Researchers at Osaka Metropolitan University have developed an AI model that can detect fatty liver disease from routine chest X-rays with over 80 % accuracy, offering a low-cost, widely accessible screening method.
Study: Performance of a Chest Radiograph–based Deep Learning Model for Detecting Hepatic Steatosis. Image Credit: sweet_tomato/Shutterstock.com
Published in Radiology Cardiothoracic Imaging, the study introduces a deep learning system that analyzes standard chest radiographs for signs of hepatic steatosis, potentially replacing or supplementing specialized scans like MRI or ultrasound in early detection efforts. By leveraging imaging that’s already part of routine care, the approach could significantly expand access to screening for this common yet often undiagnosed condition.
Background
Fatty liver disease affects about 25 % of the global population and can progress to serious conditions such as cirrhosis or liver cancer if left untreated. While current diagnostic tools like ultrasound, CT scans, and MRI are effective, they also require costly equipment and dedicated facilities. In contrast, chest X-rays are widely available and inexpensive, but have traditionally been overlooked for liver assessment.
Recognizing that chest radiographs frequently capture part of the liver, Associate Professors Sawako Uchida-Kobayashi and Daiju Ueda saw an opportunity. Their team at Osaka Metropolitan University conducted a retrospective analysis to explore whether AI could identify signs of fatty liver in these underutilized images. Using 6599 chest X-rays from 4414 patients, they trained a model to detect hepatic steatosis based on controlled attenuation parameter (CAP) scores—an established measure of liver fat content.
Research Methodology and AI Development
The study analyzed data collected over a decade (2013–2023), focusing on patients who had both chest X-rays and CAP-based liver assessments. CAP scores, derived from vibration-controlled transient elastography, served as the gold standard for determining liver fat levels.
From a total of 6599 X-ray–CAP pairs, the researchers split the dataset into training (80 %), tuning (10 %), and internal test (10 %) sets from one institution, along with an external test set from a separate hospital. The AI model was trained to identify visual patterns in the chest X-rays that aligned with CAP-confirmed cases of hepatic steatosis.
Crucially, the model didn’t require any changes to how chest X-rays are captured. It simply made use of the upper right abdominal region that appears in standard images. The system achieved strong results across both internal and external test sets, with area under the curve (AUC) scores of 0.83 and 0.82, respectively. Specificity reached 82 % and 76 %, while sensitivity ranged from 68 % to 76 %, closely matching the 60–80 % sensitivity range typical of ultrasound in diagnosing fatty liver.
Clinical Implications and Future Applications
This approach opens up new possibilities for large-scale screening. Chest X-rays are performed far more frequently—often 100 times more—than abdominal ultrasounds in many healthcare settings. Integrating the AI model into existing radiology workflows could allow clinicians to flag potential liver issues while reviewing images ordered for other reasons, such as lung or heart conditions.
For clinics and hospitals with limited resources, this system could be especially valuable. It provides a scalable solution to identify high-risk patients without requiring additional imaging, reducing unnecessary follow-ups while prioritizing those who need further evaluation. The researchers also suggest that future iterations of the model could incorporate demographic or lab data to enhance accuracy even further.
Professor Uchida-Kobayashi emphasized the potential for this tool to reshape preventive care by making early detection more accessible. The team is now pursuing clinical trials to test the system in real-world settings and exploring its use for other conditions that may show up incidentally on chest X-rays. As obesity and fatty liver disease continue to rise worldwide, early identification will be key to preventing irreversible liver damage.
Conclusion
By using existing chest X-rays and a deep learning model, the study presents a practical and effective way to screen for fatty liver disease, without adding cost or increasing radiation exposure. The AI system delivers diagnostic performance comparable to specialized imaging, with the potential to bring early detection to more patients around the world. It’s a clear example of how AI can uncover hidden value in routine medical data, turning familiar tools into more powerful diagnostic assets.
Journal Reference
Ueda, D., Sawako Uchida-Kobayashi, Yamamoto, A., Walston, S. L., Motoyama, H., Fujii, H., Watanabe, T., Miki, Y., & Kawada, N. (2025). Performance of a Chest Radiograph–based Deep Learning Model for Detecting Hepatic Steatosis. Radiology Cardiothoracic Imaging, 7(3). DOI: 10.1148/ryct.240402. http://dx.doi.org/10.1148/ryct.240402
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.