This work proposes an efficient way to adapt MRI sampling patterns to a given anatomy and imaging context using a small set of representative training data. Such techniques were shown to help shorten MRI experiments while guaranteeing high image quality. An often encountered drawback of such methods are high computation times. We extend the recently proposed OEDIPUS framework by making use of the Bayesian Fisher information matrix. Based on the latter we devise an algorithm, which can be more than an order of magnitude faster than OEDIPUS for practical applications. This opens up the possibility to generate tailored sampling patterns for applications for which this would be infeasible otherwise. We evaluate our method in the context of multi-echo gradient echo imaging. The resulting sampling patterns show superior image reconstruction results compared to those obtained by other popularly used sampling schemes.
