A foundation model-driven multi-view collaborative framework for semi-supervised 3D medical image segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• 3D medical image segmentation is crucial for clinical decisions but requires extensive manual annotation.
• Semi-supervised learning (SSL) offers a solution by using limited labeled data with abundant unlabeled data.
• Annotation costs and time constraints hinder the widespread application of 3D segmentation.

Purpose of the Study:

• To develop an efficient semi-supervised 3D medical image segmentation framework.
• To reduce the dependency on high-quality voxel-level annotations.
• To enhance segmentation accuracy and generalizability across different medical imaging modalities.

Main Methods:

• A foundation model-driven multi-view collaborative learning framework was proposed.
• Exploited zero-shot capabilities of Segment Anything Model (SAM)-like foundation models.
• Integrated axial, sagittal, and coronal views using a collaborative fusion module.

Main Results:

• The proposed method outperformed existing SAM-based semi-supervised approaches on MRI brain tumor and PET heart segmentation tasks.
• Demonstrated improved boundary precision for organ and tumor delineation.
• Showcased strong transferability across different imaging modalities.

Conclusions:

• The foundation model-driven, multi-view collaborative learning paradigm advances semi-supervised 3D medical image segmentation.
• Provides a scalable and clinically relevant solution reducing annotation burden.
• Maintains high segmentation accuracy across diverse medical imaging modalities.