Scholarlink

For a proper diagnosis and treatment planning it is necessary to accurately analysis the dental X-ray images, in which interpretation by manual method is time-consuming and often inaccurate. In this work, will give a novel state-of-the-art automatic dental segmentation framework which integrates two well-known deep learning (DL) models Mask R-CNN and U-Net with an encoder of EfficientNet-B0. We applied Mask R-CNN for accurate instance segmentation and then U-Net to achieve pixel-wise fine-grained accuracy (describing the structures and lesions). Combining these two approaches, we developed a system that is more reliable and much effortless. Experiments show that this dual approach is superior to single model approaches and the performance consistently outperforms high precision as indicated by high Dice similarity coefficient as well as Intersection over Union (IoU) scores. Such findings open the possibility of using computerized deep learning techniques as a new standard in the practice of clinical dentistry.

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

An EfficientNet Encoder-Based Hybrid U-Net-Mask R-CNN Technique for Dental Panoramic Image Segmentation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • D. P. Dharmendra,
  • T. M. Rajesh,
  • B. Jayasudha

摘要

For a proper diagnosis and treatment planning it is necessary to accurately analysis the dental X-ray images, in which interpretation by manual method is time-consuming and often inaccurate. In this work, will give a novel state-of-the-art automatic dental segmentation framework which integrates two well-known deep learning (DL) models Mask R-CNN and U-Net with an encoder of EfficientNet-B0. We applied Mask R-CNN for accurate instance segmentation and then U-Net to achieve pixel-wise fine-grained accuracy (describing the structures and lesions). Combining these two approaches, we developed a system that is more reliable and much effortless. Experiments show that this dual approach is superior to single model approaches and the performance consistently outperforms high precision as indicated by high Dice similarity coefficient as well as Intersection over Union (IoU) scores. Such findings open the possibility of using computerized deep learning techniques as a new standard in the practice of clinical dentistry.