The project aims to provide and validate new approaches for machine learning in prioritised task scheduled working queues in mega-kernels executed on single instruction multiple data (SIMD) computing units. A working demonstrator using a complex real world algorithm for motion correction in fetal MRI will be used and validated on real, motion corrupted MRI data. With the proposed learning strategies, it is expected to provide accurate reconstructions of the fetal anatomy in-utero and a general framework for the parallelisation of otherwise highly complex computational methods. The fundamental GPU computing methods provide a versatile framework, which will be extended with machine learning methods to automatically and intelligently define task priorities.
EPSRC — EP/N024494/1