Deep Learning Boosts Brain MRI Contrast Without Increasing Gadolinium Dose

A deep-learning (DL) algorithm may significantly improve lesion visualization and image quality on contrast-enhanced brain MRI without requiring additional gadolinium-based contrast agent (GBCA), according to a multicenter, multireader study published in the Journal of Magnetic Resonance Imaging.

The retrospective study evaluated an FDA-cleared DL-based contrast-boosting algorithm using brain MRI examinations from 110 patients (mean age, 47 ± 22 years). Images were acquired at 0.3T, 1.5T, and 3T. Pre-contrast and standard post-contrast images were processed to generate contrast-boosted (CB) images, which were then compared with standard-contrast (SC) images.

Three board-certified radiologists with 9, 15, and 18 years of experience evaluated the images for lesion enhancement, border delineation, internal morphology, overall image quality, artifacts, and vessel conspicuity.

Quantitative analysis showed that CB images significantly improved contrast-to-noise ratio by 729.17%, lesion-to-brain ratio by 87.91%, and contrast enhancement percentage by 165.81% compared with SC images (all P < .05).

Among 61 cases with lesions, CB images also received significantly higher scores for lesion contrast enhancement (3.84 vs 3.16), border delineation (3.78 vs 3.15), and internal morphology (3.56 vs 3.16). Overall image quality improved from 3.07 with standard contrast to 3.55 with contrast boosting.

A physics-based analysis estimated that CB images produced a 92.5% ± 62.7% increase in apparent contrast compared with SC images, approximating the enhancement expected from a double contrast dose. However, the researchers emphasized that this was a model-based estimate rather than a direct comparison with prospectively acquired double-dose images.

False-positive findings were primarily associated with registration inconsistencies, existing imaging artifacts, and increased vessel conspicuity. Most had minor or no diagnostic impact, and readers reported that the original SC images allowed them to rule out false-positive lesions in all cases.

The researchers concluded that DL-based contrast boosting produced “significantly superior lesion visualization performance” compared with standard-dose contrast-enhanced brain MRI and could potentially support applications requiring greater lesion sensitivity without increasing the administered contrast dose. Future studies are needed to establish diagnostic accuracy and clinical utility.