Model based learning for accelerated, limited-view 3D photoacoustic tomography.

Recent advances in deep learning for tomographic reconstructions have showngreat potential to create accurate and high quality images with a considerablespeed-up. In this work we present a deep neural network that is specificallydesigned to provide high resolution 3D images from restricted photoacousticmeasurements. The network is designed to represent an iterative scheme andincorporates gradient information of the data fit to compensate for limitedview artefacts. Due to the high complexity of the photoacoustic forwardoperator, we separate training and computation of the gradient information. Asuitable prior for the desired image structures is learned as part of thetraining. The resulting network is trained and tested on a set of segmentedvessels from lung CT scans and then applied to in-vivo photoacousticmeasurement data.