G-3427

2025-10-19 19:21

Written by ARCIMS 26 ARCIMS 26 in Sunday 2025-10-19 19:21

Deep Learning-Based Brain Tissue Segmentation in Multiple Sclerosis Using MRI Imaging

 Reza Erfani Far 1 © ℗, Sogand Abbasi Azizi 2, Kaveh Samimi 3   

 1Student Research Committee, Iran University of Medical Sciences, Tehran, Iran. 2. Resident of Neurology, Department of Neurology, Rasoul Akram Hospital, Iran University of Medical Sciences, Tehran, Iran.

 Student Research Committee, Kermanshah University of Medical Sciences, Kermanshah, Iran

 Assistant Professor of Radiology, Department of Radiology, School of Medicine, Rasoul Akram Hospital, Iran University of Medical Sciences, Tehran, Iran.

Email: reza.efn@gmail.com
 

 


 
Abstract

Introduction: Multiple Sclerosis (MS) is a chronic inflammatory and autoimmune disease of the central nervous system, characterized by demyelination and the formation of lesions in the brain's white matter. Early diagnosis and disease monitoring heavily rely on advanced analysis of brain MRI images. One of the main challenges is accurate segmentation of MS lesions from other brain tissues such as gray and white matter. With advancements in image processing and machine learning, several new techniques have been introduced to improve segmentation accuracy. Methods and Materials: In this study, brain MRI images of 67 MS patients were obtained from the Rasoul Akram Hospital (2025), Imaging center (PACS) dataset. After preprocessing steps including skull stripping, intensity normalization, and spatial registration, three segmentation methods were evaluated: (1) automatic thresholding, (2) K-Means clustering, and (3) a deep learning model based on the U-Net architecture. Their performance was compared using metrics such as Dice Similarity Coefficient (DSC), accuracy, sensitivity, and specificity. Results: The U-Net model significantly outperformed the other methods. The average DSC for U-Net was 0.89, compared to 0.76 for K-Means and 0.65 for thresholding. U-Net achieved a sensitivity of 91.5% and an overall accuracy of 92.3% (p 0.001). In contrast, K-Means achieved a sensitivity of 82.7% and accuracy of 85.4% (p 0.001). The thresholding method, while computationally simple, demonstrated the lowest performance in lesion detection. Conclusion and Discussion: This study demonstrates that deep learning techniques, particularly the U-Net architecture, offer superior performance in segmenting MS lesions in brain MRI. These methods not only enhance the precision of lesion detection but also hold great potential for identifying early pathological changes. Integrating such advanced segmentation techniques into clinical workflows could significantly accelerate diagnosis and improve treatment planning for MS patients.


Keywords: Multiple Sclerosis, Deep Learning, Brain Segmentation, MRI images

Feedback

What is your opinion? Click on the stars you want.

Comments (0)

No Comment yet. Be the first!

Post a comment