Automated classification of brain diseases using the Restricted Boltzmann Machine and the Generative Adversarial Network

Abstract

Background:Early diagnosis of brain diseases is very important. Brain disease classification is a common and complex topic in biomedical engineering. Therefore, machine learning methods are the most reliable and common methods for the automatic detection of brain diseases. Materials and methods:The brain diseases dataset consists of four classes: atrophy, normal, white matter density, and ischemia. First, data augmentation and normalization is applied to the dataset. Then, deep feature extraction is applied to this preprocessed dataset with the Restricted Boltzmann Machine (RBM) method. Deep feature data are given as input to the generator unit of the Generative Adversarial Network (GAN) method. Finally, classification is performed by Tree, Linear discriminant, Naïve Bayes, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Ensemble, Neural Network classifiers and K-mean. Results:In the classification applied before feature extraction with the RBM method, the highest accuracy is calculated in the SVM classifier with 97.94%, and the lowest accuracy is calculated in the Naïve Bayes classifier with 80.43%. After applying the RBM feature extraction, the highest accuracy is calculated as 99.65% in the SVM classifier and 83.74% in the lowest Naïve Bayes classifier. Conclusion:This study shows that the performance criteria values of the presented methods is improved with RBM-GAN.