Towards High-Quality MRI Reconstruction With Anisotropic Diffusion-Assisted Generative Adversarial Networks and Its...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Image Reconstruction

Background:

• Fast Magnetic Resonance Imaging (MRI) reconstruction is crucial for improving clinical diagnostics by reducing scan times.
• Existing Generative Adversarial Networks (GANs) for MRI reconstruction face challenges with noise, leading to blurred details and artifacts.

Purpose of the Study:

• To develop a novel deep learning framework for enhanced MRI image reconstruction.
• To address noise and detail preservation issues in current MRI reconstruction methods.
• To improve the authenticity and accuracy of generated MR images.

Main Methods:

• Proposed a novel deep framework: Anisotropic Diffusion-Assisted Generative Adversarial Networks (AD-GANs).
• Integrated an Anisotropic Diffused Reconstruction Module to minimize reconstruction losses and denoise outputs.
• Implemented multi-modal learning to aggregate information from diverse MRI modalities.
• Optimized a joint loss function within a unified framework to preserve high-frequency information and structural details.

Main Results:

• The AD-GANs framework achieved superior performance in reconstructing high-quality MR images.
• Demonstrated significant improvements in Peak Signal-to-Noise Ratio (PSNR) and multi-scale Structural Similarity Index Measure (mSSIM).
• Achieved average PSNR of 35.785 dB and mSSIM of 0.9765 on the MRNet dataset, outperforming baselines by at least 2.9 dB and 0.07.

Conclusions:

• The proposed AD-GANs framework effectively enhances MRI reconstruction quality.
• The novel approach minimizes noise while preserving crucial image details, enabling more authentic MR image generation.
• The framework shows promise for improving clinical diagnostic accuracy through advanced AI-driven image reconstruction.