Med-SRNet: GAN-Based Medical Image Super-Resolution via High-Resolution Representation Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• High-resolution (HR) medical imaging is crucial for early disease diagnosis.
• Acquiring clear HR medical images is challenging due to technical and environmental limitations.

Purpose of the Study:

• To develop a novel medical image super-resolution (SR) method using generative adversarial networks (GANs).
• To improve the quality and detail of medical images for better diagnostic accuracy.

Main Methods:

• A novel medical image super-resolution (SR) method, Med-SRNet, based on generative adversarial network (GAN) was developed.
• The GAN generator incorporates High-Resolution Network (HRNet) to maintain HR representations with multi-scale fusions.
• Deconvolution operations were used to recover HR representations, enhancing feature aggregation.

Main Results:

• Med-SRNet demonstrated superior performance compared to existing methods on medical image datasets.
• The method achieved higher Peak Signal to Noise Ratio (PSNR) and Structural Similarity (SSIM) values.
• Significant improvements were observed on brain and lung CT datasets, particularly at 8x super-resolution.

Conclusions:

• Med-SRNet effectively enhances medical image quality through advanced GAN-based super-resolution.
• The proposed method shows significant potential for improving diagnostic capabilities in medical imaging.
• The approach offers a robust solution for generating realistic high-frequency details in medical images.