Fine Perceptive GANs for Brain MR Image Super-Resolution in Wavelet Domain.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Image Processing

Background:

• Magnetic Resonance (MR) imaging is crucial for clinical and brain studies.
• Acquiring high-resolution MR images is challenging due to hardware, time, and cost limitations.

Purpose of the Study:

• To propose fine perceptive generative adversarial networks (FP-GANs) for generating super-resolution (SR) MR images.
• To improve the quality and detail of MR images for enhanced clinical applications.

Main Methods:

• FP-GANs utilize a divide-and-conquer approach, processing low-frequency and high-frequency components separately.
• The method decomposes MR images into approximation and texture subbands using wavelet transforms.
• Generators with residual-in-residual dense blocks and a texture-enhancing module are employed for SR reconstruction.

Main Results:

• FP-GANs successfully generate super-resolution MR images with finer structure recovery.
• The proposed model outperforms existing methods quantitatively and qualitatively on MultiRes_7T and ADNI datasets.
• SR results from FP-GANs demonstrated value in subsequent classification tasks.

Conclusions:

• FP-GANs offer a novel and effective solution for MR image super-resolution.
• The method enhances the diagnostic potential of MR imaging by improving image quality.
• This approach has implications for advancing clinical analysis and brain exploration through improved MR imaging techniques.