DAMO Launches MAOSS Fatty Liver AI Model

🔑 Enhanced Key Takeaways

• •MAOSS was jointly developed by Alibaba DAMO Academy with Shengjing Hospital of China Medical University and Gulou Hospital of Nanjing University[1].
• •Research on MAOSS was published in Nature Communications on February 11, 2026, detailing multi-modal AI for steatotic liver disease screening using non-contrast CT scans[3].
• •MAOSS achieved AUC scores of 0.904-0.929 for steatosis detection and 0.824-0.888 for significant fibrosis, validated against histology and MRI-PDFF gold standards[3].

🛠️ Technical Deep Dive

• •Trained on large dataset: 968 histopathologically confirmed cases and 1103 radiologically confirmed cases; validated on 660 histology and 375 MRI-PDFF cases[3].
• •Uses unenhanced CT scans to extract high-dimensional features like liver texture and density for simultaneous steatosis and fibrosis staging[1][3].
• •In multi-center validation, AUC for liver steatosis staging: 0.904-0.917 (vs. radiologists' 0.709); identifies 52.4% of stage 2 fibrosis high-risk patients vs. 16.6% in traditional pathways[1][2].
• •Integrates into clinical pathways to identify 36% more fibrosis progression risk patients; Cox model shows higher cirrhosis hazard ratio (5.54) for intermediate-high risk group[3].

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline