AI-Assisted Post Contrast Brain MRI: Eighty Percent Reduction in Contrast Dose

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Growing safety concerns regarding gadolinium-based contrast agents (GBCAs) necessitate dose reduction strategies in MRI.
• Maintaining diagnostic accuracy is crucial when reducing GBCA doses.

Purpose of the Study:

• To evaluate a deep learning (DL) method for synthesizing full-dose contrast-enhanced T1-weighted (T1w) MRI images from low-dose (20% standard) acquisitions.
• To assess the diagnostic performance of DL-generated images compared to standard-dose images.

Main Methods:

• A multicentric prospective study included 101 patients with various brain pathologies.
• A DL network processed pre-contrast and low-dose T1w sequences to generate synthesized full-dose T1w images (DL-T1w).
• DL-T1w images were compared to actual full-dose T1w images via quantitative metrics (SSIM, PSNR) and a reader study by three neuroradiologists.

Main Results:

• No significant difference in overall image quality between DL-T1w and full-dose T1w images (P=0.08).
• DL-T1w images showed significantly higher image Signal-to-Noise Ratio (SNR) and vessel conspicuity (P<0.05).
• Statistical noninferiority was demonstrated for border delineation, internal morphology, and contrast enhancement (P<0.001), with DL-T1w achieving 86% SSIM and 27 dB PSNR.

Conclusions:

• The proposed DL method effectively synthesizes postcontrast T1w MRI images comparable to full-dose images.
• This DL approach enables significant reduction in gadolinium contrast agent dose without compromising diagnostic quality.
• The findings support the potential of DL for safer and more efficient contrast-enhanced MRI.