همایش ملی بیوانفورماتیک ایران

صفحه اصلی / 4th international edition and 13th Iranian Conference on Bioinformatics

A Transformer-based Model for Diagnosis of Multiple Sclerosis using MRI Images

نویسندگان :

Elham Bagheri¹ Leila Safari²

1- Department of Electrical and Computer Engineering University of Zanjan, 2- Department of Electrical and Computer Engineering University of Zanjan,

کلمات کلیدی :

Multiple Sclerosis (MS)،Diagnosis،MRI images،Transformer Based Models،Vision Transformer(ViT)

چکیده :

Multiple Sclerosis (MS) is an autoimmune inflammatory disease that leads to the degradation of myelin in the central nervous system (brain and spinal cord), resulting in issues with movement, sensation, and balance. This disease has significant effects on individuals’ quality of life. In this study, a Vision Transformer (ViT)-based model is proposed for diagnosing MS using MRI images. One of the primary challenges in this research is the lack of a standard dataset for implementing the proposed model. To address this issue, the two techniques of transfer learning and fine-tuning have been employed. Two relatively large datasets, including over 21,000 general image samples and 5,000 pneumonia images, were used for pre-training the proposed model. Additionally, a smaller dataset containing 3,427 MRI images associated with MS, labeled with patient classes and healthy classes in both axial and sagittal views ( four classes in total), was used to fine-tune the ViT model for MS diagnosis. The results of MRI image classification with pre-trained models show an accuracy of 0.100 for the two-class scenario with both pre-training datasets, 0.96 for the four-class scenario with the pneumonia dataset, and 0.95 for the four-class scenario with the general dataset. These results indicate that dataset selection for fine-tuning the model in ViT implementation is highly significant. Moreover, the high performance of the proposed method confirms the effectiveness of the emerging ViT model and its notable superiority over traditional neural network architectures, such as deep convolutional networks.