Study on Resting-State fMRI Data of Mild Traumatic Brain Injury Based on Machine Learning

Wuyi Liu

Abstract: This study aims to select abnormal function connections in the resting-state function connection network (FCN) of mild traumatic brain injury (mTBI) in the acute phase, then analyze the function connections changes among different brain networks and the corresponding pathophysiological mechanisms. Sixty-nine patients with mTBI and sixty healthy subjects with age, sex, and education level matched were recruited for the study. All subjects accepted resting-state fMRI scan, neuropsychological assessment, PSQI, and treadmill testing. This study adopted the elastic net to select subjects? abnormal function connections, which were input into a support vector machine for classification validation. Furthermore, we performed a correlation analysis between the mTBI group and abnormal functional connectivity and psychometric scores. The classifier achieved accuracy = 84.62%, recall = 85.27%, precision = 84.62%, and F1 score = 84.72%. The decreased function connections between networks include visual network (VN)-visual network (VN), visual network (VN)-frontoparietal network (FPN), somatomotor network (SMN)-ventral attention network (VAN). Furthermore, there are decreased function connections between the cerebellum network (CN) with ventral attention network (VAN), somatomotor network (SMN), default mode network (DMN), and frontoparietal network (FPN). Abnormal functional connections are widely distributed in various networks. The interaction of various networks may lead to motor, executive, cognitive and emotional dysfunction in patients with mTBI. Abnormal functional connections affect brain networks and functions through global distribution and local aggregation. The most affected network areas are cerebellum, visual, and somatomotor networks. The cerebellum network and other networks have extensive coordination and interaction.

Keywords: Brain network, Elastic net, Mild traumatic brain injury, Resting-state fMRI, Support vector machine