ReeGAN: MRI image edge-preserving synthesis based on GANs trained with misaligned data

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Magnetic Resonance Imaging (MRI) provides multi-angle and multi-dimensional insights vital for medical diagnostics.
• Cross-modality conversion research in MRI is significant but often overlooks misaligned data and edge information.

Purpose of the Study:

• To develop a robust MRI cross-modality conversion method that addresses challenges posed by misaligned data and preserves crucial edge information.
• To enhance the quality and reliability of synthesized target modality MRI images.

Main Methods:

• A generative adversarial network (GAN) with multi-feature fusion was proposed.
• The method incorporates an auxiliary registration network to adapt to noisy labels from misaligned data.
• Auxiliary edge information was injected to improve synthesized image quality.

Main Results:

• The proposed GAN effectively preserves edge information during MRI cross-modality conversion.
• The method demonstrates robustness when trained on noisy or misaligned data.
• Comprehensive experiments and ablation studies validated the technique's effectiveness.

Conclusions:

• The developed method offers a significant advancement in MRI cross-modality conversion, particularly for scenarios involving data imperfections.
• This approach enhances the utility of MRI by improving the quality of converted images and addressing common data alignment issues.