A Deep-Learning-Derived Ultrasound Framework for Hepatic Steatosis Detection and Integrated Fibrosis Assessment: A Multicenter Study

Background/Objectives: We aimed to develop an integrated B-mode ultrasound framework for hepatic steatosis detection and fibrosis assessment in patients with suspected metabolic dysfunction-associated steatotic liver disease (MASLD). Methods: This multicenter study prospectively included 734 participants with 2485 ultrasound images from eight centers (April 2023–November 2025). A Global Attention Mechanism-enhanced Cross-Attention Multi-Scale Vision Transformer was used to derive a steatosis signature in Dataset A (Centers 1–3, n = 245), with external testing in Datasets B (Centers 4–5, n = 294) and C (Centers 6–8, n = 195). The signature was then integrated with liver elasticity and clinical and laboratory variables to develop a fibrosis prediction model in Dataset B, with external testing in Dataset C. Results: The steatosis signature achieved AUROCs of 0.887 (95% CI, 0.831–0.938) and 0.968 (95% CI, 0.932–0.994) in Datasets B and C, respectively, significantly outperforming the hepatic steatosis index, fatty liver index, and the junior radiologist (AUROC range: 0.728–0.794; all p < 0.05), while performing comparably to the senior radiologist (AUROC range: 0.864–0.932). The fibrosis model achieved an AUROC of 0.849 (95% CI, 0.762–0.916) in Dataset C and significantly outperformed elasticity, fibrosis-4 index, aspartate aminotransferase-to-platelet ratio index, and nonalcoholic fatty liver disease fibrosis score (AUROC range: 0.431–0.697; all p < 0.05). For F0 versus F1, the model also achieved an AUROC of 0.841 (95% CI, 0.745–0.922), significantly outperforming all comparators (all p < 0.05). Conclusions: This integrated framework provides a noninvasive approach for steatosis case-finding and early fibrosis risk stratification in patients with suspected MASLD.

Authors

Institutions

Publication Details

Journal
Biomedicines
Published
2026-11-17
DOI
https://doi.org/10.3390/biomedicines14112656
Primary Topic
Liver Disease Diagnosis and Treatment
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Deep-Learning-Derived Ultrasound Framework for Hepatic Steatosis Detection and Integrated Fibrosis Assessment: A Multicenter Study

Liyun Xue, Zunguo Du, Wentao Kong, Hong Ding et al.
8 links
Biomedicines
Liver Disease Diagnosis and Treatment
article

A Deep-Learning-Derived Ultrasound Framework for Hepatic Steatosis Detection and Integrated Fibrosis Assessment: A Multicenter Study

Liyun Xue, Zunguo Du, Wentao Kong, Hong Ding, Xin Li, Jing Liang, Dong Jiang, Han Liu, Yuli Zhu
article en

Abstract

Background/Objectives: We aimed to develop an integrated B-mode ultrasound framework for hepatic steatosis detection and fibrosis assessment in patients with suspected metabolic dysfunction-associated steatotic liver disease (MASLD). Methods: This multicenter study prospectively included 734 participants with 2485 ultrasound images from eight centers (April 2023–November 2025). A Global Attention Mechanism-enhanced Cross-Attention Multi-Scale Vision Transformer was used to derive a steatosis signature in Dataset A (Centers 1–3, n = 245), with external testing in Datasets B (Centers 4–5, n = 294) and C (Centers 6–8, n = 195). The signature was then integrated with liver elasticity and clinical and laboratory variables to develop a fibrosis prediction model in Dataset B, with external testing in Dataset C. Results: The steatosis signature achieved AUROCs of 0.887 (95% CI, 0.831–0.938) and 0.968 (95% CI, 0.932–0.994) in Datasets B and C, respectively, significantly outperforming the hepatic steatosis index, fatty liver index, and the junior radiologist (AUROC range: 0.728–0.794; all p < 0.05), while performing comparably to the senior radiologist (AUROC range: 0.864–0.932). The fibrosis model achieved an AUROC of 0.849 (95% CI, 0.762–0.916) in Dataset C and significantly outperformed elasticity, fibrosis-4 index, aspartate aminotransferase-to-platelet ratio index, and nonalcoholic fatty liver disease fibrosis score (AUROC range: 0.431–0.697; all p < 0.05). For F0 versus F1, the model also achieved an AUROC of 0.841 (95% CI, 0.745–0.922), significantly outperforming all comparators (all p < 0.05). Conclusions: This integrated framework provides a noninvasive approach for steatosis case-finding and early fibrosis risk stratification in patients with suspected MASLD.

BiomedicinesVol. 14(11)
Jiangsu University (CN), Second Military Medical University (CN), Fudan University (CN), Affiliated Hospital of Jiangsu University (CN), Changhai Hospital (CN), Zhongshan Hospital (CN), Eastern Hepatobiliary Surgery Hospital (CN), Huashan Hospital (CN)
National Natural Science Foundation of China, Huashan Hospital
Good health and well-being
Liver Disease Diagnosis and Treatment
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon