Scholarlink

Diagnosing multiple sclerosis (MS) accurately is highly challenging due to symptom overlap with other demyelinating diseases. Here, we present DemyeliNeXt, an explainable few-shot learning framework designed to classify MS and other demyelinating diseases from MRI scans. This framework employs a prototypical network with a 3D DenseNet-121 backbone and uses Deep SHAP for feature importance visualization. We train our DemyeliNeXt on a dataset from African populations and we test it for different datasets including MICCAI MSSEG2 public dataset. Our findings demonstrate robust performance across diverse datasets highlighting the model’s potential to enhance diagnosis accuracy and generalizability in various clinical settings.

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

Explainable Few-Shot Learning for Multiple Sclerosis Detection in Low-Data Regime

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Montassar Ben Dhifallah,
  • Dalel Kanzari,
  • Salma Naija,
  • Sana Ben Amor,
  • Ahmed Zrig,
  • Mezri Maatouk,
  • Mabrouk Abdelaali,
  • Jamel Saad,
  • Asma Achour,
  • Sofiane Gaied Chortane,
  • Maher Hadhri,
  • Ahmed Dahmoul,
  • Azza Ben Ali,
  • Sahar Selim,
  • Ahmed Nebli

摘要

Diagnosing multiple sclerosis (MS) accurately is highly challenging due to symptom overlap with other demyelinating diseases. Here, we present DemyeliNeXt, an explainable few-shot learning framework designed to classify MS and other demyelinating diseases from MRI scans. This framework employs a prototypical network with a 3D DenseNet-121 backbone and uses Deep SHAP for feature importance visualization. We train our DemyeliNeXt on a dataset from African populations and we test it for different datasets including MICCAI MSSEG2 public dataset. Our findings demonstrate robust performance across diverse datasets highlighting the model’s potential to enhance diagnosis accuracy and generalizability in various clinical settings.