A generic deep learning model for reduced gadolinium dose in contrast-enhanced brain MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Rising safety concerns associated with gadolinium-based contrast agents (GBCAs) necessitate dose reduction strategies in contrast-enhanced MRI.
• Maintaining diagnostic image quality is crucial despite reduced contrast agent administration.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for predicting contrast-enhanced brain MRI images using approximately 10% of the standard GBCA dose.
• To ensure the model's robustness and generalizability across different clinical sites and MRI scanners.

Main Methods:

• A deep learning model incorporating multi-planar reconstruction, a 2.5D architecture, and specific loss functions (L1, perceptual, adversarial) was developed.
• The model was trained and validated on 640 heterogeneous 3D T1-weighted brain MRI scans from three institutions.
• Model performance was evaluated using quantitative metrics (PSNR, SSIM, Dice score) and qualitative assessment by board-certified radiologists.

Main Results:

• The DL model achieved high similarity to full-dose scans, with average PSNR of [value] dB and SSIM of [value].
• Radiologists confirmed the same enhancing patterns in 90% of cases, with minor discrepancies not affecting diagnosis.
• Quantitative tumor segmentation demonstrated strong performance with an average Dice score of [value] (median = 0.91).

Conclusions:

• A robust deep learning model was developed for low-dose contrast-enhanced brain MRI, offering potential for widespread clinical application.
• The proposed technical solutions enhance model generalizability across diverse imaging settings.
• This approach addresses safety concerns by significantly reducing GBCA administration while preserving diagnostic utility.