DTI at four directions: an application at ultra-low field

Motivation: The project aimed to tackle extended DTI scan durations, worsened by low SNR at low fields, striving to boost efficiency while preserving results' accuracy at lower SNR levels. Goal(s): The study sought to create an ML-based approach to shorten DT-MRI scans while ensuring reliable tensor estimation despite low SNR challenges at ULF. Approach: ML models, trained on synthetic data, predicted diffusivities and principal eigenvectors from four diffusion-weighted images, factoring in simulated noise and gradient rotations for noise and motion. Results: The models estimated diffusivities and fibre orientations with fewer data, showing promise for ULF tractography. Suggesting shorter DTI scans are possible at ULF. Impact: Our results are relevant to clinicians and researchers using low-field MRI, potentially enabling faster DT-MRI scans, opening avenues for efficient DTI in challenging settings, and making fibre mapping more accessible with reduced acquisition scan times.