Scholarlink

<p>Dental disease classification (DDC) from panoramic radiographs is complicated due to various factors such as the variability of the images, class imbalance, and many diseases occurring in the same image. This research proposes a deep learning method that combines multi-label classification with dataset integration to help overcome these problems. Two merged datasets were utilized, the panoramic dental imaging and analysis dataset (PDIAD), to test six pre-trained CNNs (DenseNet-121/201, ResNet V2-50/101, EfficientNet-B0, and VGG19) for the identification of four dental problems: missing teeth, impaction, cavities, and fillings. Measures of performance including (accuracy, F1-score, precision, recall, and training time) were assessed. The experiments suggest that CNNs can successfully detect these problems, where DenseNet-121 obtsined (81.75% accuracy and 79% F1-score) on panoramic radiography and vgg19 led the results on PDIAD (85.75% accuracy, 81.5% F1-score). EfficientNet-B0 was the fastest training model, while DenseNet-201 was the slowest. This research has enhanced the generalization and robustness of real-world applications to a great extent by utilizing multiple datasets and labels as opposed to previous single-dataset or single-label approaches. These results emphasize the capability of deep learning in DDC and the significance of dataset merging and multi-label modeling.</p>

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

Automated Diagnosis of Dental Diseases Using Deep Learning on Radiographic Images

  • 期刊论文

  • 发表日期：

    2025年8月18日

查看全文
  • Rawda Tarek,
  • Ayman Elshenawy,
  • Mohammad Ismail Assadwy,
  • Mohamed Ashraf Madkour

摘要

Dental disease classification (DDC) from panoramic radiographs is complicated due to various factors such as the variability of the images, class imbalance, and many diseases occurring in the same image. This research proposes a deep learning method that combines multi-label classification with dataset integration to help overcome these problems. Two merged datasets were utilized, the panoramic dental imaging and analysis dataset (PDIAD), to test six pre-trained CNNs (DenseNet-121/201, ResNet V2-50/101, EfficientNet-B0, and VGG19) for the identification of four dental problems: missing teeth, impaction, cavities, and fillings. Measures of performance including (accuracy, F1-score, precision, recall, and training time) were assessed. The experiments suggest that CNNs can successfully detect these problems, where DenseNet-121 obtsined (81.75% accuracy and 79% F1-score) on panoramic radiography and vgg19 led the results on PDIAD (85.75% accuracy, 81.5% F1-score). EfficientNet-B0 was the fastest training model, while DenseNet-201 was the slowest. This research has enhanced the generalization and robustness of real-world applications to a great extent by utilizing multiple datasets and labels as opposed to previous single-dataset or single-label approaches. These results emphasize the capability of deep learning in DDC and the significance of dataset merging and multi-label modeling.