Scholarlink

Metacarpophalangeal fractures, being common injuries of the hand, are subtle and may not be diagnosed on routine radiographs leading to misdiagnosis and reduced function owing to variable image quality. In this work, the study introduced MetPhaFrac (Metacarpal Phalanges Fracture), an ensemble model that fuses Mask Region-based Convolutional Neural Network (R-CNN) with Inception v3 as backbone, which allows to detect and instance segment metacarpal phalanges fractures accurately and worked with a primary dataset of 2,245 anonymised X-ray images (1,450 fractured and 795 non-fractured). The model was trained on an NVIDIA A100 GPU using PyTorch, AdamW was used as an optimiser. The model achieved 93.3% mean average precision (mAP) @0. 5 with 94.7% precision, 92.2% recall and 91.1% F1-score that surpassing other benchmark models. This artificial intelligence (AI) approach, enhances diagnostic precision with pixel-level segmentation, with future work targeting real-time clinical integration.

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

MetPhaFrac: Detection of Metacarpophalangeal Fractures Using Mask R-CNN Inception v3: An AI-Driven Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Takhellambam Sylvia,
  • Sanjay Kumar Dubey,
  • Khelchandra Thongam,
  • Guruaribam Rishikanta Sharma

摘要

Metacarpophalangeal fractures, being common injuries of the hand, are subtle and may not be diagnosed on routine radiographs leading to misdiagnosis and reduced function owing to variable image quality. In this work, the study introduced MetPhaFrac (Metacarpal Phalanges Fracture), an ensemble model that fuses Mask Region-based Convolutional Neural Network (R-CNN) with Inception v3 as backbone, which allows to detect and instance segment metacarpal phalanges fractures accurately and worked with a primary dataset of 2,245 anonymised X-ray images (1,450 fractured and 795 non-fractured). The model was trained on an NVIDIA A100 GPU using PyTorch, AdamW was used as an optimiser. The model achieved 93.3% mean average precision (mAP) @0. 5 with 94.7% precision, 92.2% recall and 91.1% F1-score that surpassing other benchmark models. This artificial intelligence (AI) approach, enhances diagnostic precision with pixel-level segmentation, with future work targeting real-time clinical integration.