Scholarlink

Multiple Sclerosis (MS) is a chronic, autoimmune, neuroinflammatory, and neurodegenerative disease of the central nervous system (CNS) with the occurrence of demyelinated lesions in the white and grey matter. The exact mechanism of the disease initiation and progression is unknown. While the visible lesions are commonly seen in the white and grey matter, many areas that appear normal in a magnetic resonance imaging (MRI) have underlying damage. These areas are referred to as normal-appearing white matter (NAWM) and normal-appearing grey matter (NAGM). In addition, the control white matter (CWM) and control grey matter (CGM) are from healthy individuals without MS for comparison. This work uses a novel deep learning model using a one-vs-rest classification strategy to differentiate among these four brain tissue types from Scanning Transmission Electron Microscopy (STEM) Images. The methodology integrates traditional convolutional neural networks (CNNs) with advanced transfer learning architectures, VGG16, ResNet50, and EfficientNetB0. Generative Adversarial Network (GAN)-based augmentation was used to generate synthetic images to address data imbalance and improve generalization. With 280 balanced and GAN-augmented images per class, the dataset was split using an 80:20 train-test ratio. Among the models tested, VGG16 performed the best, obtaining the highest overall classification accuracy of 93.19% with a precision and AUROC close to 0.9. This study presents a promising approach for automating brain tissue classification in MS patients. The ability to precisely distinguish between tissue types enables early detection and targeted treatment, which may improve outcomes for future MS patients by assisting clinicians with disease monitoring and treatment planning. This approach could support integration into future digital pathology platforms for MS diagnosis.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

GAN-Based Transfer Learning Model for Brain Tissue Classification in Progressive Multiple Sclerosis Using Electron Microscopy Images

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • S. Anjanaa,
  • K. M. Dhyaneshwar,
  • Namasivaya Naveen Shanmuga Sundaram

摘要

Multiple Sclerosis (MS) is a chronic, autoimmune, neuroinflammatory, and neurodegenerative disease of the central nervous system (CNS) with the occurrence of demyelinated lesions in the white and grey matter. The exact mechanism of the disease initiation and progression is unknown. While the visible lesions are commonly seen in the white and grey matter, many areas that appear normal in a magnetic resonance imaging (MRI) have underlying damage. These areas are referred to as normal-appearing white matter (NAWM) and normal-appearing grey matter (NAGM). In addition, the control white matter (CWM) and control grey matter (CGM) are from healthy individuals without MS for comparison. This work uses a novel deep learning model using a one-vs-rest classification strategy to differentiate among these four brain tissue types from Scanning Transmission Electron Microscopy (STEM) Images. The methodology integrates traditional convolutional neural networks (CNNs) with advanced transfer learning architectures, VGG16, ResNet50, and EfficientNetB0. Generative Adversarial Network (GAN)-based augmentation was used to generate synthetic images to address data imbalance and improve generalization. With 280 balanced and GAN-augmented images per class, the dataset was split using an 80:20 train-test ratio. Among the models tested, VGG16 performed the best, obtaining the highest overall classification accuracy of 93.19% with a precision and AUROC close to 0.9. This study presents a promising approach for automating brain tissue classification in MS patients. The ability to precisely distinguish between tissue types enables early detection and targeted treatment, which may improve outcomes for future MS patients by assisting clinicians with disease monitoring and treatment planning. This approach could support integration into future digital pathology platforms for MS diagnosis.