Introduction: According to the CDC, there were approximately 214,110 traumatic brain injury (TBI)-related hospitalizations in 2020. Post-concussive syndrome (PCS) refers to lingering symptoms that persist beyond the expected TBI recovery period. These symptoms—affecting 11–82% of post-TBI patients—include headaches, sleep disturbances, dizziness, fatigue, anxiety, tinnitus, blurry vision, memory issues, emotional lability, and photophobia. Given PCS’s prevalence and impact, objective identification and treatment are needed. Currently, PCS is diagnosed primarily through subjective questionnaires, with few objective diagnostic tools available. Advances in artificial intelligence (AI) and brain imaging offer opportunities to improve PCS diagnosis.
Methods: Our dataset of 216 human brains includes DTI metrics for specific grey matter regions in the ATAG, Brodmann, Brainseg, and CerebrA atlases. Diffusion tensor imaging (DTI) metrics analyzed include Fractional Anisotropy, Axial Diffusivity, Mean Diffusivity, and Radial Diffusivity. The normative dataset was sourced from the NIMH Healthy Research Volunteer Dataset and normalized to the PCS dataset. A machine learning model was implemented in Python to analyze the dataset, yielding accuracy, sensitivity, and specificity scores. The dataset was split 80/20 for training and testing. A Feature Importance analysis ranked factors contributing to PCS diagnosis, and a Receiver Operating Characteristic (ROC) curve was generated.
Results: The best model achieved 88% accuracy, with sensitivity and specificity of 90% and 77%, respectively. The area under the ROC curve was 0.95. Feature Importance analysis identified key DTI metrics in the cingulate cortex and temporal lobe as significant in PCS diagnosis.
Conclusion : In conclusion, we successfully implemented an AI model to predict PCS diagnosis using DTI metrics, demonstrating that DTI imaging may confirm brain connectivity disorders. Further studies are needed to validate these findings.
.jpg)
