Enhancing Multiple Sclerosis Diagnosis: Deep Transfer Learning Techniques for MRI Image Analysis
Omar Alomari1, Baran Baytar2, Yılmaz Kaya3, Fatma Sibel Bayramoğlu4, Cemil Aydin1, Merve Akçay5
1Hamidiye International School of Medicine, University of Health Sciences, Istanbul 34668, Turkey., 2Radiology Department, Bingol State Hospital, Saray, Hastane Cd. No:60, 12000 Bingöl, Türkiye, 3Department of Computer Engineering, Batman University Batıraman Campus 72000 Center 72, 060 Batman Center/Batman, Türkiye, 4Radiology Department, Sultan II. Abdulhamid Han Training and Research Hospital, Istanbul 34668, Turkey., 5Department of Pediatrics, University of Health Sciences, Darica Farabi Training and Research Hospital, Kocaeli, Turkey
Objective:
To investigate the potential of artificial intelligence (AI) and deep transfer learning techniques to enhance the detection and analysis of multiple sclerosis (MS)-specific lesions in MRI scans.
Background:
MS is a chronic, autoimmune disease characterized by demyelination in the central nervous system, leading to a range of neurological symptoms and significant long-term disability. Current diagnostic methods rely on clinical evaluation and MRI findings, which often fall short in accurately assessing axonal damage. AI has shown promise in medical imaging for improving diagnostic precision and could potentially address these limitations.
Design/Methods:
This retrospective study was conducted between 2020 and 2023. MRI data were collected from 39 male and 71 female MS patients and 32 healthy controls. We applied various deep transfer learning models (Xception, MobileNet, AlexNet, VGG19, ResNet152, and NasNet) to analyze axial and sagittal FLAIR images. Model performance was assessed using sensitivity, specificity, accuracy, and ROC curves.
Results:
For axial FLAIR images, the Xception model achieved the highest accuracy of 95.07%, while NasNet had the lowest at 70.77%. In sagittal FLAIR images, Xception again led with an accuracy of 98.10%. High specificity rates were noted, with Xception and VGG19 achieving 100% specificity. The ROC analysis indicated robust classification performance, with AUC values of 0.95 for Xception in axial images and 0.98 for both Xception and MobileNet in sagittal images.
Conclusions:
This study highlights the effective application of deep transfer learning methods in enhancing the diagnostic process for MS. While results are promising, the study's limited patient dataset underscores the need for larger, more diverse datasets to improve model performance and generalizability in clinical settings.
Disclaimer: Abstracts were not reviewed by Neurology® and do not reflect the views of Neurology® editors or staff.