Large Scale MRI Collection and Segmentation of Cirrhotic Liver

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Hepatology

Background:

• Liver cirrhosis is the advanced stage of chronic liver disease, posing significant mortality risks.
• Magnetic resonance imaging (MRI) is crucial for non-invasive assessment, but segmenting cirrhotic livers is challenging due to anatomical changes and signal variability.
• Current deep learning advancements are hindered by a lack of extensive, annotated datasets for cirrhotic liver analysis.

Purpose of the Study:

• To introduce CirrMRI600+, the first large-scale, expert-annotated dataset for cirrhotic liver segmentation using MRI.
• To provide benchmark results from deep learning models to establish performance standards for automated analysis.
• To facilitate the development and validation of computational tools for cirrhotic liver assessment.

Main Methods:

• Compiled a dataset of 628 high-resolution abdominal MRI scans, including T1- and T2-weighted sequences.
• Included expert-validated segmentation labels for cirrhotic livers across nearly 40,000 annotated slices.
• Collected associated demographic, clinical, and histopathological data where available.

Main Results:

• Established CirrMRI600+ as a comprehensive resource with 628 annotated MRI scans.
• Presented benchmark results from 11 deep learning models, setting performance baselines.
• Demonstrated the dataset's utility for training and evaluating AI models for liver cirrhosis analysis.

Conclusions:

• CirrMRI600+ addresses the critical need for large-scale annotated data in computational liver disease research.
• The dataset will accelerate the development of automated methods for cirrhosis visual staging.
• This resource supports personalized treatment planning through advanced AI-driven liver analysis.