Rapid Liver Fibrosis Evaluation Using the UNet-ResNet50-32 × 4d Model in Magnetic Resonance Elastography:...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Hepatology

Background:

• Liver fibrosis is a critical indicator of chronic liver disease progression.
• Magnetic Resonance Elastography (MRE) offers a non-invasive method for assessing liver fibrosis.
• Accurate fibrosis staging is essential for effective patient management.

Purpose of the Study:

• To develop and evaluate deep learning (DL) models for automated liver fibrosis segmentation in MRE images.
• To compare the performance of a standard U-Net model against a UNet-ResNet50-32 × 4d architecture.
• To establish a reliable AI tool for quantitative fibrosis assessment.

Main Methods:

• Retrospective analysis of 319 MRE scans from patients (2018-2020).
• Segmentation of MRE images using conventional U-Net and UNet-ResNet50-32 × 4d models.
• Performance evaluation using correlation coefficients, Intersection over Union (IoU), and Dice scores.

Main Results:

• The UNet-ResNet50-32 × 4d model achieved high accuracy, with correlation coefficients of 0.952 (training) and 0.943 (validation).
• A Dice score of 85.68% confirmed the model's robust segmentation performance.
• The advanced model showed strong agreement with ground truth annotations.

Conclusions:

• The UNet-ResNet50-32 × 4d model is a reliable tool for rapid and accurate liver fibrosis assessment.
• Automated MRE analysis using DL can streamline clinical workflows.
• This technology supports timely clinical decisions in managing chronic liver diseases.