Deep learning assisted blunt liver injury diagnosis with ultrasound: an animal study

Blunt liver injury usually needs fast and accurate diagnosis for timely rescue, which can be achieved through ultrasound scanning. However, the recognition of liver injury from ultrasound images is highly dependent on subjective judgment, especially when there is no bleeding in the viscera. In this study, we explore animal models for validated standards of blunt liver injury, thereby evaluating the feasibility of the deep learning network in assessing indirect (hemoperitoneum) and direct (trauma focus) ultrasonographic traumatic signs for objective and reliable diagnosis. 54 healthy New Zealand rabbits were used to establish animal models with different levels of blunt liver injury. We collected ultrasound images before and after their modeling, then calculated the hemoperitoneum index, i.e., the sum of the maximum depth, to establish the corresponding anatomical standard. A deep-learning-based model, ResNet50, was then utilized to automatically detect the presence of hemoperitoneum or trauma focus and further grade the injury based on ultrasound images. There is a significant positive correlation between a hemoperitoneum index and the weight of the hemoperitoneum after animal modeling ( r = 0.788, p < 0.001). In addition, the difference in the hemoperitoneum index is positively correlated with the grade of blunt liver injury ( r = 0.413, p = 0.007). The accuracy, sensitivity, and specificity of the ResNet50 model were 0.842, 0.901, and 0.712 for determining the presence or absence of hemoperitoneum and 0.864, 0.852, and 0.881 for trauma focus. Furthermore, the model could discriminate the degree of liver injury with an accuracy of 0.732. This study demonstrates the feasibility of developing a validated standard for liver injury using animal models. The results also suggest the potential and feasibility of using a deep learning-based model to identify liver injury beyond the detection of hemoperitoneum from ultrasound images.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-17
DOI
https://doi.org/10.1038/s41598-026-79868-2
Primary Topic
Abdominal Trauma and Injuries
Type
article
Field-Weighted Citation Impact
0.00

Funders

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

Deep learning assisted blunt liver injury diagnosis with ultrasound: an animal study

Guangquan Zhou, Tuo Liu, Faqin Lv, Xuan Zhang et al.
1 links
Scientific Reports
Abdominal Trauma and Injuries
article

Deep learning assisted blunt liver injury diagnosis with ultrasound: an animal study

Guangquan Zhou, Tuo Liu, Faqin Lv, Xuan Zhang, MD Zhaoming Zhong, Yichen Zhu, Shihao Hua, Tianci Zhang
article en

Abstract

Blunt liver injury usually needs fast and accurate diagnosis for timely rescue, which can be achieved through ultrasound scanning. However, the recognition of liver injury from ultrasound images is highly dependent on subjective judgment, especially when there is no bleeding in the viscera. In this study, we explore animal models for validated standards of blunt liver injury, thereby evaluating the feasibility of the deep learning network in assessing indirect (hemoperitoneum) and direct (trauma focus) ultrasonographic traumatic signs for objective and reliable diagnosis. 54 healthy New Zealand rabbits were used to establish animal models with different levels of blunt liver injury. We collected ultrasound images before and after their modeling, then calculated the hemoperitoneum index, i.e., the sum of the maximum depth, to establish the corresponding anatomical standard. A deep-learning-based model, ResNet50, was then utilized to automatically detect the presence of hemoperitoneum or trauma focus and further grade the injury based on ultrasound images. There is a significant positive correlation between a hemoperitoneum index and the weight of the hemoperitoneum after animal modeling ( r = 0.788, p < 0.001). In addition, the difference in the hemoperitoneum index is positively correlated with the grade of blunt liver injury ( r = 0.413, p = 0.007). The accuracy, sensitivity, and specificity of the ResNet50 model were 0.842, 0.901, and 0.712 for determining the presence or absence of hemoperitoneum and 0.864, 0.852, and 0.881 for trauma focus. Furthermore, the model could discriminate the degree of liver injury with an accuracy of 0.732. This study demonstrates the feasibility of developing a validated standard for liver injury using animal models. The results also suggest the potential and feasibility of using a deep learning-based model to identify liver injury beyond the detection of hemoperitoneum from ultrasound images.

Scientific Reports
Chinese PLA General Hospital (CN), Southeast University (CN)
National Natural Science Foundation of China, Jiangsu Provincial Key Research and Development Program, National Key Research and Development Program of China
Good health and well-being
Abdominal Trauma and Injuries
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