High spatiotemporal-resolution abdominal 4D-MRI through respiratory-synchronized frame collaborative reconstruction

Area of Science:

• Medical Imaging
• Radiotherapy Guidance
• Image Reconstruction

Background:

• Four-dimensional magnetic resonance imaging (4D-MRI) is crucial for abdominal radiotherapy.
• Current 4D-MRI suffers from low spatial/temporal resolution and motion artifacts due to undersampling.
• Existing methods fail to fully utilize redundant frame information for detail restoration.

Purpose of the Study:

• Develop a novel technique to improve spatiotemporal resolution in abdominal 4D-MRI.
• Mitigate spatial undersampling by effectively leveraging multi-frame information.
• Enhance image quality for better clinical implementation in radiotherapy.

Main Methods:

• Introduced a multi-frame collaborative reconstruction network (MCR-Net) for 4D-MRI.
• Integrated an Inter-frame mutual-attention mechanism (IMM) to exploit inter-frame correlations and suppress noise.
• Utilized a structure-aware consolidation module (SaCM) for enhanced structural detail recovery and artifact reduction.

Main Results:

• MCR-Net significantly outperformed nine state-of-the-art methods in visual quality and quantitative accuracy (MAE, SSIM, PSNR).
• Achieved superior performance metrics: 3.77% (MAE), 1.03% (SSIM), and 6.74% (PSNR) improvement over next-best methods.
• Demonstrated enhanced registration accuracy (10.66% MAE, 3.60% SSIM, 1.94% NCC) and maintained quality under increased undersampling.

Conclusions:

• MCR-Net effectively suppresses artifacts and recovers anatomical structures from undersampled 4D-MRI data.
• The method has the potential to significantly improve 4D-MRI's spatiotemporal resolution.
• MCR-Net can advance clinical applications in abdominal radiotherapy guidance.