A BCLC Staging System for Hepatocellular Carcinoma using Swin Transformer and CT Imaging

Area of Science:

• Medical Imaging
• Oncology
• Artificial Intelligence

Background:

• The Barcelona Clinic Liver Cancer (BCLC) staging system is vital for hepatocellular carcinoma (HCC) management.
• Accurate BCLC staging aids radiologists, but dedicated datasets are scarce.
• Existing datasets often have imbalanced BCLC class distribution due to varied curation purposes.

Purpose of the Study:

• To develop an automated BCLC staging system utilizing the Swin Transformer model.
• To investigate the integration of disparate datasets for BCLC staging.
• To address the challenge of maintaining class distribution in integrated datasets for clinical applicability.

Main Methods:

• Implementation of the Swin Transformer deep learning architecture.
• Integration of two distinct medical imaging datasets.
• Evaluation of class distribution preservation during dataset merging.

Main Results:

• Development of a BCLC staging system with 55.81% (±7.8%) accuracy.
• Demonstration of challenges and strategies for integrating datasets with differing objectives.
• Highlighting the importance of class distribution in practical AI model development for HCC.

Conclusions:

• The Swin Transformer model shows potential for BCLC staging in HCC.
• Dataset integration for medical AI requires careful consideration of class balance.
• Further research is needed to improve accuracy and clinical utility of AI-based BCLC staging.