MDRN: Multi-distillation residual network for efficient MR image super-resolution.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Super-resolution (SR) magnetic resonance imaging (MRI) provides detailed anatomical information.
• Current SR methods often rely on complex convolutional networks, posing training challenges and high computational demands unsuitable for medical settings.

Purpose of the Study:

• To develop an efficient and accurate SR method for MRI that addresses the limitations of existing approaches.
• To introduce a novel Multi-Distillation Residual Network (MDRN) optimized for computational efficiency and reconstruction quality.

Main Methods:

• Proposed a Multi-Distillation Residual Network (MDRN) for enhanced feature refinement.
• Designed a novel feature multi-distillation residual block incorporating a contrast-aware channel attention module.
• Focused residual features on low-vision information to maximize network performance.

Main Results:

• MDRN demonstrated superior trade-off between reconstruction accuracy and computation cost compared to state-of-the-art methods.
• Achieved higher peak signal-noise ratio (up to 1.82 dB) at 4x scale.
• Required lower GPU memory and runtime than competing SR methods.

Conclusions:

• MDRN offers a significant advancement in SR MRI, balancing high reconstruction quality with computational efficiency.
• The proposed method is well-suited for resource-constrained medical environments.
• Source code will be publicly available for further research and application.