Efficient Classification of Multiple Sclerosis and Idiopathic Transverse Myelitis Using CNN Feature Extraction and Walrus Optimizer on MRI Scans

Abstract

Acute myelopathies, including Multiple Sclerosis (MS) and Idiopathic Transverse Myelitis (ITM), present significant diagnostic challenges due to overlapping clinical and imaging features. Accurate differentiation between these conditions and normal individuals is crucial for timely and appropriate treatment. In this study, we propose a method that utilizes Convolutional Neural Networks (CNNs), specifically ResNet50 and DenseNet201, to extract features from MRI images, followed by feature selection using the Walrus Optimizer for classification of MS, ITM, and healthy controls. A dataset of 2,746 MR images, including 128 MS patients, 131 ITM patients, and 150 healthy controls, was used for training and validation. The dataset, consisting of sagittal and axial views, captures the unique lesion characteristics of each condition, such as the length and location of spinal lesions. Our approach achieved an accuracy of over 90%, demonstrating the effectiveness of CNN-based feature extraction combined with advanced optimization techniques. This AI-driven method offers a significant advancement in non-invasive diagnostics, potentially reducing the need for additional procedures and enabling earlier and more precise clinical interventions. Our findings highlight the potential of combining CNNs with feature selection algorithms in the field of medical imaging, providing a reliable tool for the differentiation of complex neurological disorders.