【摘要】 目的:探究深度学习网络模型(ResNet-18)用于判定口腔鳞状细胞癌(oral squamous cell carcinoma, OSCC)最差浸润方式(worst pattern of invasion, WPOI)的可行性及效果。方法:应用ResNet-18模型对收集的491张OSCC患者数字化病理切片进行研究，训练其识别并区分非肿瘤区域、WPOI 1～3级、WPOI 4～5级，利用分类准确率对模型进行评估。结果:ResNet-18神经网络可以有效区分OSCC非肿瘤区域、WPOI 1～3级、WPOI 4～5级，其准确率可达99.5%。结论:深度学习网络模型ResNet-18可以有效区分OSCC非肿瘤区域、WPOI 1～3级、WPOI 4～5级，辅助医师提高诊断速度。更多还原

【Abstract】 Objective: To explore the feasibility and effect of the deep learning network model(ResNet-18) to determine the worst infiltration mode(worst pattern of invasion, WPOI) of oral squamous cell carcinoma(oral squamous cell carcinoma, OSCC). Methods: The 491 digital pathological sections collected by ResNet-18 model were trained to identify and distinguish non-tumor areas, WPOI 1-3 and WPOI 4-5, and the model was evaluated using the classification accuracy. Results: ResNet-18 neural network can effectively distinguish non-tumor areas of OSCC, WPOI 1-3 and WPOI 4-5, with an accuracy of 99.5%. Conclusion: The deep learning network model ResNet-18 can effectively distinguish the non-tumor areas of OSCC, WPOI 1-3, and WPOI 4-5, and assist physicians to improve the diagnosis speed.更多还原