HACR-Net: An Efficient hybrid attention network for MRI image super-resolution

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• High-resolution Magnetic Resonance Imaging (MRI) is crucial for clinical diagnosis but faces hardware and time constraints.
• Super-resolution (SR) techniques aim to reconstruct high-resolution MRI from low-resolution inputs, yet often fail to capture fine details and complex dependencies.

Purpose of the Study:

• To introduce the Hybrid Attention and Channel Retention Network (HACR-Net) for improved MRI super-resolution.
• To address limitations in existing SR methods regarding shallow feature extraction, contextual modeling, and anatomical detail preservation.

Main Methods:

• Developed HACR-Net incorporating a Hybrid Attention Module (HAM) for enhanced feature extraction using channel and spatial attention.
• Integrated a Multiscale Feature Aggregation Block (MFAB) to capture diverse image details from global structures to high frequencies.
• Employed a Channel Retention Attention Block (CRAB) with a bottleneck design to preserve fine contextual details and minimize information loss.

Main Results:

• HACR-Net demonstrated high-performance MRI image reconstruction on IXI and BraTS2018 datasets.
• The proposed network achieved this with a significantly reduced parameter count (1.67M) and computational cost (81.3G FLOPs).

Conclusions:

• HACR-Net offers an effective solution for MRI super-resolution, outperforming existing methods.
• The network's efficiency in terms of model size and computational load makes it a promising tool for clinical applications.