An image-domain deep-learning denoising technique for accelerated parallel brain MRI: prospective clinical...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Parallel imaging accelerates MRI scans but can amplify noise and introduce aliasing artifacts.
• High acceleration factors in MRI can compromise image quality.
• Developing methods to mitigate artifacts from accelerated MRI is crucial for clinical applications.

Purpose of the Study:

• To evaluate a vendor-agnostic AI approach for artifact removal in highly accelerated MRI.
• To improve image quality in accelerated MRI scans.
• To reduce MRI scan times without sacrificing diagnostic information.

Main Methods:

• AI model trained on retrospectively degraded accelerated MRI images.
• Evaluation using retrospective and prospective highly accelerated MRI datasets from multiple vendors.
• Qualitative assessment by radiologists and quantitative analysis of noise, contrast, and resolution.

Main Results:

• AI model outputs were rated superior to inputs (P < .001) and comparable or superior to baseline images.
• Quantitative metrics showed improved signal-to-noise and contrast-to-noise ratios (P < .001).
• Average scan time reduction of 29% (19%-41% per sequence) was achieved.

Conclusions:

• The vendor-agnostic AI method effectively reduces MRI scan time.
• Image quality was preserved or enhanced, demonstrating robustness across different vendors and scanners.
• This AI approach offers potential for cost reduction and improved patient experience in MRI.