Effectiveness of Deep Learning Classifiers in Histopathological Diagnosis of Oral Squamous Cell Carcinoma by Pathologists

Research Square (Research Square)(2023)

引用 0|浏览6
暂无评分
摘要
Abstract Objective: The study aims to identify valid histological classifiers from histopathological images of oral squamous cell carcinoma using convolutional neural network (CNN) deep learning models, and attempts to show how the learning results of the identified valid deep learning classifier models can be used as a reference to help oral pathologists improve their diagnostic performances. Methods: Histopathological samples of oral squamous cell carcinoma were prepared by an oral pathologist. Images were divided into tiles on a virtual slide, and labels (squamous cell carcinoma, normal, and others) were applied to the images containing cells. The CNNs used were VGG16 and ResNet50 with the optimizers SGD and SAM, both with and without a learning rate scheduler. The conditions for achieving good CNN performances were identified by examining the performance metrics. Furthermore, we used ROCAUC to statistically evaluate the improvement in the diagnostic performance of six oral pathologists by using the results obtained from the selected CNN model for assisted diagnosis. Results: Of all model combinations, VGG16 with SAM showed the highest performance. The performance metrics obtained for this optimal model were accuracy = 0.8622 and AUC = 0.9602. The diagnostic performances of the oral pathologists were significantly improved statistically when the diagnostic results of the best model were used as supplementary diagnoses (p-value = 0.031). Conclusions: It was found that by referring to the learning results of the best model classifier via deep learning, the diagnostic accuracy of the pathologists can be improved. This study contributes to the application of highly reliable deep learning models to the field of oral pathological diagnosis.
更多
查看译文
关键词
deep learning classifiers,deep learning,squamous cell carcinoma,histopathological diagnosis
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要