MP-Net: A Multi-Center Privacy-Preserving Network for Medical Image Segmentation

Area of Science:

• Medical Imaging
• Computer Science
• Data Security

Background:

• Multi-center collaborations are crucial for robust medical image analysis.
• Existing methods like federated learning face challenges in data transmission and privacy.
• Secure and efficient collaborative learning frameworks are needed for medical imaging.

Purpose of the Study:

• To present the Multi-Center Privacy-Preserving Network (MP-Net) for secure medical image segmentation.
• To develop a novel framework for multi-center collaborative learning that enhances data privacy and reduces data transmission.
• To enable secure, ciphertext-based image segmentation through computable image encryption.

Main Methods:

• MP-Net utilizes a three-layer model: encryption, segmentation, and decryption networks.
• Image data is encrypted into ciphertext using an encryption network.
• An improved U-Net performs segmentation on the encrypted image data, followed by decryption to obtain the segmentation mask.

Main Results:

• The MP-Net framework enables secure utilization of multi-center data for segmentation.
• The method significantly reduces data transmission volume compared to traditional federated learning.
• Evaluations on cardiac MRI and CTPA datasets demonstrate the framework's effectiveness.

Conclusions:

• MP-Net provides a novel and secure approach to multi-center medical image segmentation.
• The framework enhances data privacy protection while maintaining segmentation accuracy.
• MP-Net facilitates the development of more robust segmentation models by securely leveraging distributed data.