Segmentation of Multiple Sclerosis Using Autoencoder and Classifier

Multiple Sclerosis (MS) is an inflammatory demyelination of the central nervous system that progresses to complete or incomplete recovery of neurological dysfunction. Accurately identifying the degree of damage to the brain matter is critical for prognosis. Progression of MS must be closely monitored requiring regular assessment of symptoms through diagnostic methods like magnetic resonance imaging (MRI). This process is quite taxing and time-consuming to be done manually. This paper presents a methodology to perform automatic segmentation of multiple sclerosis via a combination approach of an unsupervised autoencoder and a pre-trained classifier for lesion detection followed by segmentation. The presented approach investigates the usability of an autoencoder for classification and segmentation. The novelty lies in reducing the reliance of the model on the availability of labeled data. This is a prevalent issue in the field of research for MS. The proposed approach has been evaluated on the FLAIR MRI images obtained from the members of Ozal University Medical Faculty in 2021 and is able to deliver good results.