MedGAN-SSM: Multimodal brain image synthesis network integration using SSM empowered GAN

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Multimodal medical image synthesis is crucial for overcoming data limitations in clinical settings.
• Existing methods struggle with data scarcity and incomplete imaging modalities.
• High-fidelity synthesis is needed to support downstream tasks like segmentation.

Purpose of the Study:

• To present MedGAN-SSM, a framework enhancing multimodal brain MRI synthesis.
• To improve image quality and anatomical consistency in generated MRIs.
• To demonstrate robustness in scenarios with missing imaging modalities.

Main Methods:

• Integration of generative adversarial networks (GANs) with state space modeling (SSM).
• Utilizing a state space module for global semantic information capture via cross-layer transmission.
• Employing a dynamic attention gate for feature adjustment and an S6 module for multi-scale feature fusion.

Main Results:

• MedGAN-SSM outperforms existing methods on BraTS2020 and IXI datasets in PSNR, SSIM, and MAE.
• The framework shows robustness in missing-modality scenarios.
• Generated high-fidelity images that improved segmentation performance.

Conclusions:

• MedGAN-SSM reliably synthesizes high-quality multimodal brain MRI, preserving anatomical details.
• The framework aids automated analyses and clinical workflows, especially under incomplete imaging conditions.
• This approach offers a robust solution for data augmentation and missing data imputation in medical imaging.