Robust Multi-Contrast MRI Medical Image Translation via Knowledge Distillation and Adversarial Attack.

Area of Science:

• Medical Imaging
• Artificial Intelligence

Background:

• Medical image translation is crucial but challenging due to noise and anatomical variations.
• Existing deep learning models (GANs, Transformers, Diffusion) struggle with perfect medical image translation.

Purpose of the Study:

• To propose a robust multi-contrast translation framework for MRI images.
• To enhance translation quality using knowledge distillation and adversarial attack methods.

Main Methods:

• Implemented a refinement network with teacher-student modules for knowledge distillation.
• Integrated an adversarial attack module to generate perturbations and improve generator robustness.
• Utilized a registration network as the teacher module to better learn noise distribution.

Main Results:

• The proposed framework demonstrated superior refinement and robustness in MRI translation.
• Ablation studies verified the effectiveness of each designed module.
• Experiments on public MRI datasets showed significant improvements over state-of-the-art methods.

Conclusions:

• The novel framework offers a significant advancement in medical image translation quality and reliability.
• The integration of knowledge distillation and adversarial attacks provides a robust solution for MRI enhancement.