A Novel Approach to Image Classification for Detecting Abnormalities in Neuroimages based on the Structural Similarity Index Measure

Abstract

Medical imaging has improved image quality and enables accurate diagnosis and treatment. Medical imaging is used in the early detection and diagnosis of mental disorders or mental illnesses, and treatment. This study performs image-based classification using the Structural Similarity Index Measure (SSIM) to detect normal and abnormal neuroimages. Two experiments were performed on the same dataset. 342 Dicom images were divided into standard and abnormal categories. At first, the SSIM between images was calculated. SVM, KNN, Naïve Bayes, and Decision Tree classifiers were applied and compared. Similarly, an artificial neural network using two optimizers, Adam and SGD, was applied to the same dataset. In theexperiments, 100% and 97% accuracy was achieved in image-based classification, while SSIM-based classification achieved 100% and 61% for different classifiers.