Automated brain disease classification using exemplar deep features

Abstract

Automated brain disease classification is one of the complex and widespread issues for machine learning and biomedical engineering. Various models and papers have been presented to solve automated brain disease detection models. This paper proposes an exemplar-based automated brain disease detection model using a computer vision technique. The presented model uses an exemplar-based deep feature generator. A pre-trained deep learning model is selected to generate features. Here, MobilNetV2 is selected as a feature extractor. The presented automated brain disease detection model contains four phases, and they are preprocessing, exemplar deep feature generator, feature selection using iterative neighborhood component analysis (INCA), and classification with support vector machine (SVM). In the preprocessing phase, the input brain images are resized to 512 x 512 sized images, and the used images are divided into 128 x 128 and 256 x 256 exemplars. By using MobileNetV2, 1000 features are generated from each exemplar and the resized image. The generated/extracted features are feed to the INCA feature selector, and the most valuable feature is selected. The selected feature is utilized as the input of the SVM classifier. An MR image dataset was collected for brain disease detection (it can be downloaded using https://www.kaggle.com/turkertuncer/brain-disorders-four-categories) to test the presented exemplar deep feature-based model. The collected dataset contains 444 MR images with three diseases and control (normal) category. Our model attained 99.10% classification accuracy using SVM. The success of the presented model is demonstrated by using these calculated accuracies.