Identifying Mild Traumatic Brain Injury Patients From MR Images Using Bag of Visual Words

摘要

Mild traumatic brain injury (mTBI) is a growing public health problem with anestimated incidence of one million people annually in US. Neurocognitive testsare used to both assess the patient condition and to monitor the patientprogress. This work aims to directly use MR images taken shortly after injuryto detect whether a patient suffers from mTBI, by incorporating machinelearning and computer vision techniques to learn features suitablediscriminating between mTBI and normal patients. We focus on 3 regions inbrain, and extract multiple patches from them, and use bag-of-visual-wordtechnique to represent each subject as a histogram of representative patternsderived from patches from all training subjects. After extracting the features,we use greedy forward feature selection, to choose a subset of features whichachieves highest accuracy. We show through experimental studies that BoWfeatures perform better than the simple mean value features which were usedpreviously.