A lightweight super-resolution network with skip-connections

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Low-resolution MRI images are common in remote areas due to limited scanner availability.
• This limitation hinders accurate diagnosis and treatment planning.
• Developing accessible image enhancement solutions is crucial for equitable healthcare.

Purpose of the Study:

• To develop a lightweight super-resolution algorithm for MRI images.
• To improve the resolution of MRI scans obtained from low-field intensity scanners.
• To enable accurate diagnoses in resource-limited clinical settings.

Main Methods:

• Compared super-resolution algorithms including SRGAN, SPSR, and LESRCNN.
• Modified the LESRCNN network by incorporating a global skip connection.
• Evaluated performance using metrics like SSMI, PSNR, PI, and LPIPS.

Main Results:

• The proposed algorithm demonstrated improved SSMI, PSNR, PI, and LPIPS compared to LESRCNN.
• The algorithm is computationally efficient with a small parameter count and low complexity.
• Clinical evaluation by MRI doctors confirmed significant improvements and potential for remote use.

Conclusions:

• The developed algorithm effectively reconstructs high-resolution MRI images from low-resolution inputs.
• Its lightweight nature makes it suitable for deployment in remote hospitals with limited computing resources.
• The algorithm holds significant clinical value, aiding diagnoses and saving patient time.