Scholarlink

Accurate segmentation of the masseter muscles (MM) is critical for the assessment of various clinical conditions. While artificial intelligence (AI)-based methods have demonstrated promising results, existing studies suffer from limitations such as small datasets and constrained model deployment bandwidth, hindering cross-study comparisons. To address these challenges, we curated a comprehensive dataset comprising over 1,000 annotated images from four distinct sources. We trained and evaluated a diverse set of architectures including the well-applied convolutional neural network (CNN)-based models as well as the recently proposed Vision Transformer and Vision Mamba based models. A standardized evaluation pipeline was developed to assess various performance metrics across models. Our results demonstrate that the YOLOv8 + SAM model exhibits superior generalization capabilities compared to others.

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

A Novel Dataset and Comprehensive AI Model Evaluation for Masseter Muscle Segmentation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yunfeng Liang,
  • Lin Zou,
  • Millie Ming Rong Goh,
  • Si Xian Ho,
  • Zhen Jie Zhang,
  • Barbara Helen Rosario,
  • Bharathi Balasundaram,
  • Sindhu John,
  • Saraswathy Suresh Babu,
  • Charlene Jin Yee Liew,
  • Aileen Lim,
  • Andy Wee An Ta,
  • Han Leong Goh

摘要

Accurate segmentation of the masseter muscles (MM) is critical for the assessment of various clinical conditions. While artificial intelligence (AI)-based methods have demonstrated promising results, existing studies suffer from limitations such as small datasets and constrained model deployment bandwidth, hindering cross-study comparisons. To address these challenges, we curated a comprehensive dataset comprising over 1,000 annotated images from four distinct sources. We trained and evaluated a diverse set of architectures including the well-applied convolutional neural network (CNN)-based models as well as the recently proposed Vision Transformer and Vision Mamba based models. A standardized evaluation pipeline was developed to assess various performance metrics across models. Our results demonstrate that the YOLOv8 + SAM model exhibits superior generalization capabilities compared to others.