Developing a Fully Automated Imaging Biomarker for HCC Risk Assessment via MRI-Based Tumor Segmentation and EPM.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Hepatocellular carcinoma (HCC) is a significant global health concern, necessitating improved early detection methods.
• Magnetic resonance imaging (MRI) is a key modality for liver imaging, but accurate tumor characterization remains challenging.
• Automated analysis of MRI could enhance diagnostic efficiency and accuracy for HCC.

Purpose of the Study:

• To evaluate the feasibility of automated tumor segmentation for early HCC detection using MRI.
• To assess the performance of a convolutional neural network (PocketNet) for liver tumor segmentation.
• To determine the clinical utility of automated imaging biomarkers derived from segmentation for HCC risk assessment.

Main Methods:

• Implementation of PocketNet, a lightweight convolutional neural network, for automated liver tumor segmentation on MRI.
• Utilizing Enhancement Pattern Mapping (EPM) as a voxel-wise imaging biomarker within segmented regions.
• Comparative analysis of automated segmentation performance against manual annotations and assessment of EPM feature predictability using XGBoost.

Main Results:

• Automated segmentation achieved performance comparable to the upper bound for larger tumors but underperformed for smaller lesions.
• Enhancement Pattern Mapping (EPM) features derived from automated segmentations showed predictive power similar to those from manual segmentations.
• EPM features demonstrated stability despite segmentation inaccuracies, indicating robustness.

Conclusions:

• Automated tumor segmentation is feasible for HCC detection using MRI, with EPM features proving robust.
• The developed automated imaging biomarker holds potential for clinical utility in HCC risk assessment.
• Further development could lead to a fully automated system for improved HCC management.