Slice-Wise Augmentation for Multi-Task Learning to Predict Neuropsychological Outcomes After Traumatic Brain Injury

Traumatic brain injury (TBI) often causes lasting cognitive impairments and requires accurate predictions for personalized rehabilitation. This study evaluated the performance of multi-task deep learning models with and without slice-wise data augmentation. 3D Cerebrovascular reactivity (CVR) maps derived from resting-state fMRI at 3 months post injury along with clinical characteristics were used to predict neuropsychological outcomes at 12 months post-injury in patients with moderate to severe TBI. The model with slice-wise augmentation of 3D CVR maps outperformed the model without augmentation, achieving a mean absolute error of 9.18 ± 1.30. The study demonstrated that slice-wise augmentation of 3D CVR maps enhances the performance of multi-task deep learning models for predicting neuropsychological outcomes one year after TBI.