基于深度迁移学习的牙源性囊肿病理图像鉴别诊断方法研究  

Deep transfer learning-based method for differential diagnosis of pathology images for odontogenic cysts

作　　者：方嘉琨 张建运 FANG Jia-kun;ZHANG Jian-yun(Peking University School and Hospital ofStomatology,National Center of Stomatology,National Clinical Research Center for Oral Diseases,National Engineering Laboratory for Digital and Material Technology of Stomatology,Beijing 100081,China)

机构地区：[1]北京大学口腔医学院(口腔医院),国家口腔医学中心,国家口腔疾病临床医学研究中心,口腔数字化医疗技术和材料国家工程实验室,北京100081

出　　处：《医疗卫生装备》2022年第6期14-17,共4页Chinese Medical Equipment Journal

基　　金：北京大学口腔医学院教育教学研究项目(YS030120)。

摘　　要：目的:基于深度迁移学习实现牙源性囊肿病理图像自动化鉴别诊断,为病理医师提供参考。方法:收集诊断为牙源性角化囊肿和正角化牙源性囊肿的数字病理图像各50张,应用自动化分割方法获取上皮组织图像,将分割后的图像切割为小图像块作为训练集、验证集和独立测试集,其中训练集和验证集用于建立模型,测试集用于评价模型预测能力。使用深度迁移学习的方法基于VGG16网络建立诊断模型,并对该模型鉴别诊断的效果进行评估。结果:基于深度迁移学习的诊断模型在测试集上准确率达到96.96%,模型可以较好地区分牙源性角化囊肿和正角化牙源性囊肿;将同一图像的小图像块的分类概率取均值作为该图像的分类概率,模型的诊断准确率为100%。结论:在较少的样本量下,基于深度迁移学习建立的诊断模型应用于牙源性囊肿病理图像鉴别准确率较高,可以作为提高病理医师诊断效率的辅助工具。Objective To realize auto differential diagnosis of pathology images for odontogenic cysts based on deep transfer learning to provide references for pathologists. Methods Totally 50 digital pathology images for odontogenic keratocyst(OKC)and another 50 ones for orthokeratinized odontogenic cyst(OOC) were collected, and epithelial tissue images were obtained by automated segmentation method. The segmented images were cut into small image patches as the training dataset, validation dataset and independent test dataset, with the training and validation datasets to build the model and the independent test dataset to evaluate the model prediction ability. A diagnostic model based on VGG16 network using a deep transfer learning approach was established and the effectiveness of the model for differential diagnosis was evaluated. Results The diagnostic model based on deep transfer learning achieved 96.96% accuracy at the test set, and the model could distinguish between OKC and OOC effectively;the classification probability of some image was determined by calculating the mean value of those of the small image patches from the image, and the diagnostic accuracy of the model reached 100%. Conclusion The deep transfer learning-based diagnostic model gains high accuracy when used for the differential diagnosis of pathology images of odontogenic cysts, and can be an auxiliary tool to enhance the diagnostic efficiency of pathologists. [Chinese Medical Equipment Journal,2022,43(6):14-17]

关 键 词：牙源性囊肿 正角化牙源性囊肿 牙源性角化囊肿 迁移学习 深度学习 深度迁移学习 病理图像 

分 类 号：R318[医药卫生—生物医学工程]