Deep Learning Image Formation in Medical Imaging

The rise in AI approaches to image processing in the last ten years, particularly via deep Convolutional Neural Networks (Deep-CNNs) opened a new paradigm for medical image acquisition, reconstruction, and processing, yielding the recently coined “Deep Imaging” concept. The Deep Imaging paradigm opened a new space of imaging system designs and image formation methods with potential to overcome limitations of conventional methods based on models of the system physics and conventional computational approaches alone. Example of application of Deep Imaging to medical image formation and processing, addressed numerous tasks including CT image denoising, mitigation of artifacts in sparsely sampled acquisition patterns or in incomplete sampling scenarios, compensation of image biases, and compensation of non-idealities caused by the elements of the anatomy being imaged, such as artifacts caused patient motion.

This presentation will cover emerging Deep Imaging approaches, with emphasis on application to volumetric x-ray imaging for point-of-care imaging and interventional radiology applications.