Multiple sclerosis diagnosis with brain MRI retrieval: A deep learning approach

Multiple Sclerosis (MS) is an auto-immune disorder affecting the central nervous system, affecting 2.8 million people worldwide. Early diagnosis is crucial due to its profound social and economic impacts. MRI is commonly used for monitoring abnormalities. This study proposes a novel Content-Based Medical Image Retrieval (CBMIR) framework using Convolutional Neural Networks (CNN) and Transfer Learning (TL) for MS diagnosis using MRI data. Our framework utilizes The Inception V3 model that is pre-trained on ImageNet and RadImageNet datasets, and we modified the model by adding a new block of six layers to reduce the features’ dimensionality in the feature extraction phase. Fine-tuning the hyper-parameters for the whole system was done using the Bayesian optimizer. We experiment with Nine different distance metrics to measure query and database image similarity. Experiments on four public MS-MRI datasets demonstrated the end-to-end deep learning framework’s generalizability without extensive pre-processing, with mAP scores of 86.20%, 93.77%, 94.18%, and 90.46%, respectively demonstrating its effectiveness in retrieval. Moreover, a comparison with related CBMIR systems confirmed the effectiveness of our model.