Enhanced Artificial Intelligence Methods for Liver Steatosis Assessment Using Machine Learning and Color Image
Concepción Gómez-Gavara1,2, Itxarone Bilbao1,2, Gemma Piella3
1Barcelona Autonoma University, Universitat Autónoma de Barcelona, Barcelona, Spain.
Clinical Transplantation
|October 9, 2024
Summary
This study developed a smartphone-based method using machine learning to accurately assess liver steatosis, improving donor liver selection for transplantation.
Area of Science:
- Transplantation research
- Medical imaging analysis
- Machine learning applications
Background:
- Significant liver steatosis negatively impacts liver transplant outcomes.
- Current assessment of donor liver steatosis relies on subjective visual inspection.
- Objective, accurate methods are needed to evaluate donor liver steatosis.
Purpose of the Study:
- To develop a rapid, robust, accurate, and cost-effective method for assessing liver steatosis.
- To utilize smartphone photography and machine learning for objective liver steatosis evaluation.
Main Methods:
- Collected smartphone photographs and liver biopsies from 192 adult brain death donor livers.
- Applied color calibration, segmentation, and feature extraction to liver images.
- Utilized a random forest machine learning classifier (LiverColor project) for analysis.
Main Results:
- The machine learning model achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.74.
- The model demonstrated 85% accuracy in determining macrosteatosis at a 30% threshold.
- Analysis included 362 photographs and 7240 image patches.
Conclusions:
- Machine learning analysis of smartphone liver images accurately determines steatosis.
- This approach offers a promising tool for objective donor liver assessment.
- The LiverColor project provides a foundation for improved liver transplantation protocols.
