SDR-Former: A Siamese Dual-Resolution Transformer for liver lesion classification using 3D multi-phase imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Automated liver lesion classification in multi-phase CT and MR scans is crucial but difficult.
• Existing methods struggle with varying phase counts and capturing complex features.

Purpose of the Study:

• To develop a novel framework for accurate liver lesion classification in 3D multi-phase CT and MR imaging.
• To address challenges posed by varying phase numbers and enhance feature representation.

Main Methods:

• A Siamese Dual-Resolution Transformer (SDR-Former) framework was proposed, utilizing a Siamese Neural Network (SNN).
• A hybrid Dual-Resolution Transformer (DR-Former) combined 3D CNN and 3D Transformer for multi-resolution analysis.
• An Adaptive Phase Selection Module (APSM) was introduced for dynamic phase weighting.

Main Results:

• The SDR-Former framework demonstrated significant efficacy in liver lesion classification across 3-phase CT and 8-phase MR datasets.
• Experimental validation confirmed the framework's ability to capture both local and global features effectively.

Conclusions:

• The proposed SDR-Former framework offers a robust and computationally efficient solution for liver lesion classification.
• A pioneering, publicly available multi-phase MR dataset for liver lesion analysis was released to support the research community.