Deep learning model based on grayscale ultrasound for predicting asymptomatic compensated advanced chronic liver disease
10.13929/j.issn.1003-3289.2025.06.021
- VernacularTitle:灰阶超声深度学习模型预测无症状代偿性晚期慢性肝病
- Author:
Sisi HUANG
1
;
Yingzi LIANG
1
;
Fangyi HUANG
1
;
Liyan WEI
1
;
Yuanyuan CHEN
1
;
Yong GAO
1
Author Information
1. 广西医科大学第一附属医院超声科,广西南宁 530021
- Publication Type:Journal Article
- Keywords:
liver diseases;
ultrasonography;
deep learning
- From:
Chinese Journal of Medical Imaging Technology
2025;41(6):947-951
- CountryChina
- Language:Chinese
-
Abstract:
Objective To explore the value of deep learning(DL)model based on grayscale ultrasound for predicting asymptomatic advanced chronic liver disease(cACLD).Methods Totally 258 patients with asymptomatic compensatory chronic liver diseases were retrospectively included,among them 117 with F3 or F4 stage liver fibrosis were classified into cACLD group,while 141 with F1 or F2 stage liver fibrosis were taken as non-cACLD group.The patients were divided into training set(n=180,including 82 cases of cACLD and 98 cases of non-cACLD)and validation set(n=78,including 35 cases of cACLD and 43 cases of non-cACLD)at the ratio of 7∶3.Univariate and multivariate logistic regression were used to screen independent clinical predictors of cACLD and construct a clinical model.Based on liver grayscale ultrasound,optimal DL features were extracted and screened,and Resnet50 network was adopted as framework,na?ve Bayes classifier was used to construct DL model,and a combined model was constructed based on clinical model and DL model.The efficacy and clinical value of each model for predicting asymptomatic cACLD were evaluated.Results Age,gamma-glutamyl transferase and platelet count were all independent clinical predictors of cACLD,and a clinical model was constructed.Totally 38 optimal DL features were screened to build a DL model.The AUC of combined model in training set and validation set was 0.950 and 0.740,of DL model was 0.944 and 0.737,respectively,being not significantly different(both P＞0.05)but all higher than that of clinical model(0.667 and 0.573,all P＜0.05).Taken 0.59-0.90 as the threshold,the net benefits of combined model in both training and validation sets were higher than that of other models.Conclusion DL model based on grayscale ultrasound could be used to effectively predict asymptomatic cACLD.Combining with clinical characteristics might improve clinical net benefit of this model.
