PIMedSeg: Progressive interactive medical image segmentation.

Area of Science:

• Medical Imaging Analysis
• Computer Vision in Healthcare
• Biomedical Engineering

Background:

• Accurate medical image segmentation is vital for diagnosis but remains challenging with automatic methods.
• Interactive segmentation offers a promising alternative to overcome limitations of fully automatic approaches.
• Existing interactive methods often require substantial user effort.

Purpose of the Study:

• To propose a novel interactive segmentation framework.
• To reduce user effort while achieving high-quality segmentation results.
• To enhance the efficiency of medical image segmentation.

Main Methods:

• A progressive workflow incorporating user-provided region clicks and edge scribbles.
• Utilizes a novel disk and curve transform for encoding user input.
• Employs a transformer-based module for feature refinement, integrating CNN outputs and input maps.

Main Results:

• Demonstrated effectiveness across diverse medical imaging modalities, including ultrasound (US), CT, and MRI.
• Outperformed state-of-the-art alternative segmentation approaches in experiments.
• Achieved high-quality segmentation with minimal user interaction.

Conclusions:

• The proposed framework offers a viable solution for high-quality medical image segmentation.
• Significantly reduces the need for extensive manual segmentation efforts.
• Provides a balance between user interaction and segmentation accuracy.