Regularized referenceless temperature estimation in PRF-shift MR thermometry

In PRF-shift MR thermometry, temperature rises induced by thermal therapy can be estimated from the phase differences between tissue in a heated state and in a pretreatment state. To overcome errors caused by motion, referenceless methods have been proposed that estimate temperature from a single image. Compared to baseline subtraction, these methods are not susceptible to inter-frame motion, and produce accurate temperature estimates in moving tissues. However, referenceless methods could be improved if they did not require knowledge of the hot spot's location, or if they took into account underlying noise statistics. Here we introduce a regularized approach to referenceless PRF-shift thermometry that is statistically motivated, generalizes readily to multiple receive coils, and does not require knowledge of the hot spot's location.