Related Experiment Video
Updated: Jan 13, 2026
05:37
Author Spotlight: A Non-Invasive Tool to Assess and Differentiate Fat Patterns in Liver Using 3D Dixon MRI
Published on: October 20, 2023
2.1K
Development and validation of an artificial intelligence model based on liver CSE-MRI fat maps for predicting
Bo Jiang1, Weijun Situ1, Zhichao Feng2
1Department of Radiology, The Second Xiangya Hospital of Central South University, Changsha, Hunan, China.
PLOS Digital Health
|January 7, 2026
Summary
An artificial intelligence (AI) model using liver MRI fat maps can non-invasively detect dyslipidemia early. This AI tool shows high accuracy in predicting lipid abnormalities, serving as a potential early warning system.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Cardiovascular Health
Background:
- Dyslipidemia is a major risk factor for cardiovascular disease.
- Current detection methods often involve invasive blood tests.
- Non-invasive early detection methods are needed to improve patient outcomes.
Purpose of the Study:
- To develop and validate an AI model for non-invasive early detection of dyslipidemia.
- To utilize liver chemical shift-encoded MRI (CSE-MRI) fat maps for lipid abnormality prediction.
- To assess the model's accuracy and generalization capabilities.
Main Methods:
- An automated AI pipeline was created using transfer learning with pre-trained networks (ResNet18, MobileNet, DenseNet, AlexNet, SqueezeNet).
- Liver CSE-MRI fat images from 1,757 scans of 89 patients were used.
- Model performance was evaluated using 8-fold cross-validation and an independent test set.
Main Results:
- The ResNet18-based AI model achieved high accuracy on the test set.
- Accuracies for predicting lipid abnormalities were: triglyceride (0.853), total cholesterol (0.833), LDL cholesterol (0.937), and HDL cholesterol (0.936).
- High F1-Scores were observed for triglyceride (0.885), LDL (0.886), and HDL (0.897) cholesterol.
Conclusions:
- The AI model utilizing liver CSE-MRI fat maps demonstrates significant potential for early dyslipidemia detection.
- The model shows high accuracy and generalization for predicting key lipid indices.
- Further enhancement of performance for all lipid indices may be achieved by expanding the training dataset.
