2019/07/26 by Guilherme Pombo, Pombo, Guilherme, Robert M. Gray +6
Computer Science · #Bayesian Methods and Mixture Models #Computation (stat.CO) #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Gaussian Processes and Bayesian Inference #Generative Adversarial Networks and Image Synthesis #Image and Video Processing (eess.IV) #Machine Learning (cs.LG) #Machine Learning (stat.ML) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1907.11559
openalex publication_date 2019/07/26 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Deep generative models are rapidly gaining traction in medical imaging.\nNonetheless, most generative architectures struggle to capture the underlying\nprobability distributions of volumetric data, exhibit convergence problems, and\noffer no robust indices of model uncertainty. By comparison, the autoregressive\ngenerative model PixelCNN can be extended to volumetric data with relative\nease, it readily attempts to learn the true underlying probability distribution\nand it still admits a Bayesian reformulation that provides a principled\nframework for reasoning about model uncertainty. Our contributions in this\npaper are two fold: first, we extend PixelCNN to work with volumetric brain\nmagnetic resonance imaging data. Second, we show that reformulating this model\nto approximate a deep Gaussian process yields a measure of uncertainty that\nimproves the performance of semi-supervised learning, in particular\nclassification performance in settings where the proportion of labelled data is\nlow. We quantify this improvement across classification, regression, and\nsemantic segmentation tasks, training and testing on clinical magnetic\nresonance brain imaging data comprising T1-weighted and diffusion-weighted\nsequences.\n