| 李时杰,宋镒凯,张智弘,梁 辉,周红文,龚颖芸,陈丰荣.基于深度学习的肝脏病理图像中肝脂肪变性分级研究[J].,2022,(19):3601-3607 |
| 基于深度学习的肝脏病理图像中肝脂肪变性分级研究 |
| Deep Learning for Liver Steatosis Diagnosis of Pathological Sections |
| 投稿时间：2022-03-28 修订日期：2022-04-24 |
| DOI：10.13241/j.cnki.pmb.2022.19.001 |
| 中文关键词: 深度学习 多示例学习 非酒精性脂肪肝诊断 数字病理 |
| 英文关键词: Deep learning Multiple instance learning Non-alcoholic fatty liver diagnosis Digital pathology |
| 基金项目:国家自然科学基金委重大研究计划培育项目（91857103） |
|
| 摘要点击次数: 1516 |
| 全文下载次数: 658 |
| 中文摘要: |
| 摘要 目的：设计基于深层神经网络模型用来分析肝脏全景病理切片图像（Whole slide images, WSI）的肝脂肪变性分级方法，以实现对非酒精性脂肪性肝病（Non-alcoholic fatty liver disease, NAFLD）病程的辅助诊断。方法：结合临床诊断，以非酒精性脂肪肝活动度积分（NAFLD activity score, NAS）为评价标准，将肝脂肪变性程度分为无、轻度、中度和重度等四级病程，本研究采用多示例学习的策略构建并训练深度神经网络模型，将训练获得的人工智能模型用来实现计算机自动化诊断肝脏病理切片中肝脂肪变性程度分级。结果：通过使用本研究中的人工智能方法可以在3分钟内对一张WSI进行完整的分析，得到该病患肝脏病理切片中肝脂肪变性分级，训练获得的人工智能模型的AUC为0.97，肝脂肪变性分级的平均准确率为78.18%，macro-F1 score、macro-Precision和macro-Recall分别为79.49、82.03和77.10，其结果展示获得的人工智能模型已满足可辅助临床诊断的水平。结论：本研究基于深度学习技术开发的人工智能方法初步实现快速自动化诊断肝脂肪变性分级，展现了其潜在的临床使用价值。 |
| 英文摘要: |
| ABSTRACT Objective: A classification method of hepatic steatosis based on a deep neural network model is designed to analyze the whole slide images (WSIs) of the liver, so as to realize the auxiliary diagnosis for the course of non-alcoholic fatty liver disease (NAFLD). Methods: Combined with clinical diagnosis, taking NAFLD activity score (NAS) as the evaluation standard, the degree of hepatic steatosis was divided into four stages: none, mild, moderate, and severe. In this study, a multiple instance learning method was used to construct and train a deep neural network model. The trained artificial intelligence (AI) model was used to realize the computer automatic diagnosis of the degree of hepatic steatosis in liver pathological sections. Results: By using the AI method in this study, a WSI could be completely analyzed within 3 minutes to obtain the degree of liver steatosis in a patient's liver pathological section. The AUC of the trained AI model was 0.97, the average accuracy of liver steatosis degree was 78.18%, and the macro-F1 score, macro-Precision, and macro-Recall were 79.49, 82.03, and 77.10, respectively. These results demonstrated that the trained AI model has satisfied with the level of auxiliary clinical diagnosis. Conclusion: The AI method based on deep learning technology developed in this study preliminarily realizes the rapid and automatic diagnosis of the classification of liver steatosis, which shows its potential clinical value. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
