| 本文已被：浏览 1090次 下载 848次 |
码上扫一扫！ |
| 人工智能在鉴别肝门部胆管癌细胞及周围神经侵袭中的应用 |
| 顾小强1,俞文隆2,陈颖3,魏培莲1,董伟4,钱建新1,于观贞1* |
|
|
(1. 上海中医药大学附属龙华医院肿瘤科, 上海 200032;
2. 海军军医大学(第二军医大学)东方肝胆外科医院外科, 上海 200438;
3. 海军军医大学(第二军医大学)长海医院消化科, 上海 200433;
4. 海军军医大学(第二军医大学)东方肝胆外科医院病理科, 上海 200438
*通信作者) |
|
| 摘要: |
| 目的 建立一种用于辅助诊断肝门部胆管癌（HC）的人工智能（AI）算法模型，评价其识别肿瘤细胞及周围神经侵犯（PNI）的能力。方法 采用AI算法对825张HC和175张非癌变组织图像（600张为训练集，300张为测试集，100张为比较数据集）进行深度学习，将不同参数的GoogLeNet和DenseNet相结合的神经网络用于HC细胞和PNI的特征提取和深度学习。比较该AI算法模型与3名病理科医师（副主任医师、主治医师、住院医师各1名）在判断肿瘤有无及肿瘤细胞百分比的差异。结果 基于深度学习的AI算法可以准确识别HC组织标本图像中的肿瘤细胞及PNI。AI算法诊断肿瘤的能力可与经验丰富的病理科副主任医师媲美，且在评估肿瘤细胞百分比方面更胜一筹。结论 AI算法模型在识别HC肿瘤细胞及PNI方面具有辅助作用。 |
| 关键词: 人工智能 肝门部胆管癌 肿瘤间质比 神经侵犯 |
| DOI：10.16781/j.0258-879x.2021.07.0724 |
| 投稿时间：2021-03-03修订日期：2021-06-28 |
| 基金项目:上海中医药大学附属龙华医院高层次人才引进科研启动经费（LH02.51.002）. |
|
| Application of artificial intelligence in identifying hilar cholangiocarcinoma and perineural invasion |
| GU Xiao-qiang1,YU Wen-long2,CHEN Ying3,WEI Pei-lian1,DONG Wei4,QIAN Jian-xin1,YU Guan-zhen1* |
(1. Department of Oncology, Longhua Hospital Affiliated to Shanghai University of Traditional Chinese Medicine, Shanghai 200032, China;
2. Department of Surgery, Eastern Hepatobiliary Surgery Hospital, Naval Medical University (Second Military Medical University), Shanghai 200438, China;
3. Department of Gastroenterology, Changhai Hospital, Naval Medical University (Second Military Medical University), Shanghai 200433, China;
4. Department of Pathology, Eastern Hepatobiliary Surgery Hospital, Naval Medical University (Second Military Medical University), Shanghai 200438, China
*Corresponding author) |
| Abstract: |
| Objective To establish an artificial intelligence (AI) algorithm model for the diagnosis of hilar cholangiocarcinoma (HC) and to evaluate its ability to recognize tumor cells and perineural invasion (PNI). Methods AI algorithm was used for deep learning on 825 HC and 175 non-cancerous tissue images (600 for training set, 300 for test set, and 100 for comparison set). The neural network combined with GoogLeNet and DenseNet with different parameters was used for feature extraction and deep learning of HC cells and PNI. The AI algorithm model was compared among 3 pathologists (1 associate chief physician, 1 attending physician and 1 resident) in judging the presence or absence of tumor and the proportion of tumor cells. Results The AI algorithm based on deep learning could accurately identify tumor cells and PNI in HC tissue sample images. The AI algorithm was as good as experienced associate chief physicians in the department of pathology in diagnosing tumors, and was better in estimating the proportion of tumor cells. Conclusion AI algorithm model is helpful in identifying HC cells and PNI. |
| Key words: artificial intelligence hilar cholangiocarcinoma tumor-stroma ratio neural invasion |
