Scholarlink

In this study, we propose a semantic segmentation framework based on convolutional neural networks (CNNs) aimed at segmenting teeth in 2D panoramic radiographs. To enhance the model’s performance, we adopt a data augmentation strategy based on the combination of Mosaic and multi-scale image scaling, which significantly enriches the training set samples. Additionally, we propose a post-processing strategy based on the model’s prediction probability map to address the issue of inaccurate points predicted by the model. The proposed framework is tested and evaluated on 2D panoramic radiographs, including 1000 unlabeled radiographs. The Dice coefficient, Intersection over Union (IoU), and \(1-H(d)\) achieve 0.9341, 0.9814, and 0.0244, respectively, resulting in a total score of 0.9607, ranking fifth on the leaderboard. This demonstrates the superiority of the proposed framework in tooth segmentation of 2D panoramic radiographs.

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

Convolutional Neural Network-Based Multi-scale Semantic Segmentation for Two-Dimensional Panoramic X-Rays of Teeth

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Qixuan Wang,
  • Yangzheng Zhao,
  • Zhuofan Zhang

摘要

In this study, we propose a semantic segmentation framework based on convolutional neural networks (CNNs) aimed at segmenting teeth in 2D panoramic radiographs. To enhance the model’s performance, we adopt a data augmentation strategy based on the combination of Mosaic and multi-scale image scaling, which significantly enriches the training set samples. Additionally, we propose a post-processing strategy based on the model’s prediction probability map to address the issue of inaccurate points predicted by the model. The proposed framework is tested and evaluated on 2D panoramic radiographs, including 1000 unlabeled radiographs. The Dice coefficient, Intersection over Union (IoU), and \(1-H(d)\) achieve 0.9341, 0.9814, and 0.0244, respectively, resulting in a total score of 0.9607, ranking fifth on the leaderboard. This demonstrates the superiority of the proposed framework in tooth segmentation of 2D panoramic radiographs.