Probabilistic medical image imputation via deep adversarial learning

The ability to impute missing images from a sequence of medical images plays an important role in enabling the detection, diagnosis and treatment of disease. Motivated by this, in this manuscript we propose a novel, probabilistic deep-learning algorithm for imputing images. Within this approach, given a sequence of contrast enhanced CT images, we train a generative adversarial network (GAN) to learn the underlying probabilistic relation between these images. Thereafter, given all but one member from a sequence, we infer the probability distribution of the missing image using Bayesian inference. We make the inference problem computationally tractable by mapping it to the low-dimensional latent space of the GAN. Thereafter, we use Markov Chain Monte Carlo (MCMC) techniques to learn and sample the inferred distribution. Moreover, we propose a novel style loss unique to contrast-enhanced computed tomography (CECT) imaging to improve the texture of the generated images, and apply these techniques to infer missing CECT images of renal masses collected during an IRB-approved retrospective study. In doing so, we demonstrate how the ability to infer the probability distribution of the missing image, as opposed to a single image recovery, can be used by the end-user to quantify the reliability of the imputed results.