浙江医学2025，Vol.47Issue(13)：1378-1383,后插2,7.DOI:10.12056/j.issn.1006-2785.2025.47.13.2025-252
基于增强磁共振的深度学习模型对进展期肝纤维化的诊断效能
Diagnostic efficacy of enhanced magnetic resonance imaging-based deep learning models on progressive liver fibrosis
黄杰 1陈勇 1赵文静 1林俊 1周静 1何承海 2姚婉贞 1周建文 3杨凯文4
作者信息
- 1. 310015 杭州师范大学附属医院放射科
- 2. 310015 杭州师范大学附属医院消化科
- 3. 310015 杭州师范大学附属医院病理科
- 4. 贵州中医药大学信息工程学院
- 折叠
摘要
Abstract
Objective To explore the diagnostic efficacy of a deep learning model constructed based on the automatic semantic segmentation results of enhanced MRI for the diagnosis of progressive liver fibrosis.Methods A total of 217 patients with chronic liver disease admitted to the Affiliated Hospital of Hangzhou Normal University from January 2014 to December 2024 were retrospectively selected and divided into 175 cases in the training set and 42 cases in the test set according to the ratio of 8∶2.An automatic semantic segmentation model was constructed to automatically segment the liver,and a deep learning model was built based on the automatic segmentation results of enhanced MRI.The similarity between automated semantic segmentation models and radiologists' manual segmentation results were compared,so was the diagnostic efficacy of deep learning models with benchmark models for patients in training and test sets.The number of incorrect diagnosis cases,accuracy,and diagnostic efficacy in a test set were compared between deep learning models and APRI and Fib-4.Results The Dice similarity coefficient between the automatic semantic segmentation model and the manual segmentation result of the radiologist was 0.973 in the training set,while 0.870 in the test set.In the training set,the AUC value(95％CI)was 0.999(0.941-0.998)in the deep learning model,while 0.995(0.928-0.993)in the baseline model,and the difference between the two models was not statistically significant(P=0.606).In the test set,the AUC(95％CI)was 0.959(0.774-1.000)in the deep learning model,which was higher than that of the baseline model,0.852(0.650-0.957),and the difference between the two models was statistically significant(P=0.031).The deep learning model incorrectly diagnosed 3 cases of progressive liver fibrosis with an accuracy of 0.929.APRI incorrectly diagnosed 7 cases of progressive liver fibrosis with an accuracy of 0.833.Fib-4 incorrectly diagnosed 3 cases each of progressive and non-progressive liver fibrosis with an accuracy of 0.857.The AUC(95％CI)of the deep learning model was higher than that of APRI of 0.750(0.583-0.917),and of Fib-4 of 0.839(0.698-0.981).The difference in diagnostic efficacy of the deep learning model compared with APRI and Fib-4(P=0.003)was statistically significant(both P＜0.001).Conclusion Deep learning models constructed based on the results of automated semantic segmentation of enhanced MRI are expected to be a noninvasive diagnostic solution for progressive liver fibrosis.关键词
深度学习/肝纤维化/进展期/MRIKey words
Deep learning/Liver fibrosis/Advanced stage/MRI引用本文复制引用
黄杰,陈勇,赵文静,林俊,周静,何承海,姚婉贞,周建文,杨凯文..基于增强磁共振的深度学习模型对进展期肝纤维化的诊断效能[J].浙江医学,2025,47(13):1378-1383,后插2,7.
