A Diagnosis Model of Typhoon-Related Post-Traumatic Stress Disorder Based on Fixel-Based Analysis in Machine Learning

Post-traumatic stress disorder (PTSD) is the most common mental disorder following traumatic experiences. Environmental disasters such as super typhoons can severely disrupt daily life and may trigger PTSD in exposed individuals. White matter alterations have been observed in patients with PTSD. Fixel-based analysis (FBA), a recently developed diffusion MRI technique, allows detailed assessment of white matter microstructure. This study aimed to evaluate the potential of FBA as an imaging biomarker in typhoon survivors, reducing the subjective bias associated with clinical symptom scales.

Whole-brain diffusion MRI data from the PTSD group (n = 27), trauma-exposed controls (TEC, n = 33), and healthy controls (HC, n = 30) were analyzed to identify white matter fiber tracts showing abnormalities in FBA metrics, including fiber density (FD), fiber cross-section (FC), and fiber density-cross section (FDC). The study then examined whether these FBA-derived features, when combined with machine learning, could improve the identification of potential PTSD biomarkers.

Compared with the HC group, patients with PTSD showed increased fiber density (FD) in the right frontopontine tract and right middle longitudinal fascicle, as well as higher fiber density-cross section (FDC) values in the bilateral frontopontine tract and left thalamo-premotor tract (Bonferroni correction, p < 0.05/18 = 0.003). To differentiate PTSD from TEC, binary and multiclass machine learning models with five-fold cross-validation were developed. The binary model (PTSD vs. TEC) achieved high performance (accuracy = 0.89, sensitivity = 0.97, specificity = 0.71, precision = 0.87, AUC = 0.95), whereas the multiclass model (PTSD vs. TEC vs. HC) demonstrated excellent results (macro-averaged precision = 0.99, recall = 0.99, F1-score = 0.99). The top 20 contributing features of the optimal model were analyzed using Shapley additive explanation (SHAP) values to illustrate model interpretability.

Most typhoon-exposed individuals with PTSD may exhibit structural alterations in brain white matter. By combining fixel-based analysis (FBA) with machine learning, this study identified diffusion markers within specific white matter tracts and demonstrated their potential diagnostic value for distinguishing PTSD from trauma-exposed controls. These findings enhance our understanding of microstructural white matter changes and their spatial distribution in PTSD and also suggest potential imaging biomarkers for its diagnosis.