TASC-SwinMT: Task-Adaptive Synergistic Cross-Task Swin Multi-Task Framework for CT and MRI Image Interpolation and...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Computational Anatomy

Background:

• Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) interpolation and segmentation are vital for clinical applications.
• Current methods often process these tasks separately, leading to inefficiencies and missed opportunities for feature sharing.
• This necessitates integrated approaches for improved medical image analysis.

Purpose of the Study:

• To develop a unified multi-task learning framework for simultaneous medical image interpolation and segmentation.
• To leverage shared spatial features between interpolation and segmentation tasks.
• To enhance the accuracy and efficiency of medical image analysis pipelines.

Main Methods:

• Proposed TASC-SwinMT, a unified multi-task framework using a shared SwinUNet encoder and task-specific decoders.
• Implemented three modules for cross-task synergistic learning and a dynamic multi-task loss function.
• Validated on Medical Segmentation Decathlon datasets (Task02_Heart, Task06_Lung).

Main Results:

• TASC-SwinMT outperformed baseline models in both interpolation and segmentation tasks.
• Achieved superior performance in lesion boundary depiction, small object segmentation, and inter-slice consistency.
• Demonstrated significantly reduced computational overhead compared to separate methods.

Conclusions:

• The proposed cross-task feature sharing and joint optimization strategy proved effective.
• TASC-SwinMT exhibits strong stability and generalization for clinical medical image analysis.
• The framework offers a reliable solution for integrated CT and MRI analysis.