Machine Learning Classification of Traumatic Brain Injury Patients and Healthy Controls Using Multiple Indices of Diffusion Tensor Imaging
Abstract Background: Diffusion tensor imaging (DTI) indices provide quantitative measures of white matter microstructural changes following traumatic brain injury (TBI). However, there is still insufficient evidence for their use as predictive measures. Recently, there has been growing interest in using machine learning (ML) approaches to aid the diagnosis of many neurological and psychiatric illnesses including TBI. The aim of this study is to examine the potential of using multiple DTI indices in conjunction with ML to automate the classification of healthy subjects and patients with TBI across a spectrum of TBI severity.Methods: Participants were adult patients with chronic TBI (n=26) and age and gender-matched healthy controls (n=26). DTI images were obtained from all the participants. Tract-based spatial statistics (TBSS) analysis was applied to the DTI images. Classification models were built using principle component analysis (PCA) and support vector machines (SVM). Receiver operator characteristic (ROC) curve analysis and area under the curve (AUC) were used to assess the classification performance of the different classifiers.Results: The whole-brain white matter TBSS analyses showed significantly decreased FA, as well as increased MD, AD, and RD in TBI patients compared with healthy controls (all p-value < 0.01). The PCA and SVM-based ML classification using combined DTI indices classified TBI patients and healthy controls with the accuracy of 90.5% with an area under the curve (AUC) of 93 +/- 0.09.Conclusion: This study demonstrates the potential of a joint DTI and ML approach for objective classification of TBI patients and healthy controls.