应用计算机辅助诊断技术对口腔鳞状细胞癌组织病理图像进行自动检测的价值

The value of automatic detection of pathological images of oral squamous cell carcinoma by computer aided diagnosis

目的:评估计算机辅助诊断技术对口腔鳞状细胞癌组织病理图像进行自动检测的准确性及其临床应用价值。方法:来自B Borooah癌症研究所的医学专家,从230例患者中收集、准备和分类的1224张口腔组织病理图片被纳入研究。本研究采用十折交叉验证对图像样本进行训练和测试,验证本研究模型的有效性。此外,本研究采用经典的ResNet50模型作为深度学习的框架,并根据切片图像的性质进行了改进,以确保自动检测的效果。结果:分类实验的结果表明,本研究所提出的深度学习模型可以快速、精确的对口腔鳞状细胞癌组织病理图像进行检测,受试者工作特征曲线(receiver operating characteristic curve,ROC)和曲线下面积(area under the curve,AUC)(最优AUC=0.91,平均AUC=0.88)显示了该方法的实验效果。此外,模型的准确率(accuracy,ACC)(0.976)、敏感性(sensitivity,SEN)(0.981)以及特异性(specificity,SPE)(0.971)也进一步显示了该研究的效果。结论:本研究所提出的深度学习框架可以很好的对口腔鳞状细胞癌进行自动检测,所得到的结果可以有效地转化为软件,对于临床辅助诊断使用有极大的帮助。

Objective:To evaluate the accuracy and clinical application value of computer aided diagnosis in automatic detection of oral squamous cell carcinoma pathological images.Methods:Medical experts from the B Borooah Cancer Institute,1224 oral pictures collected,prepared and classified from 230 patients were included in the study.This study uses 10-fold cross-validation to train and test the image data to verify the effectiveness of the research model.In addition,this research uses the classic ResNet50 model as the deep learning framework and improves it according to the properties of sliced images to ensure the effect of automatic detection.Results:The results of the classification experiment show that the deep learning model proposed in this research can quickly and accurately detect oral squamous cell carcinoma,the receiver operating characteristic curve(ROC)and the area under the line(AUC)(optimum AUC=0.91,average AUC=0.88)shows the experimental effect of this method.In addition,the accuracy(ACC=0.976),sensitivity(SEN=0.981),and specificity(SPE=0.971)of the model further show the effect of the study.Conclusion:The deep learning framework proposed in this research can automatically detect oral squamous cell carcinoma.The results obtained can be effectively transformed into software for clinical use as an artificial intelligence-assisted diagnosis tool.