Enhancing compressed sensing and parallel imaging accelerated magnetic resonance angiography using a dual-domain...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Neuroscience

Background:

• Time-of-flight magnetic resonance angiography (TOF-MRA) is crucial for diagnosing cerebrovascular diseases, requiring high spatial resolution and sensitivity.
• Accelerating MRA scans is vital for reducing patient scan times and increasing throughput.
• Achieving diagnostic-quality MRA images at high undersampling rates presents a significant challenge.

Purpose of the Study:

• To develop a deep learning-based post-processing method for generating high-quality MRA images from highly undersampled data.
• To improve the detection sensitivity of vascular abnormalities through enhanced MRA image quality.

Main Methods:

• A prospective study utilized compressed sensing (CS) and sensitivity encoding (SENSE) to accelerate MRA scans at various factors.
• A Dual-Domain Projection Generative Adversarial Network (DDPGAN) was developed, processing 2D patches from accelerated MRA reconstructions.
• Quantitative metrics (PSNR, SSIM, NMSE, SNR) and qualitative reader assessments were used to evaluate image quality and diagnostic confidence.

Main Results:

• DDPGAN demonstrated superior performance over other deep learning models in both quantitative and qualitative evaluations.
• For CS×10, DDPGAN achieved high metrics: PSNR=31.98 dB, SSIM=0.88, NMSE=0.052, SNR=56.82.
• Qualitative assessments showed significant improvements in image clarity and diagnostic confidence, with accelerated scans surpassing standard scans in some cases.

Conclusions:

• The DDPGAN framework effectively enhances accelerated MRA image quality, preserving fine details and boosting diagnostic confidence.
• The dual-domain design of DDPGAN facilitates efficient, high-quality MRA examinations.
• DDPGAN shows strong clinical potential, particularly for evaluating cerebrovascular diseases.