Article: Improving multiple sclerosis identification with an advanced U-Net architecture featuring dilated convolutions Journal: International Journal of Computational Biology and Drug Design (IJCBDD) 2025 Vol.16 No.4 pp.366 - 391 Abstract: Multiple sclerosis (MS) is a disease that impacts the central nervous system (CNS), which can lead to brain, spinal cord, and optic nerve problems. A total of 2.8 million are estimated to suffer from MS. Every five minutes, a new case of MS is reported globally. Numerous deep neural network models have been developed using different types of MS data, including MRI and clinical data. However, there is no standard approach available for the identification of abnormalities (lesions) in the deep grey matter (DGM) and mesial temporal lobe (MTL) of the brain using MRI. This research proposed a U-Net-based modified architecture with dilated convolution operation for detecting MS. In this, the images from the BioGPS dataset are trained using the proposed U-Net model for extracting image features automatically. These outcome features are fed into softmax classification for performing pixel-wise probabilities of class labels. This network learns special features effectively, requiring much less computation than the traditional U-Net. This work gained 97.26% accuracy in predicting the abnormalities in MS, which is comparatively higher than the existing methods. Inderscience Publishers - linking academia, business and industry through research

Title: Improving multiple sclerosis identification with an advanced U-Net architecture featuring dilated convolutions

Authors: M. Divya; J. Dhilipan; A. Saravanan

Addresses: Department of Computer Science and Applications, SRM Institute of Science and Technology, Ramapuram, Chennai, 600089, Tamil Nadu, India ' Department of Computer Science and Applications, SRM Institute of Science and Technology, Ramapuram, Chennai, 600089, Tamil Nadu, India ' Faculty of Engineering and Technology, Sri Ramachandra Institute of Higher Education and Research (SRIHER), Porur, Chennai, 600116, Tamil Nadu, India

Abstract: Multiple sclerosis (MS) is a disease that impacts the central nervous system (CNS), which can lead to brain, spinal cord, and optic nerve problems. A total of 2.8 million are estimated to suffer from MS. Every five minutes, a new case of MS is reported globally. Numerous deep neural network models have been developed using different types of MS data, including MRI and clinical data. However, there is no standard approach available for the identification of abnormalities (lesions) in the deep grey matter (DGM) and mesial temporal lobe (MTL) of the brain using MRI. This research proposed a U-Net-based modified architecture with dilated convolution operation for detecting MS. In this, the images from the BioGPS dataset are trained using the proposed U-Net model for extracting image features automatically. These outcome features are fed into softmax classification for performing pixel-wise probabilities of class labels. This network learns special features effectively, requiring much less computation than the traditional U-Net. This work gained 97.26% accuracy in predicting the abnormalities in MS, which is comparatively higher than the existing methods.

Keywords: multiple sclerosis lesions; modified U-net; dilated convolution; pixel-wise classification; BioGPS dataset; DGM; deep grey matter; MTL; mesial temporal lobe.

DOI: 10.1504/IJCBDD.2025.151211

International Journal of Computational Biology and Drug Design, 2025 Vol.16 No.4, pp.366 - 391

Received: 02 May 2024
Accepted: 01 Jan 2025
Published online: 19 Jan 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Improving multiple sclerosis identification with an advanced U-Net architecture featuring dilated convolutions

Keep up-to-date