Fast Super-Resolution in MRI Images Using Phase Stretch Transform, Anchored Point Regression and Zero-Data Learning

Medical imaging is fundamentally challenging due to absorption and scattering in tissues and by the need to minimize illumination of the patient with harmful radiation. Common problems are low spatial resolution, limited dynamic range and low contrast. These predicaments have fueled interest in enhancing medical images using digital post processing. In this paper, we propose and demonstrate an algorithm for real-time inference that is suitable for edge computing. Our locally adaptive learned filtering technique named Phase Stretch Anchored Regression (PhSAR) combines the Phase Stretch Transform for local features extraction in visually impaired images with clustered anchored points to represent image feature space and fast regression based learning. In contrast with the recent widely-used deep neural network for image super-resolution, our algorithm achieves significantly faster inference and less hallucination on image details and is interpretable. Tests on brain MRI images using zero-data learning reveal its robustness with explicit PSNR improvement and lower latency compared to relevant benchmarks.