L-TGVN: New AI slashes MRI time using patient's prior scans
No more long, uncomfortable MRI sessions—AI reconstructs from 10x fewer measurements
L-TGVN aims to solve one of MRI's biggest pain points: long acquisition times. The model uses a deep variational network that intelligently incorporates a patient's most recent prior scan as side information, while dynamically controlling how much influence that prior has based on consistency with newly acquired measurements. This design makes it robust to temporal changes like pathology progression, misalignment, and protocol drift across visits.
In tests against matched-capacity baselines—including prior-guided methods and non-longitudinal approaches—L-TGVN showed consistent improvements in quantitative metrics and better preservation of fine details, even at challenging acceleration factors. The code is publicly available, and the work has been accepted at MICCAI 2026, signaling strong peer validation from the medical imaging community.
- Uses patient's prior MRI as personalized prior, reducing scan time by up to 10x with no need for explicit pre-registration
- Robust to temporal changes (pathology progression), misalignment, and protocol differences across visits
- Accepted at MICCAI 2026; code released on GitHub for reproducibility
Why It Matters
Faster, personalized MRI means lower costs, higher patient comfort, and more throughput—without sacrificing diagnostic quality.