Scholarlink

Common manual diagnostic procedures in the dental industry are time-consuming and subjective. Although current deep learning algorithms such as ResNet-50 and ResUNet have potential, they are still facing noise sensitivity, imprecision of irregular lesion boundary segmentation and poor generalization. Our solution to the limitations is an Attention-Guided Multi-Stage Hybrid Model (AMSHM). There are three major innovations launched through our pipeline. A Dual-Stage Denoising Module combines Layer Division Non-Zero Elimination with an adaptive soft-switching median filter to useful noise reduction. A Hybrid Attention ResNet is used to classify images and a Cascaded Attention ResUNet is used to classify images with more specific image segmentation through the use of nested skip connections and attention mechanisms. When tested on a larger dataset of 20,000 diverse radiographs the proposed model attained state-of-the-art performance with classification accuracy of 91.2% and segmentation accuracy of 98.7% exceeding that of the base model. The method greatly improves the strength of precision and clinical relevance of AI-based dental caries 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

A Multi-stage Attention-Guided Hybrid Model for Dental Caries Detection and Segmentation in Radiographic Imaging

  • 章节

  • 发表日期：

    2026年

查看全文
  • Hanuman Maurya,
  • Radhey Shyam,
  • Natthan Singh,
  • Nagendra Pratap Singh,
  • Parvesh Saini

摘要

Common manual diagnostic procedures in the dental industry are time-consuming and subjective. Although current deep learning algorithms such as ResNet-50 and ResUNet have potential, they are still facing noise sensitivity, imprecision of irregular lesion boundary segmentation and poor generalization. Our solution to the limitations is an Attention-Guided Multi-Stage Hybrid Model (AMSHM). There are three major innovations launched through our pipeline. A Dual-Stage Denoising Module combines Layer Division Non-Zero Elimination with an adaptive soft-switching median filter to useful noise reduction. A Hybrid Attention ResNet is used to classify images and a Cascaded Attention ResUNet is used to classify images with more specific image segmentation through the use of nested skip connections and attention mechanisms. When tested on a larger dataset of 20,000 diverse radiographs the proposed model attained state-of-the-art performance with classification accuracy of 91.2% and segmentation accuracy of 98.7% exceeding that of the base model. The method greatly improves the strength of precision and clinical relevance of AI-based dental caries diagnosis.