accessible upon registration and compliance with data usage policies. Ethics Approval and Consent to Participate This study was conducted in accordance with the principles of the Declaration of Helsinki. All data utilized in this research were obtained from the ADNI and ADNI-DOD databases, which have received prior ethics approval from their respective institutional review boards. Informed consent was obtained from all participants by ADNI and ADNI-DOD at the time of data collection. References Al Jowf GI, Ahmed ZT, Reijnders RA, de Nijs L, Eijssen LMT: To Predict, Prevent, and Manage Post-Traumatic Stress Disorder (PTSD): A Review of Pathophysiology, Treatment, and Biomarkers . Int J Mol Sci 2023, 24 (6). Kunimatsu A, Yasaka K, Akai H, Kunimatsu N, Abe O: MRI findings in posttraumatic stress disorder . Journal of magnetic resonance imaging : JMRI 2020, 52 (2):380-396. Li L, Zhang Y, Zhao Y, Li Z, Kemp GJ, Wu M, Gong Q: Cortical thickness abnormalities in patients with post-traumatic stress disorder: A vertex-based meta-analysis . Neurosci Biobehav Rev 2022, 134 :104519. Prins D, Plank T, Baseler HA, Gouws AD, Beer A, Morland AB, Greenlee MW, Cornelissen FW: Surface-Based Analyses of Anatomical Properties of the Visual Cortex in Macular Degeneration . PLoS One 2016, 11 (1):e0146684. Cornwell H, Toschi N, Hamilton-Giachritsis C, Staginnus M, Smaragdi A, Gonzalez-Madruga K, Mackes N, Rogers J, Martinelli A, Kohls G et al : Identifying cortical structure markers of resilience to adversity in young people using surface-based morphometry . Soc Cogn Affect Neurosci 2024, 19 (1). Nisar S, Bhat AA, Hashem S, Syed N, Yadav SK, Uddin S, Fakhro K, Bagga P, Thompson P, Reddy R et al : Genetic and Neuroimaging Approaches to Understanding Post-Traumatic Stress Disorder . Int J Mol Sci 2020, 21 (12). Yuan M, Li L, Zhu H, Zheng B, Lui S, Zhang W: Cortical morphological changes and associated transcriptional signatures in post-traumatic stress disorder and psychological resilience . BMC medicine 2024, 22 (1):431. Crombie KM, Ross MC, Letkiewicz AM, Sartin-Tarm A, Cisler JM: Differential relationships of PTSD symptom clusters with cortical thickness and grey matter volumes among women with PTSD . Scientific reports 2021, 11 (1):1825. Akiki TJ, Averill CL, Wrocklage KM, Schweinsburg B, Scott JC, Martini B, Averill LA, Southwick SM, Krystal JH, Abdallah CG: The Association of PTSD Symptom Severity with Localized Hippocampus and Amygdala Abnormalities . Chronic stress (Thousand Oaks, Calif) 2017, 1 . Davatzikos C: Machine learning in neuroimaging: Progress and challenges . Neuroimage 2019, 197 :652-656. Yuan J, Ran X, Liu K, Yao C, Yao Y, Wu H, Liu Q: Machine learning applications on neuroimaging for diagnosis and prognosis of epilepsy: A review . J Neurosci Methods 2022, 368 :109441. Walter M, Alizadeh S, Jamalabadi H, Lueken U, Dannlowski U, Walter H, Olbrich S, Colic L, Kambeitz J, Koutsouleris N et al : Translational machine learning for psychiatric neuroimaging . Prog Neuropsychopharmacol Biol Psychiatry 2019, 91 :113-121. Andreev AV, Kurkin SA, Stoyanov D, Badarin AA, Paunova R, Hramov AE: Toward interpretability of machine learning methods for the classification of patients with major depressive disorder based on functional network measures . Chaos 2023, 33 (6). O’Leary B, Shih C-H, Chen T, Xie H, Cotton AS, Xu KS, Morey R, Wang X, Workgroup E-PGCP: Classification of PTSD and Non-PTSD Using Cortical Structural Measures in Machine Learning Analyses—Preliminary Study of ENIGMA-Psychiatric Genomics Consortium PTSD Workgroup . In: International Conference on Brain Informatics: 2020 : Springer; 2020: 118-127. Lundberg SM, Lee S-I: A unified approach to interpreting model predictions . In: Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, California, USA: Curran Associates Inc.; 2017: 4768–4777. Fujihara K, Takei Y: [FreeSurfer as a Platform for Associating Brain Structure with Function] . Brain Nerve 2018, 70 (7):841-848. Garcia-Saldivar P, Garimella A, Garza-Villarreal EA, Mendez FA, Concha L, Merchant H: PREEMACS: Pipeline for preprocessing and extraction of the macaque brain surface . Neuroimage 2021, 227 :117671. Fischl B, Salat DH, Van Der Kouwe AJ, Makris N, Ségonne F, Quinn BT, Dale AM: Sequence-independent segmentation of magnetic resonance images . Neuroimage 2004, 23 :S69-S84. Ségonne F, Dale AM, Busa E, Glessner M, Salat D, Hahn HK, Fischl B: A hybrid approach to the skull stripping problem in MRI . Neuroimage 2004, 22 (3):1060-1075. Joy A, Nagarajan R, Daar ES, Paul J, Saucedo A, Yadav SK, Guerrero M, Haroon E, Macey P, Thomas MA: Alterations of gray and white matter volumes and cortical thickness in treated HIV-positive patients . Magnetic Resonance Imaging 2023, 95 :27-38. Storey JD: The positive false discovery rate: a Bayesian interpretation and the q-value . The annals of statistics 2003, 31 (6):2013-2035. Handelman GS, Kok HK, Chandra RV, Razavi AH, Lee MJ, Asadi H: eDoctor: machine learning and the future of medicine . J Intern Med 2018, 284 (6):603-619. Moro S, Cortez P, Rita P: A data-driven approach to predict the success of bank telemarketing . Decision Support Systems 2014, 62 :22-31. Yang L, Li H, Meng Y, Shi Y, Ge A, Zhang G, Liu C: Dynamic changes in brain structure in patients with post-traumatic stress disorder after motor vehicle accident: A voxel-based morphometry-based follow-up study . Front Psychol 2022, 13 :1018276. Liu Y, Li YJ, Luo EP, Lu HB, Yin H: Cortical thinning in patients with recent onset post-traumatic stress disorder after a single prolonged trauma exposure . PLoS One 2012, 7 (6):e39025. Crombie KM, Ross MC, Letkiewicz AM, Sartin-Tarm A, Cisler JM: Differential relationships of PTSD symptom clusters with cortical thickness and grey matter volumes among women with PTSD . Scientific reports 2021, 11 (1):1825. Cortese AM, Cacciante L, Schuler AL, Turolla A, Pellegrino G: Cortical Thickness of Brain Areas Beyond Stroke Lesions and Sensory-Motor Recovery: A Systematic Review . Front Neurosci 2021, 15 :764671. Li S, Wang S, Li X, Li Q, Li X: Abnormal surface morphology of the central sulcus in children with attention-deficit/hyperactivity disorder . Front Neuroanat 2015, 9 :114. Sussman D, Pang EW, Jetly R, Dunkley BT, Taylor MJ: Neuroanatomical features in soldiers with post-traumatic stress disorder . BMC Neurosci 2016, 17 :13. Sun D, Davis SL, Haswell CC, Swanson CA, LaBar KS, Fairbank JA, Morey RA: Brain Structural Covariance Network Topology in Remitted Posttraumatic Stress Disorder . Front Psychiatry 2018, 9 :90. Sylvester S, Sagehorn M, Gruber T, Atzmueller M, Schöne B: SHAP value-based ERP analysis (SHERPA): Increasing the sensitivity of EEG signals with explainable AI methods . Behav Res Methods 2024, 56 (6):6067-6081. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5777371","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":433312309,"identity":"2ccd82b8-66f5-411f-8fd1-8b63e0e606a0","order_by":0,"name":"Yulong Jia","email":"","orcid":"","institution":"Department of Radiology and Nuclear medicine, Xuanwu Hospital, Capital Medical University, 100053, Beijing, China","correspondingAuthor":false,"prefix":"","firstName":"Yulong","middleName":"","lastName":"Jia","suffix":""},{"id":433312311,"identity":"0b539dd5-7598-4af0-9193-30c13e4d4163","order_by":1,"name":"Beining Yang","email":"","orcid":"","institution":"Department of Radiology and Nuclear medicine, Xuanwu Hospital, Capital Medical University, 100053, Beijing, China","correspondingAuthor":false,"prefix":"","firstName":"Beining","middleName":"","lastName":"Yang","suffix":""},{"id":433312314,"identity":"acde40fa-b09c-49df-aa8e-61b59f32a62e","order_by":2,"name":"Haotian Xin","email":"","orcid":"","institution":"Department of Radiology and Nuclear medicine, Xuanwu Hospital, Capital Medical University, 100053, Beijing, China","correspondingAuthor":false,"prefix":"","firstName":"Haotian","middleName":"","lastName":"Xin","suffix":""},{"id":433312316,"identity":"55d32033-166c-4172-be54-9687ff949013","order_by":3,"name":"Qunya Qi","email":"","orcid":"","institution":"Department of Radiology and Nuclear medicine, Xuanwu Hospital, Capital Medical University, 100053, Beijing, China","correspondingAuthor":false,"prefix":"","firstName":"Qunya","middleName":"","lastName":"Qi","suffix":""},{"id":433312319,"identity":"939abfe4-860e-433c-9a56-f7f6117c6065","order_by":4,"name":"Yu Wang","email":"","orcid":"","institution":"Department of Radiology and Nuclear medicine, Xuanwu Hospital, Capital Medical University, 100053, Beijing, China","correspondingAuthor":false,"prefix":"","firstName":"Yu","middleName":"","lastName":"Wang","suffix":""},{"id":433312320,"identity":"6a4c420a-babc-4de6-b929-c3311d0109dd","order_by":5,"name":"Chaoyang Huang","email":"","orcid":"","institution":"Department of Radiology and Nuclear medicine, Xuanwu Hospital, Capital Medical University, 100053, Beijing, China","correspondingAuthor":false,"prefix":"","firstName":"Chaoyang","middleName":"","lastName":"Huang","suffix":""},{"id":433312321,"identity":"aae0fc1a-0cfa-44f8-9cca-6aaaf4ec1332","order_by":6,"name":"Jie Lu","email":"","orcid":"","institution":"Department of Radiology and Nuclear medicine, Xuanwu Hospital, Capital Medical University, 100053, Beijing, China","correspondingAuthor":false,"prefix":"","firstName":"Jie","middleName":"","lastName":"Lu","suffix":""},{"id":433312323,"identity":"a3b9dbff-e8db-42bc-9daa-99ce6b0e48f4","order_by":7,"name":"Nan Chen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3UlEQVRIiWNgGAWjYDACCQglxyZ//uMDBgMStBjzSzAYG5CkJXHmDAYzCaLcxT+7+djDHxV3jA1uN6RV/ii4I8/AfvjoBryW3DmWbiBx5pmcwZ0Dx27zGDwzbOBJS7uBT4uBRI6ZhGHbYWODA4lttxkMDjM2SPCYEdCS/00i8d/hxA0HktkKfxgctidCSw6bxMGGw0Dvp7Ex8BgcTiSoReJGmplkw7HDxvw8Z5ilgVqS2wj5hX9G8jPJHzWH5djYexg//vhz2Laf/fAxvFowARtpykfBKBgFo2AUYAMAu55MZBGonEwAAAAASUVORK5CYII=","orcid":"","institution":"Department of Radiology and Nuclear medicine, Xuanwu Hospital, Capital Medical University, 100053, Beijing, China","correspondingAuthor":true,"prefix":"","firstName":"Nan","middleName":"","lastName":"Chen","suffix":""}],"badges":[],"createdAt":"2025-01-07 03:08:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5777371/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5777371/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":79177833,"identity":"7c855143-e85d-4429-ab55-377ee136fd23","added_by":"auto","created_at":"2025-03-25 10:07:07","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":92064,"visible":true,"origin":"","legend":"\u003cp\u003eROC curve of the multi-model (\u003cem\u003ep\u003c/em\u003e＜0.05,Averaged over 1,000 Permutation Tests).\u003c/p\u003e","description":"","filename":"Figure1.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5777371/v1/6df192ab3f29684430e34b57.jpg"},{"id":79180119,"identity":"b5e41ec5-1b26-4983-9e1c-923d7dd82fa1","added_by":"auto","created_at":"2025-03-25 10:23:07","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":84196,"visible":true,"origin":"","legend":"\u003cp\u003eFeature Contributions to Model Classification,The left panel depicts SHAP value distributions for individual subjects, with color gradients representing feature values from low (blue) to high (red). Features are ranked by their overall impact on model predictions. The right panel shows the mean absolute SHAP values, quantifying the average contribution magnitude of each brain region.\u003c/p\u003e","description":"","filename":"Figure2.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5777371/v1/c65bc2b29a9f53ab82f1571e.jpg"},{"id":80220653,"identity":"5da0d026-55c3-4b0d-a387-9f92053d607d","added_by":"auto","created_at":"2025-04-09 10:38:52","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2318627,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5777371/v1/f7db1ae7-1832-44e6-8b24-f36ba2ea6aba.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"PTSD Classification Based on Surface-Based Morphometry: Integrating SHAP Analysis for Key Feature Identification","fulltext":[{"header":"Introduction","content":"\u003cp\u003ePTSD is a complex psychiatric disorder that emerges following traumatic events, often resulting in chronic symptoms such as intrusive recollections, hyperarousal, emotional detachment, and avoidance behaviors, all of which significantly impair the quality of life[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. In recent years, neuroimaging has provided critical insights into the neurobiological underpinnings of PTSD, revealing that trauma impacts specific brain regions involved in emotion regulation, memory consolidation, and cognitive processing[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Among the various neuroimaging techniques, SBM has become increasingly valuable for understanding PTSD-related structural changes[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Unlike traditional volumetric analysis, SBM provides fine-grained assessments of the cortical surface, including measures such as cortical thickness, surface area, and curvature[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. These features are particularly informative for identifying subtle cortical alterations that may be linked to PTSD symptomatology, offering a more nuanced view of how PTSD manifests at the neuroanatomical level[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eNotably, SBM studies in PTSD have revealed significant structural changes in cortical regions that contribute to PTSD's characteristic symptoms. For instance, reductions in cortical thickness in the prefrontal cortex, a region critical for cognitive control and emotional regulation, are frequently associated with diminished ability to manage intrusive thoughts and emotional responses in PTSD patients[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Additionally, structural irregularities in the hippocampus, such as reduced surface area, have been linked to impairments in memory and stress response, while changes in occipital regions involved in visual processing may influence trauma-related hyperarousal and threat perception[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Despite these findings, a gap remains in directly correlating SBM-derived features with specific PTSD symptoms, as traditional group comparisons often overlook how individual cortical features relate to symptom severity or symptom clusters[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Addressing this gap could enhance our understanding of the neural basis of specific PTSD symptoms and improve the targeting of therapeutic interventions.\u003c/p\u003e \u003cp\u003eAdvances in ML further enable more sophisticated analysis of SBM features in PTSD, particularly for classification purposes[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. By analyzing high-dimensional data, ML techniques can identify patterns within complex SBM datasets that traditional approaches might miss[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], facilitating the differentiation of PTSD from healthy controls or even other psychiatric conditions[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eHowever, while ML models can achieve high classification accuracy, their lack of interpretability has limited their clinical utility[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. In clinical contexts, understanding how and why certain cortical features contribute to PTSD classification is essential for establishing confidence in the model's predictions and for advancing patient-centered applications[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. To address this limitation, recent ML developments have introduced interpretability techniques like SHAP (SHapley Additive exPlanations)[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], which can clarify the contribution of each SBM feature to the model\u0026rsquo;s predictions. SHAP values are particularly useful in neuroimaging research because they provide a transparent way to identify and rank the most important SBM features for PTSD classification, making it possible to link these features directly with PTSD symptoms or symptom clusters.\u003c/p\u003e \u003cp\u003eIn this study, we aim to bridge the gap between neuroimaging analysis and clinical interpretability by building an ML model that uses SBM features for PTSD classification and then employing SHAP analysis to interpret the model\u0026rsquo;s decision-making process. Specifically, our objectives are to construct a robust ML model that utilizes SBM-derived measures, such as cortical thickness and curvature, and to apply SHAP analysis to determine which of these features are most associated with PTSD, thereby linking structural brain changes to specific symptomatology. By doing so, we hope to enhance the clinical applicability of SBM research in PTSD by making the model\u0026rsquo;s features and their contributions to PTSD classification both transparent and interpretable. This approach has the potential to improve diagnostic precision in PTSD while also providing clinicians with a clearer understanding of the structural brain changes associated with specific PTSD symptoms, advancing both the research and clinical landscapes of PTSD neuroimaging.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eParticipant Selection\u003c/h2\u003e \u003cp\u003eThis study utilized data from the Brain Aging in ADNI-DoD database and the ADNI database (adni.loni.usc.edu), focusing on individuals aged 60 to 80 years. Given the purpose of this study to investigate PTSD-related structural differences, we carefully selected participants based on predefined inclusion and exclusion criteria:\u003c/p\u003e \u003cp\u003e(1) PTSD Group (n\u0026thinsp;=\u0026thinsp;62)\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eParticipants were selected from the ADNI-DoD database, which was designed to study the effects of PTSD and traumatic brain injury (TBI) in an aging veteran population.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePTSD status was determined using CAPS-IV total severity scores, with a threshold score of CAPS-IV\u0026thinsp;\u0026ge;\u0026thinsp;45 used to identify clinically significant PTSD symptoms, requiring at least 1 B symptom (re-experiencing), 3 C symptoms (avoidance), 2 D symptoms (hyperarousal), symptoms persisting over 1 month, and significant social or occupational impairment, as confirmed by clinical psychiatrists in the DOD-ADNI dataset. Only individuals without a history of neurodegenerative disease (e.g., Alzheimer\u0026rsquo;s disease, Parkinson\u0026rsquo;s disease) were included.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eIndividuals with severe cognitive impairment (MoCA\u0026thinsp;\u0026lt;\u0026thinsp;26, MMSE\u0026thinsp;\u0026lt;\u0026thinsp;24) were excluded to minimize confounding effects of cognitive decline.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eTo reduce potential confounding effects of TBI, individuals with a history of moderate-to-severe TBI were excluded. Mild TBI cases were reviewed, and sensitivity analyses were conducted to assess potential impacts.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003e(2) Healthy Control Group (n\u0026thinsp;=\u0026thinsp;39)\u003c/p\u003e\u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eParticipants were selected from the ADNI database to match the PTSD group in terms of age and education level.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eInclusion required no history of PTSD or significant traumatic exposure, confirmed through self-report and clinical evaluation.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eParticipants with clinically significant cognitive impairment (MoCA\u0026thinsp;\u0026lt;\u0026thinsp;26, MMSE\u0026thinsp;\u0026lt;\u0026thinsp;24) were excluded.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eIndividuals with major psychiatric disorders (e.g., major depressive disorder, schizophrenia) or neurological conditions were excluded.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eFrom the extensive ADNI and ADNI-DoD databases, finally we identified 101 subjects who met the predefined inclusion and exclusion criteria. This selection was based on a comprehensive search for individuals with high-quality T1-weighted MRI scans and complete clinical data that aligned with our study requirements.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eMRI data acquisition and Surface-based morphometry\u003c/h3\u003e\n\u003cp\u003eMRI data used in this study were derived from the existing Alzheimer's Disease Neuroimaging Initiative\u0026ndash;Department of Defense (ADNI-DoD) dataset; thus, no new MRI data were collected. The ADNI-DoD database was properly cited, and raw imaging data from this dataset consisted of high-resolution T1-weighted MRI scans.\u003c/p\u003e \u003cp\u003eMRI scans from the ADNI-DoD database were processed utilizing surface-based morphometry (SBM) analysis through FreeSurfer software[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Specifically, original T1-weighted images in DICOM format were first converted into Neuroimaging Informatics Technology Initiative (NIfTI) format, facilitating subsequent data preprocessing and analysis steps. Preprocessing of the structural MRI data included cortical reconstruction and segmentation using FreeSurfer\u0026rsquo;s standardized processing pipeline[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. This process was automated with a batch script, ensuring consistent data handling across participants[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Cortical reconstruction and volumetric segmentation were performed automatically, producing cortical thickness, surface area, curvature, and volume data aligned to the standard fsaverage template. Each derived cortical metric was spatially smoothed using a Gaussian kernel with a full-width at half maximum (FWHM) of 10 mm to enhance comparability across participants. The smoothed data were saved for further analysis. Region of interest (ROI)-level metrics were extracted according to standardized anatomical terminology defined by the Destrieux Atlas[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Statistical analyses were conducted using general linear modeling(GLM) to examine structural differences between PTSD patients and healthy controls, accounting for covariates including age and gender. Corrections for multiple comparisons were rigorously performed using false discovery rate (FDR) and Monte Carlo simulation methods[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. After statistical analysis, significant anatomical clusters were identified and summarized for further interpretation and reporting.\u003c/p\u003e\n\u003ch3\u003eClassification model selection\u003c/h3\u003e\n\u003cp\u003eTo identify the optimal classification model, we utilized the scikit-learn library to review and test machine learning models based on significant differences in SBM features for MRI-based PTSD classification. The candidate models included Support Vector Machine (SVM) with linear and RBF kernels, Random Forest (RF), Logistic Regression (LR), Decision Tree, and Extreme Gradient Boosting (XGBoost)[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. We split the dataset into training and test sets at a 9:1 ratio. Hyperparameter tuning for each classifier was conducted using embedded 10-fold cross-validation on the training set. To ensure the statistical validity of the classification performance, we ran 1,000 permutation tests, enabling us to evaluate the robustness of the model by comparing true results to performance under randomized label assignments. Finally, we assessed and compared the classifiers\u0026rsquo; performance on the test dataset.\u003c/p\u003e \u003cp\u003eTo interpret the contribution of each feature in the classification of PTSD, we employed SHAP values. SHAP values provide a consistent measure of feature importance by quantifying the contribution of each feature to the model\u0026rsquo;s output on an individual basis[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. We computed SHAP values for the best-performing model using the SHAP Python library, which integrates with the scikit-learn framework to explain model predictions. The SHAP analysis involved calculating the average impact of each feature across all predictions, highlighting the most influential brain regions for PTSD classification. By visualizing SHAP values, we identified features that significantly contributed to model decisions, providing insights into which cortical characteristics were most predictive of PTSD. This feature importance analysis was integral in understanding the model\u0026rsquo;s decision-making process and validating the clinical relevance of identified biomarkers.\u003c/p\u003e\n\u003ch3\u003eCorrelation Analysis Between SHAP-Ranked Brain Regions and PTSD Symptoms\u003c/h3\u003e\n\u003cp\u003eTo further explore the relationship between model-identified brain regions and PTSD symptom severity, we conducted a Spearman correlation analysis. Specifically, we compared all differential brain regions with CAPS total scores as well as B, C, and D symptom cluster scores from the CAPS assessment, using Spearman's rank correlation to capture potential monotonic relationships.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eDemographic and clinical characteristics\u003c/h2\u003e \u003cp\u003eA total of 62 participants with PTSD and 39 healthy controls (HC) were included in this study. No significant differences were found between the PTSD and HC groups in terms of age (PTSD: 70.2\u0026thinsp;\u0026plusmn;\u0026thinsp;5.8 years; HC: 69.8\u0026thinsp;\u0026plusmn;\u0026thinsp;5.6 years; \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05). Additionally, no significant differences were identified between the PTSD and HC groups regarding cognitive assessments, including the MoCA scores (PTSD: 27.5\u0026thinsp;\u0026plusmn;\u0026thinsp;1.5; HC: 27.6\u0026thinsp;\u0026plusmn;\u0026thinsp;1.2; p\u0026thinsp;=\u0026thinsp;0.261) and MMSE scores (PTSD: 28.0\u0026thinsp;\u0026plusmn;\u0026thinsp;1.8; HC: 28.1\u0026thinsp;\u0026plusmn;\u0026thinsp;1.5; \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.233). Similarly, no significant differences were observed in depressive symptoms as assessed by the Geriatric Depression Scale (GDS; PTSD: 1.85\u0026thinsp;\u0026plusmn;\u0026thinsp;1.85; HC: 1.85\u0026thinsp;\u0026plusmn;\u0026thinsp;1.80; \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.315). None of the participants reported significant cognitive impairment or depressive symptoms. Notably, there was no significant difference in the gender distribution or history of traumatic brain injury between groups (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05)(Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eAbnormal brain structural changes in PTSD patients\u003c/h3\u003e\n\u003cp\u003eSignificant structural differences were observed between the PTSD group and the control group across 16 brain regions. Decreased cortical thickness was found in the left Lingual Gyrus, Inferior Temporal Gyrus, Parahippocampal Gyrus, and the right Lingual Gyrus. Additionally, reductions in surface area were identified in the left Lingual Gyrus, as well as the right Calcarine Sulcus and Occipital Pole, while curvature decreases were detected in the left Anterior Occipital Sulcus. Volume reductions were prominent in the left and right Lingual Gyrus.\u003c/p\u003e \u003cp\u003eIn contrast, increased cortical thickness was observed in the bilateral Central Sulcus, left Intraparietal and Transverse Parietal Sulcus, and right Inferior Precentral Sulcus, along with volume increases in the bilateral Central Sulcus. These findings were statistically corrected at the cluster level using FDR correction (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01).(Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cb\u003eDemographic and Clinical Characteristics of PTSD and HC\u003c/b\u003e\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePTSD (n\u0026thinsp;=\u0026thinsp;62)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHC (n\u0026thinsp;=\u0026thinsp;39)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge (years, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e70.2\u0026thinsp;\u0026plusmn;\u0026thinsp;5.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e69.8\u0026thinsp;\u0026plusmn;\u0026thinsp;5.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.687\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEducation (years, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15.4\u0026thinsp;\u0026plusmn;\u0026thinsp;2.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e15.8\u0026thinsp;\u0026plusmn;\u0026thinsp;2.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.756\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMoCA Score (mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e27.5\u0026thinsp;\u0026plusmn;\u0026thinsp;1.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e27.6\u0026thinsp;\u0026plusmn;\u0026thinsp;1.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.261\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMMSE Score (mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e28.0\u0026thinsp;\u0026plusmn;\u0026thinsp;1.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e28.1\u0026thinsp;\u0026plusmn;\u0026thinsp;1.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.233\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGDS Score (mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.85\u0026thinsp;\u0026plusmn;\u0026thinsp;0.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.80\u0026thinsp;\u0026plusmn;\u0026thinsp;0.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.315\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTBI History (n, %)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCAPS-IV Total Score (mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e58.7\u0026thinsp;\u0026plusmn;\u0026thinsp;8.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSignificant differential brain regions in surface-based morphometry\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBrain Region\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCoordinates\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProperty\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eT value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft Lingual Gyrus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(-5.6, -68.1, 4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eArea\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-6.183\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft Anterior Occipital Sulcus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(-41.3, -64.7, 1.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCurv\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-5.767\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.039\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft Lingual Gyrus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(-12.8, -81.1, -6.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThickness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-2.882\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft Inferior Temporal Gyrus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(-54.7, -48.4, -9.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThickness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-4.786\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.003\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft Parahippocampal Gyrus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(-33.8, -11.0, -27.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThickness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-3.939\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.009\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft Central Sulcus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(-11.9, -31.5, 58.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThickness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.419\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.011\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft Central Sulcus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(-44.0, -12.5, 27.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThickness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.138\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.017\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft Intraparietal and Transverse Parietal Sulcus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(-22.7, -58.5, 33.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThickness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.502\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.031\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft Lingual Gyrus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(-13.4, -63.9, 1.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eVolume\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-5.811\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft Central Sulcus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(-41.3, -20.1, 35.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eVolume\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.768\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.005\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRight Inferior Precentral Sulcus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(38.2, 9.1, 41.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThickness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.573\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRight Lingual Gyrus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(12.4, -78.9, -4.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThickness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-3.394\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.004\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRight Central Sulcus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(25.5, -26.6, 52.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThickness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.015\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.028\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRight Calcarine Sulcus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(21.0, -56.4, 2.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eArea\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-3.272\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRight Occipital Pole\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(15.0, -93.1, -4.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eArea\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-3.085\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRight Lingual Gyrus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(6.7, -66.7,5.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eVolume\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-4.625\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eModel Performance Comparison\u003c/h2\u003e \u003cp\u003eSix machine learning models (Random Forest, SVM (Linear), SVM (RBF), XGBoost, Logistic Regression, and Decision Tree) were evaluated using ten-fold cross-validation and tested on an independent dataset. The performance metrics, including AUC, accuracy, sensitivity, and specificity, for each model are summarized as follows:\u003c/p\u003e \u003cp\u003eThe Random Forest model achieved an average AUC of 0.90, accuracy of 0.79, sensitivity of 0.88, and specificity of 0.66 during cross-validation. On the test set, it maintained a high accuracy of 0.91 and an AUC of 0.75, with 100% sensitivity and 75% specificity.The SVM (Linear) model showed a slightly higher average AUC of 0.93, accuracy of 0.79, sensitivity of 0.91, and specificity of 0.59 in cross-validation. On the test set, it delivered an accuracy of 0.91, sensitivity of 1.00, an AUC of 0.79, and specificity of 0.75.The SVM (RBF) model demonstrated a high sensitivity of 0.98 in cross-validation, but its specificity was low (0.12), resulting in an average AUC of 0.92. On the test set, despite achieving perfect sensitivity (1.00), the model performed poorly in terms of specificity (0.00), with an accuracy of 0.64 and an AUC of 0.79. The XGBoost model performed with an average AUC of 0.86, accuracy of 0.74, sensitivity of 0.82, and specificity of 0.64 during cross-validation. On the test set, it showed an accuracy of 0.82, sensitivity of 1.00, and an AUC of 0.64, with a lower specificity of 0.50. The Logistic Regression model exhibited strong performance with an average AUC of 0.93, accuracy of 0.70, sensitivity of 0.95, and specificity of 0.33 during cross-validation. On the test set, it achieved an AUC of 0.86, accuracy of 0.91, sensitivity of 1.00, and specificity of 0.75. The Decision Tree model, despite its lower average AUC of 0.67, accuracy of 0.68, sensitivity of 0.71, and specificity of 0.62 in cross-validation, showed a better performance on the test set with an accuracy of 0.82, sensitivity of 1.00, AUC of 0.75, and specificity of 0.50(Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e2\u003c/span\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of Classification Performance Among Different Machine Learning Models for PTSD Diagnosis\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAUC (Training Set)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAccuracy (Training Set)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSensitivity (Training Set)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSpecificity (Training Set)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC (Test Set)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eAccuracy (Test Set)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSensitivity (Test Set)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eSpecificity (Test Set)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.75\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVM (Linear)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.75\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVM (RBF)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.50\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.75\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDecision Tree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.50\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eSHAP Analysis of Feature Importance\u003c/h2\u003e \u003cp\u003eThe SHAP value analysis revealed several key brain regions that contribute to the classification of PTSD versus the control group. The intraparietal sulcus and posterior transverse segment and the medial temporal hippocampal gyrus showed the highest positive contributions to PTSD classification, highlighting their significant role in the model. Additionally, regions associated with the central sulcus had a significant impact on the classification results in both positive and negative directions. Among the volume and thickness-related features, the central region volume, lingual gyrus volume, and inferior temporal gyrus thickness were also identified as important contributors. Other significant brain regions included the lingual gyrus, occipital pole, and anterior occipital cortex, whose structural changes play a crucial role in classification(Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eAssociation between differential SBM and clinical measures\u003c/h2\u003e \u003cp\u003eCorrelation analysis revealed a significant association between the curvature of the anterior occipital sulcus and the C symptom cluster of CAPS. A Spearman correlation analysis indicated a p-value of 0.04, suggesting a moderate relationship between structural variations in this brain region and the severity of C symptoms in PTSD.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, we sought to advance the understanding of PTSD by integrating SBM with ML techniques to classify PTSD and identify critical brain features, enhancing interpretability through SHAP analysis. Our findings not only corroborate existing evidence of PTSD-related structural brain changes but also introduce a novel, interpretable framework for leveraging SBM-derived features in PTSD classification. By pinpointing key regions such as the occipital-temporal areas and central sulcus, and linking these to specific symptom clusters, our approach offers new insights into the neurobiological underpinnings of PTSD and its potential diagnostic and therapeutic applications.\u003c/p\u003e \u003cp\u003eOur results align with a substantial body of PTSD neuroimaging research that has consistently identified structural alterations in regions involved in emotion regulation, memory, and sensory processing[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e], Specifically, we observed reduced cortical thickness, surface area, and volume in the left and right occipital-temporal regions, including the lingual gyrus and parahippocampal gyrus, alongside curvature changes in occipital areas. These findings resonate with prior studies reporting cortical thinning in the occipital-temporal cortex and hippocampus, which are implicated in emotional dysregulation, memory impairments, and heightened threat perception\u0026mdash;hallmark symptoms of PTSD[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. For instance, the lingual gyrus, a region critical for visual processing, has been associated with trauma-related hyperarousal and intrusive visual recollections[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], while the parahippocampal gyrus\u0026rsquo; role in contextual memory processing may underlie difficulties in distinguishing safe from threatening environments in PTSD patients[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. The observed volume reductions in these areas further support the hypothesis of trauma-induced structural atrophy, potentially reflecting chronic stress effects on neuroplasticity and neuronal integrity[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. These changes may disrupt the brain\u0026rsquo;s ability to modulate sensory inputs and emotional responses, contributing to the persistent hypervigilance and avoidance behaviors seen in PTSD.\u003c/p\u003e \u003cp\u003eOur correlation analysis revealed a significant association between the curvature of the anterior occipital sulcus and the C symptom cluster (avoidance) of CAPS. While direct literature on the anterior occipital sulcus in PTSD is limited, this finding partially aligns with studies like Crombie et al.[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], which linked occipital region structural changes to specific PTSD symptom clusters, including avoidance. The anterior occipital sulcus, located near visual processing areas, may influence avoidance behaviors by altering sensory integration of environmental cues. A potential reason for this correlation could be that changes in curvature reflect disrupted visual-spatial processing, leading to heightened avoidance of trauma-related stimuli as patients struggle to contextualize or filter sensory input. This suggests a nuanced role for occipital regions in PTSD symptomatology, though further research is needed to validate this link and explore underlying mechanisms.\u003c/p\u003e \u003cp\u003eConversely, we found increased cortical thickness and volume in the bilateral central sulcus and precentral gyrus, regions associated with motor function and sensory integration. This observation is less commonly reported in PTSD literature but may suggest compensatory plasticity, where the brain adapts to trauma-related deficits in other areas. Similar compensatory mechanisms have been noted in other neuropsychiatric conditions, such as ADHD and stroke recovery, where increased thickness in specific regions is thought to offset functional impairments[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].In the context of PTSD, this could reflect an adaptive response to heightened motor reactivity (e.g., startle responses) or efforts to maintain cognitive control amidst emotional dysregulation. However, this finding warrants cautious interpretation, as it contrasts with the predominant narrative of cortical thinning in PTSD. It may indicate heterogeneity in PTSD neurobiology, potentially linked to differences in trauma type, duration, or individual resilience factors, as suggested by recent studies[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Future longitudinal research could clarify whether these increases represent a protective mechanism or a maladaptive response to chronic stress.\u003c/p\u003e \u003cp\u003eThe integration of ML in this context allowed us to classify PTSD patients based on their SBM-derived features with considerable accuracy. Among the various classifiers tested, the RF model demonstrated the highest performance, achieving an average AUC of 0.90 and an impressive sensitivity of 0.88 during cross-validation. The model's ability to generalize, as shown by its high test set accuracy of 0.91, highlights the power of using ML for PTSD classification. This supports the idea that SBM features, when paired with advanced ML techniques, can serve as valuable biomarkers for PTSD diagnosis[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. However, it is also important to note the varying specificity across models. While RF showed a balanced trade-off, other models like the SVM (RBF) had very high sensitivity but low specificity, which limits their clinical utility due to the potential for false positives. These findings underscore the importance of evaluating both sensitivity and specificity in the context of clinical diagnosis[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], where the cost of false positives (e.g., misdiagnosing healthy individuals) can be significant.\u003c/p\u003e \u003cp\u003eSHAP analysis provided additional interpretability to the machine learning model by quantifying the contribution of individual brain regions to the model's decision-making process[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. The top-ranked brain regions, including the left and right occipital-temporal regions and the left central sulcus, showed significant contributions to PTSD classification. These regions, which had the highest SHAP values, were further correlated with PTSD symptom severity, specifically the B (intrusive recollections), C (avoidance), and D (hyperarousal) symptom clusters. The correlation analysis revealed that cortical thickness in the left occipital-temporal regions was most strongly associated with symptom severity, particularly in the avoidance and hyperarousal dimensions. This suggests that alterations in these regions may be closely tied to the core symptoms of PTSD, particularly those related to emotional regulation and threat detection.\u003c/p\u003e \u003cp\u003eOur results highlight several important implications for both research and clinical practice. First, the combination of SBM with machine learning techniques provides a promising avenue for identifying neurobiological markers of PTSD that are not only accurate but also interpretable, paving the way for more personalized diagnostic and therapeutic strategies. The use of SHAP analysis in particular offers a transparent means of identifying which features drive classification decisions, allowing for more targeted interventions based on specific brain alterations.\u003c/p\u003e \u003cp\u003eHowever, We acknowledge several limitations of the current study. Firstly, the dataset utilized in this analysis was derived from the existing ADNI and ADNI-DoD databases, which, while rigorous, inherently present certain constraints, including age range restrictions (participants were aged between 60\u0026ndash;80 years), and limited gender diversity (predominantly male veteran population). These demographic factors restrict the generalizability of our findings to broader populations. Secondly, although careful participant screening was performed to minimize confounding factors such as cognitive impairment and TBI, residual confounding may persist due to subtle cognitive differences or unreported mild TBI cases. While we explicitly excluded individuals with moderate-to-severe TBI and significant cognitive impairment, residual confounding from unreported mild TBI or subclinical cognitive decline remains possible.Additionally, although our analysis employed rigorous k-fold cross-validation methods and controlled for nuisance variables such as age and gender using GLM, scanner site differences and other subtle unmeasured factors inherent in multisite data could still influence the observed structural differences. Lastly, the identified structural differences between PTSD and control groups may not be exclusively specific to PTSD pathology, as they could be influenced by other factors inherent to the veteran cohort, such as comorbid psychiatric conditions or subthreshold clinical symptoms that were not explicitly measured or controlled. Given these limitations, our findings should be considered exploratory, emphasizing the need for replication and further validation in larger, more diverse, and rigorously controlled longitudinal studies.\u003c/p\u003e \u003cp\u003eGiven these limitations, our findings should be regarded as exploratory. The methodological approach employed in this study, specifically integrating SHAP analysis with SBM-derived structural features and applying machine learning techniques with rigorous cross-validation, provides a novel framework for identifying clinically relevant cortical regions in PTSD. Nevertheless, further methodological exploration is warranted. Future longitudinal research utilizing multimodal neuroimaging, enhanced covariate control, and larger, more diverse populations is essential for validating our results, clarifying causative pathways, and ultimately improving the clinical interpretability and utility of machine learning-based classification approaches in PTSD research.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the National Defense Science and Technology Special Zone Innovation Project (Contract Number:21-163-00-TS-018-007-01).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no conflicts of interest in this manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics Approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Ethics Committee of Xuanwu hospital, Capital Medical University.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to Participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eInformed consent was obtained from all individual participants included in the study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to Publish\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll participants have consented to the publication of this manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNan Chen were responsible for the experimental and methodological design; Yulong Jia was responsible for writing the article and processing the experimental data; Benin Yang, Haotian Xin, Qunya Qi, Yu Wang for data processing; Chaoyang Huang and Jie Lu were responsible for proofreading the article.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical trial number\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot Applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of Data and Materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe ADNI and ADNI-DOD data used in this study were obtained from the Alzheimer\u0026rsquo;s Disease Neuroimaging Initiative (ADNI) and the ADNI Department of Defense (ADNI-DOD) database, respectively. These datasets are shared without embargo through the LONI Image and Data Archive (IDA) and are accessible upon registration and compliance with data usage policies.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics Approval and Consent to Participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was conducted in accordance with the principles of the Declaration of Helsinki. All data utilized in this research were obtained from the ADNI and ADNI-DOD databases, which have received prior ethics approval from their respective institutional review boards. Informed consent was obtained from all participants by ADNI and ADNI-DOD at the time of data collection.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAl Jowf GI, Ahmed ZT, Reijnders RA, de Nijs L, Eijssen LMT: \u003cstrong\u003eTo Predict, Prevent, and Manage Post-Traumatic Stress Disorder (PTSD): A Review of Pathophysiology, Treatment, and Biomarkers\u003c/strong\u003e. \u003cem\u003eInt J Mol Sci \u003c/em\u003e2023, \u003cstrong\u003e24\u003c/strong\u003e(6).\u003c/li\u003e\n\u003cli\u003eKunimatsu A, Yasaka K, Akai H, Kunimatsu N, Abe O: \u003cstrong\u003eMRI findings in posttraumatic stress disorder\u003c/strong\u003e. \u003cem\u003eJournal of magnetic resonance imaging : JMRI \u003c/em\u003e2020, \u003cstrong\u003e52\u003c/strong\u003e(2):380-396.\u003c/li\u003e\n\u003cli\u003eLi L, Zhang Y, Zhao Y, Li Z, Kemp GJ, Wu M, Gong Q: \u003cstrong\u003eCortical thickness abnormalities in patients with post-traumatic stress disorder: A vertex-based meta-analysis\u003c/strong\u003e. \u003cem\u003eNeurosci Biobehav Rev \u003c/em\u003e2022, \u003cstrong\u003e134\u003c/strong\u003e:104519.\u003c/li\u003e\n\u003cli\u003ePrins D, Plank T, Baseler HA, Gouws AD, Beer A, Morland AB, Greenlee MW, Cornelissen FW: \u003cstrong\u003eSurface-Based Analyses of Anatomical Properties of the Visual Cortex in Macular Degeneration\u003c/strong\u003e. \u003cem\u003ePLoS One \u003c/em\u003e2016, \u003cstrong\u003e11\u003c/strong\u003e(1):e0146684.\u003c/li\u003e\n\u003cli\u003eCornwell H, Toschi N, Hamilton-Giachritsis C, Staginnus M, Smaragdi A, Gonzalez-Madruga K, Mackes N, Rogers J, Martinelli A, Kohls G\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eIdentifying cortical structure markers of resilience to adversity in young people using surface-based morphometry\u003c/strong\u003e. \u003cem\u003eSoc Cogn Affect Neurosci \u003c/em\u003e2024, \u003cstrong\u003e19\u003c/strong\u003e(1).\u003c/li\u003e\n\u003cli\u003eNisar S, Bhat AA, Hashem S, Syed N, Yadav SK, Uddin S, Fakhro K, Bagga P, Thompson P, Reddy R\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eGenetic and Neuroimaging Approaches to Understanding Post-Traumatic Stress Disorder\u003c/strong\u003e. \u003cem\u003eInt J Mol Sci \u003c/em\u003e2020, \u003cstrong\u003e21\u003c/strong\u003e(12).\u003c/li\u003e\n\u003cli\u003eYuan M, Li L, Zhu H, Zheng B, Lui S, Zhang W: \u003cstrong\u003eCortical morphological changes and associated transcriptional signatures in post-traumatic stress disorder and psychological resilience\u003c/strong\u003e. \u003cem\u003eBMC medicine \u003c/em\u003e2024, \u003cstrong\u003e22\u003c/strong\u003e(1):431.\u003c/li\u003e\n\u003cli\u003eCrombie KM, Ross MC, Letkiewicz AM, Sartin-Tarm A, Cisler JM: \u003cstrong\u003eDifferential relationships of PTSD symptom clusters with cortical thickness and grey matter volumes among women with PTSD\u003c/strong\u003e. \u003cem\u003eScientific reports \u003c/em\u003e2021, \u003cstrong\u003e11\u003c/strong\u003e(1):1825.\u003c/li\u003e\n\u003cli\u003eAkiki TJ, Averill CL, Wrocklage KM, Schweinsburg B, Scott JC, Martini B, Averill LA, Southwick SM, Krystal JH, Abdallah CG: \u003cstrong\u003eThe Association of PTSD Symptom Severity with Localized Hippocampus and Amygdala Abnormalities\u003c/strong\u003e. \u003cem\u003eChronic stress (Thousand Oaks, Calif) \u003c/em\u003e2017, \u003cstrong\u003e1\u003c/strong\u003e.\u003c/li\u003e\n\u003cli\u003eDavatzikos C: \u003cstrong\u003eMachine learning in neuroimaging: Progress and challenges\u003c/strong\u003e. \u003cem\u003eNeuroimage \u003c/em\u003e2019, \u003cstrong\u003e197\u003c/strong\u003e:652-656.\u003c/li\u003e\n\u003cli\u003eYuan J, Ran X, Liu K, Yao C, Yao Y, Wu H, Liu Q: \u003cstrong\u003eMachine learning applications on neuroimaging for diagnosis and prognosis of epilepsy: A review\u003c/strong\u003e. \u003cem\u003eJ Neurosci Methods \u003c/em\u003e2022, \u003cstrong\u003e368\u003c/strong\u003e:109441.\u003c/li\u003e\n\u003cli\u003eWalter M, Alizadeh S, Jamalabadi H, Lueken U, Dannlowski U, Walter H, Olbrich S, Colic L, Kambeitz J, Koutsouleris N\u003cem\u003e et al\u003c/em\u003e: \u003cstrong\u003eTranslational machine learning for psychiatric neuroimaging\u003c/strong\u003e. \u003cem\u003eProg Neuropsychopharmacol Biol Psychiatry \u003c/em\u003e2019, \u003cstrong\u003e91\u003c/strong\u003e:113-121.\u003c/li\u003e\n\u003cli\u003eAndreev AV, Kurkin SA, Stoyanov D, Badarin AA, Paunova R, Hramov AE: \u003cstrong\u003eToward interpretability of machine learning methods for the classification of patients with major depressive disorder based on functional network measures\u003c/strong\u003e. \u003cem\u003eChaos \u003c/em\u003e2023, \u003cstrong\u003e33\u003c/strong\u003e(6).\u003c/li\u003e\n\u003cli\u003eO\u0026rsquo;Leary B, Shih C-H, Chen T, Xie H, Cotton AS, Xu KS, Morey R, Wang X, Workgroup E-PGCP: \u003cstrong\u003eClassification of PTSD and Non-PTSD Using Cortical Structural Measures in Machine Learning Analyses\u0026mdash;Preliminary Study of ENIGMA-Psychiatric Genomics Consortium PTSD Workgroup\u003c/strong\u003e. In: \u003cem\u003eInternational Conference on Brain Informatics: 2020\u003c/em\u003e: Springer; 2020: 118-127.\u003c/li\u003e\n\u003cli\u003eLundberg SM, Lee S-I: \u003cstrong\u003eA unified approach to interpreting model predictions\u003c/strong\u003e. In: \u003cem\u003eProceedings of the 31st International Conference on Neural Information Processing Systems.\u003c/em\u003e Long Beach, California, USA: Curran Associates Inc.; 2017: 4768\u0026ndash;4777.\u003c/li\u003e\n\u003cli\u003eFujihara K, Takei Y: \u003cstrong\u003e[FreeSurfer as a Platform for Associating Brain Structure with Function]\u003c/strong\u003e. \u003cem\u003eBrain Nerve \u003c/em\u003e2018, \u003cstrong\u003e70\u003c/strong\u003e(7):841-848.\u003c/li\u003e\n\u003cli\u003eGarcia-Saldivar P, Garimella A, Garza-Villarreal EA, Mendez FA, Concha L, Merchant H: \u003cstrong\u003ePREEMACS: Pipeline for preprocessing and extraction of the macaque brain surface\u003c/strong\u003e. \u003cem\u003eNeuroimage \u003c/em\u003e2021, \u003cstrong\u003e227\u003c/strong\u003e:117671.\u003c/li\u003e\n\u003cli\u003eFischl B, Salat DH, Van Der Kouwe AJ, Makris N, S\u0026eacute;gonne F, Quinn BT, Dale AM: \u003cstrong\u003eSequence-independent segmentation of magnetic resonance images\u003c/strong\u003e. \u003cem\u003eNeuroimage \u003c/em\u003e2004, \u003cstrong\u003e23\u003c/strong\u003e:S69-S84.\u003c/li\u003e\n\u003cli\u003eS\u0026eacute;gonne F, Dale AM, Busa E, Glessner M, Salat D, Hahn HK, Fischl B: \u003cstrong\u003eA hybrid approach to the skull stripping problem in MRI\u003c/strong\u003e. \u003cem\u003eNeuroimage \u003c/em\u003e2004, \u003cstrong\u003e22\u003c/strong\u003e(3):1060-1075.\u003c/li\u003e\n\u003cli\u003eJoy A, Nagarajan R, Daar ES, Paul J, Saucedo A, Yadav SK, Guerrero M, Haroon E, Macey P, Thomas MA: \u003cstrong\u003eAlterations of gray and white matter volumes and cortical thickness in treated HIV-positive patients\u003c/strong\u003e. \u003cem\u003eMagnetic Resonance Imaging \u003c/em\u003e2023, \u003cstrong\u003e95\u003c/strong\u003e:27-38.\u003c/li\u003e\n\u003cli\u003eStorey JD: \u003cstrong\u003eThe positive false discovery rate: a Bayesian interpretation and the q-value\u003c/strong\u003e. \u003cem\u003eThe annals of statistics \u003c/em\u003e2003, \u003cstrong\u003e31\u003c/strong\u003e(6):2013-2035.\u003c/li\u003e\n\u003cli\u003eHandelman GS, Kok HK, Chandra RV, Razavi AH, Lee MJ, Asadi H: \u003cstrong\u003eeDoctor: machine learning and the future of medicine\u003c/strong\u003e. \u003cem\u003eJ Intern Med \u003c/em\u003e2018, \u003cstrong\u003e284\u003c/strong\u003e(6):603-619.\u003c/li\u003e\n\u003cli\u003eMoro S, Cortez P, Rita P: \u003cstrong\u003eA data-driven approach to predict the success of bank telemarketing\u003c/strong\u003e. \u003cem\u003eDecision Support Systems \u003c/em\u003e2014, \u003cstrong\u003e62\u003c/strong\u003e:22-31.\u003c/li\u003e\n\u003cli\u003eYang L, Li H, Meng Y, Shi Y, Ge A, Zhang G, Liu C: \u003cstrong\u003eDynamic changes in brain structure in patients with post-traumatic stress disorder after motor vehicle accident: A voxel-based morphometry-based follow-up study\u003c/strong\u003e. \u003cem\u003eFront Psychol \u003c/em\u003e2022, \u003cstrong\u003e13\u003c/strong\u003e:1018276.\u003c/li\u003e\n\u003cli\u003eLiu Y, Li YJ, Luo EP, Lu HB, Yin H: \u003cstrong\u003eCortical thinning in patients with recent onset post-traumatic stress disorder after a single prolonged trauma exposure\u003c/strong\u003e. \u003cem\u003ePLoS One \u003c/em\u003e2012, \u003cstrong\u003e7\u003c/strong\u003e(6):e39025.\u003c/li\u003e\n\u003cli\u003eCrombie KM, Ross MC, Letkiewicz AM, Sartin-Tarm A, Cisler JM: \u003cstrong\u003eDifferential relationships of PTSD symptom clusters with cortical thickness and grey matter volumes among women with PTSD\u003c/strong\u003e. \u003cem\u003eScientific reports \u003c/em\u003e2021, \u003cstrong\u003e11\u003c/strong\u003e(1):1825.\u003c/li\u003e\n\u003cli\u003eCortese AM, Cacciante L, Schuler AL, Turolla A, Pellegrino G: \u003cstrong\u003eCortical Thickness of Brain Areas Beyond Stroke Lesions and Sensory-Motor Recovery: A Systematic Review\u003c/strong\u003e. \u003cem\u003eFront Neurosci \u003c/em\u003e2021, \u003cstrong\u003e15\u003c/strong\u003e:764671.\u003c/li\u003e\n\u003cli\u003eLi S, Wang S, Li X, Li Q, Li X: \u003cstrong\u003eAbnormal surface morphology of the central sulcus in children with attention-deficit/hyperactivity disorder\u003c/strong\u003e. \u003cem\u003eFront Neuroanat \u003c/em\u003e2015, \u003cstrong\u003e9\u003c/strong\u003e:114.\u003c/li\u003e\n\u003cli\u003eSussman D, Pang EW, Jetly R, Dunkley BT, Taylor MJ: \u003cstrong\u003eNeuroanatomical features in soldiers with post-traumatic stress disorder\u003c/strong\u003e. \u003cem\u003eBMC Neurosci \u003c/em\u003e2016, \u003cstrong\u003e17\u003c/strong\u003e:13.\u003c/li\u003e\n\u003cli\u003eSun D, Davis SL, Haswell CC, Swanson CA, LaBar KS, Fairbank JA, Morey RA: \u003cstrong\u003eBrain Structural Covariance Network Topology in Remitted Posttraumatic Stress Disorder\u003c/strong\u003e. \u003cem\u003eFront Psychiatry \u003c/em\u003e2018, \u003cstrong\u003e9\u003c/strong\u003e:90.\u003c/li\u003e\n\u003cli\u003eSylvester S, Sagehorn M, Gruber T, Atzmueller M, Sch\u0026ouml;ne B: \u003cstrong\u003eSHAP value-based ERP analysis (SHERPA): Increasing the sensitivity of EEG signals with explainable AI methods\u003c/strong\u003e. \u003cem\u003eBehav Res Methods \u003c/em\u003e2024, \u003cstrong\u003e56\u003c/strong\u003e(6):6067-6081.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"PTSD, Neuroimaging, SBM, machine learning, SHAP analysis","lastPublishedDoi":"10.21203/rs.3.rs-5777371/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5777371/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003ePost-Traumatic Stress Disorder (PTSD) is associated with neurobiological alterations, which can be examined using surface-based morphology (SBM). While machine learning (ML) approaches have shown potential in classifying PTSD based on SBM features, further exploration is needed to improve interpretability and clinical relevance.\u003c/p\u003e\u003ch2\u003eObjectives\u003c/h2\u003e \u003cp\u003eThis study seeks to integrate ML-based classification of PTSD with SHAP analysis to identify important SBM features and their potential associations with PTSD symptomatology, providing insights into the structural changes underlying PTSD.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003e High-resolution T1-weighted MRI data from 101 participants (62 PTSD, 39 healthy controls) were analyzed using FreeSurfer\u0026rsquo;s SBM pipeline, extracting cortical thickness, surface area, and curvature features from the aparc.a2009s atlas. Several ML models, including Random Forest, SVM, and XGBoost, were trained and evaluated using ten-fold cross-validation. SHAP analysis was applied to determine feature importance, and correlation analyses were conducted to examine relationships between key features and PTSD symptom severity.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eSixteen cortical regions were identified with significant structural differences in PTSD, including reduced cortical thickness in the left lingual gyrus and increased thickness in the bilateral central sulcus. The Random Forest model achieved the highest accuracy (91%) in PTSD classification. SHAP analysis highlighted the left lingual gyrus and parahippocampal gyrus as key features. Correlation analysis suggested potential links between these features and specific PTSD symptom clusters.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThe integration of SBM and interpretable ML methods provides a promising approach for investigating structural brain changes in PTSD. While further validation is needed, these findings contribute to a better understanding of PTSD neurobiology and may support future research on diagnostic and therapeutic strategies.\u003c/p\u003e","manuscriptTitle":"PTSD Classification Based on Surface-Based Morphometry: Integrating SHAP Analysis for Key Feature Identification","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-03-25 10:07:02","doi":"10.21203/rs.3.rs-5777371/v1","editorialEvents":[{"type":"communityComments","content":1}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"80bd173f-c988-49bc-92ef-936bdcc83e59","owner":[],"postedDate":"March 25th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-04-09T10:38:20+00:00","versionOfRecord":[],"versionCreatedAt":"2025-03-25 10:07:02","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5777371","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5777371","identity":"rs-5777371","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}