Fourier Convolution Block with global receptive field for MRI reconstruction

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Signal Processing

Background:

• Under-sampled Magnetic Resonance Imaging (MRI) accelerates scans but requires advanced reconstruction techniques.
• Convolutional Neural Networks (CNNs) excel at MRI reconstruction but often have limited receptive fields (RF), hindering global feature capture.
• Global feature extraction is vital for mitigating aliasing artifacts in MRI reconstruction.

Purpose of the Study:

• To develop a novel global Fourier Convolution Block (FCB) for enhancing MRI image reconstruction.
• To address the limitations of restricted receptive fields (RF) in CNN-based MRI reconstruction methods.
• To improve the capture of global image features and reduce aliasing artifacts.

Main Methods:

• Proposed a global Fourier Convolution Block (FCB) that transforms spatial convolutions to the frequency domain for a whole-image receptive field (RF).
• Integrated FCB into four popular CNN architectures for MRI reconstruction.
• Evaluated performance on brain and knee MRI datasets, comparing against baseline models.

Main Results:

• FCB demonstrated an improved effective receptive field (RF) in CNN models.
• Models incorporating FCB achieved superior Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) compared to baseline models.
• Enhanced recovery of image details and textures was observed in reconstructed MRI scans.

Conclusions:

• The proposed global Fourier Convolution Block (FCB) effectively enhances CNN-based MRI reconstruction by expanding the receptive field (RF).
• FCB offers a computationally efficient method for improving image quality, detail, and texture recovery in under-sampled MRI.
• This approach holds promise for accelerating MRI scans and improving clinical diagnostic accuracy.