A Vessel-Based Long Short-Term Memory (LSTM) Approach for Detecting Multiple Sclerosis

Multiple Sclerosis (MS) is a chronic immune-mediated condition affecting the central nervous system, often resulting in various disturbances, including visual impairments. The early and accurate diagnosis of MS is critical for effective treatment and management. Scanning Laser Ophthalmoscopy (SLO) is a non-invasive technique that offers high-quality retinal images and holds promise for early MS detection. This study explores a vessel-based approach using Long Short-Term Memory (LSTM) networks to detect MS in SLO images. The study enrolled 106 Healthy Controls (HCs) and 39 MS patients (78 eyes), after implementing quality control measures and excluding poor-quality or damaged images, resulting in a total of 265 images (73 MS and 192 HC). We present a novel approach for the early detection of MS in SLO images using LSTM networks. This approach involves two key steps: 1) Pre-training a deep neural network on a source dataset and 2) Fine-tuning the network on the target dataset consisting of SLO images. We examine the significance of vessel segmentation in MS detection and investigate the effectiveness of our proposed method in enhancing diagnostic models. Our approach achieves an impressive accuracy rate of 97.44% when evaluated on a test dataset comprising SLO images. By conducting experiments on SLO datasets and employing the vessel-based approach with LSTM, our empirical findings confirm that this method facilitates early MS detection with exceptional accuracy. These models demonstrate the capability to accurately identify the disease with high precision and appropriate sensitivity.