Diagnosis of Dementia Using Shapley Additive Explanations Based Feature Selection Method

Alzheimer's is a disease that is often diagnosed in persons 65 years of age and older. The main aim of this work is to classify Alzheimer's disease with reduced computational complexity. Labeled 1188 transverse MR Images of the brain belonging to four severity classes are obtained from the Alzheimer Disease Neuroimaging Initiative (ADNI) database. The MR images are preprocessed and segmented using hybrid algorithms. Various texture features are extracted along with deep features from the segmented ventricle images. The Shapley Additive exPlanations (SHAP) framework is used for the feature selection process. The classification of selected features is carried out by a classifier merger. It consists of 3 classifiers with a majority voting merging technique. The results show that using a minimal set of 63 features out of 2115, an accuracy of 0.9891 is obtained with reduced complexity.