Progressive Generative Adversarial Networks for Medical Image Super\n resolution

Abstract

Anatomical landmark segmentation and pathology localization are important\nsteps in automated analysis of medical images. They are particularly\nchallenging when the anatomy or pathology is small, as in retinal images and\ncardiac MRI, or when the image is of low quality due to device acquisition\nparameters as in magnetic resonance (MR) scanners. We propose an image\nsuper-resolution method using progressive generative adversarial networks\n(P-GAN) that can take as input a low-resolution image and generate a high\nresolution image of desired scaling factor. The super resolved images can be\nused for more accurate detection of landmarks and pathology. Our primary\ncontribution is in proposing a multistage model where the output image quality\nof one stage is progressively improved in the next stage by using a triplet\nloss function. The triplet loss enables stepwise image quality improvement by\nusing the output of the previous stage as the baseline. This facilitates\ngeneration of super resolved images of high scaling factor while maintaining\ngood image quality. Experimental results for image super-resolution show that\nour proposed multistage P-GAN outperforms competing methods and baseline GAN.\n