DiffCNN: A collaborative framework of diffusion model and CNN for semi-supervised medical image segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Teacher-student architectures are common in semi-supervised medical image segmentation.
• Existing methods face challenges with teacher subnet optimization and handling noisy images due to CNN limitations.

Purpose of the Study:

• To propose DiffCNN, a collaborative framework using diffusion models and CNNs for improved semi-supervised medical image segmentation.
• To address limitations of traditional teacher-student architectures in noisy medical image segmentation.

Main Methods:

• DiffCNN employs distinct CNN and diffusion subnets for collaborative learning.
• The diffusion subnet learns mask distributions to mitigate noise.
• Adversarial learning enhances the diffusion subnet's performance by aligning with real masks.

Main Results:

• DiffCNN demonstrates superior performance compared to state-of-the-art methods on three medical image segmentation datasets.
• The collaborative framework effectively extracts complementary information and handles noisy images.

Conclusions:

• DiffCNN offers a robust and effective approach for semi-supervised medical image segmentation.
• The integration of diffusion models and CNNs presents a promising direction for medical image analysis.