Project DetailsTissue contrast in magnetic resonance imaging (MRI) can be enhanced by pulse-sequence optimization, which recent methods automate. However, their reliance on target images limits exploration of new contrasts. We develop a pulse-sequence optimization method that enhances tissue contrast without any target image. First, we propose a contrast-promoting image metric that generalizes the contrast-to-noise ratio by measuring f-divergences between tissue intensity distributions. From this metric, we construct a target-free loss function guided by tissue probability maps and prove it is robust to probability map errors beyond those expected in practice. We demonstrate our target-free optimization on both simulated MRI data and real-world T1- and T2-weighted data acquired in vivo and from phantoms at 3T. Across simulated and real-world experiments, starting from poor contrast, our method iteratively found parameters that improved contrast. It also succeeded under challenging conditions: large errors in the probability maps combined with irregular intensity distributions; and a small lesion-like tissue with no initial contrast. Our target-free method enhances tissue contrast and explores contrasts not prescribed in advance. It is particularly valuable when high-quality targets are unavailable, such as in low-field MRI.
