Three-dimensional rigid head motion correction using the Beat Pilot Tone and Gaussian Processes

In this work we propose a framework based on Gaussian Processes to extract quantitative motion information from Beat Pilot Tones and subsequently correct rigid head motion in 2D and 3D. In a calibration phase, low-resolution images are acquired and registered to build a training set. Next, Gaussian Processes are trained to infer rigid parameters from multi-channel BPTs, exploiting automatic relevance determination of input channels. In the inference phase, rigid parameters are inferred per readout from the BPT, and high-resolution scans are corrected for motion. In practice the method could reduce the need for anesthetics and/or re-scans in e.g. pediatric patients.