Deep learning-based methods may minimize GBCA dosage in brain MRI.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Gadolinium-based contrast agents (GBCAs) are crucial for brain MRI.
• Reducing GBCA dosage is desirable for patient safety and cost-effectiveness.
• Deep learning (DL) offers potential for image reconstruction with reduced contrast agents.

Purpose of the Study:

• To assess the clinical performance of a DL-based method for brain MRI with reduced GBCA dose.
• To understand the capabilities and limitations of DL in minimizing GBCA usage.
• To evaluate diagnostic accuracy and image quality with a DL-synthesized full-dose contrast-enhanced MRI.

Main Methods:

• Eighty-three patients undergoing brain contrast-enhanced (CE) MRI were included.
• Three datasets were acquired: zero-dose, 10% GBCA (low-dose), and 100% GBCA (full-dose).
• A DL model was trained to synthesize full-dose images from zero- and low-dose data, with 53 cases used for testing.

Main Results:

• DL-synthesized images achieved 90.6% accuracy in matching lesion detection compared to true full-dose images.
• The DL method demonstrated high accuracy in identifying single enhanced lesions (94.4%).
• Agreement scores for image quality, SNR, lesion conspicuity, and enhancement ranged from 0.63 to 0.89.

Conclusions:

• The DL method is a feasible approach to significantly reduce GBCA dosage in brain MRI without compromising diagnostic information.
• The DL method shows potential for routine radiological diagnosis in specific clinical scenarios.
• Further algorithmic improvements are needed to address missed small lesions in cases with multiple enhancing lesions.