Division of Oral and Maxillofacial Surgery, Faculty of Dentistry, Thammasat University, Pathum Thani, Thailand.
College of Interdisciplinary Studies, Thammasat University, Patum Thani, Thailand.
J Oral Pathol Med. 2021 Oct;50(9):911-918. doi: 10.1111/jop.13227. Epub 2021 Aug 16.
Oral cancer is a deadly disease among the most common malignant tumors worldwide, and it has become an increasingly important public health problem in developing and low-to-middle income countries. This study aims to use the convolutional neural network (CNN) deep learning algorithms to develop an automated classification and detection model for oral cancer screening.
The study included 700 clinical oral photographs, collected retrospectively from the oral and maxillofacial center, which were divided into 350 images of oral squamous cell carcinoma and 350 images of normal oral mucosa. The classification and detection models were created by using DenseNet121 and faster R-CNN, respectively. Four hundred and ninety images were randomly selected as training data. In addition, 70 and 140 images were assigned as validating and testing data, respectively.
The classification accuracy of DenseNet121 model achieved a precision of 99%, a recall of 100%, an F1 score of 99%, a sensitivity of 98.75%, a specificity of 100%, and an area under the receiver operating characteristic curve of 99%. The detection accuracy of a faster R-CNN model achieved a precision of 76.67%, a recall of 82.14%, an F1 score of 79.31%, and an area under the precision-recall curve of 0.79.
The DenseNet121 and faster R-CNN algorithm were proved to offer the acceptable potential for classification and detection of cancerous lesions in oral photographic images.
口腔癌是全球最常见恶性肿瘤之一，是一种致命疾病，在发展中国家和中低收入国家已成为日益严重的公共卫生问题。本研究旨在使用卷积神经网络（CNN）深度学习算法开发用于口腔癌筛查的自动分类和检测模型。
该研究纳入了 700 例回顾性采集的口腔颌面中心临床口腔照片，分为 350 例口腔鳞状细胞癌图像和 350 例正常口腔黏膜图像。分别使用 DenseNet121 和 faster R-CNN 构建分类和检测模型。将 490 张图像随机选为训练数据，另外 70 张和 140 张图像分别选为验证和测试数据。
DenseNet121 模型的分类准确率达到了 99%，灵敏度为 98.75%，特异度为 100%，召回率为 100%，F1 评分为 99%，受试者工作特征曲线下面积为 99%。faster R-CNN 模型的检测准确率达到了 76.67%，灵敏度为 82.14%，特异度为 100%，召回率为 79.31%，精确率-召回率曲线下面积为 0.79。
DenseNet121 和 faster R-CNN 算法在口腔摄影图像中对癌性病变的分类和检测具有一定的应用潜力。
