Liver Cirrhosis Stage Estimation from MRI With Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Hepatology

Background:

• Liver cirrhosis is severe liver scarring, often diagnosed late, leading to complications like cancer and reduced life expectancy.
• Early diagnosis of liver cirrhosis is challenging but crucial for preventing severe health outcomes.
• Current diagnostic methods for liver cirrhosis can be invasive or lack sensitivity in early stages.

Purpose of the Study:

• To develop and evaluate an end-to-end deep learning framework for automated staging of liver cirrhosis using multi-sequence MRI.
• To improve the accuracy and efficiency of liver cirrhosis diagnosis, particularly in early stages.
• To establish a new benchmark for AI-based liver cirrhosis staging.

Main Methods:

• An integrated deep learning framework utilizing multi-scale feature learning and sequence-specific attention mechanisms.
• Training and validation on the CirrMRI600+ dataset, comprising 628 high-resolution MRI scans from 339 patients.
• Comparative analysis against traditional radiomics approaches and evaluation of various deep learning architectures (VGGs, ResNets, Mamba).

Main Results:

• State-of-the-art performance in three-stage liver cirrhosis classification.
• Achieved 72.8% accuracy on T1W and 63.8% accuracy on T2W MRI sequences.
• Demonstrated superior performance compared to traditional radiomics methods and effective learning of stage-specific imaging biomarkers.

Conclusions:

• The proposed deep learning framework shows significant potential for accurate and automated liver cirrhosis staging from MRI.
• The study establishes new benchmarks for AI in liver cirrhosis assessment, paving the way for clinical applications.
• The findings highlight the capability of deep learning to identify subtle imaging biomarkers indicative of liver fibrosis progression.