Uncertainty modelling in deep learning for safer neuroimage enhancement: Demonstration in diffusion MRI

• Published in: NeuroImage (Orlando, Fla.) Vol. 225; p. 117366
• Main Authors: Tanno, Ryutaro, Worrall, Daniel E., Kaden, Enrico, Ghosh, Aurobrata, Grussu, Francesco, Bizzi, Alberto, Sotiropoulos, Stamatios N., Criminisi, Antonio, Alexander, Daniel C.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 15.01.2021; Elsevier Limited; Elsevier
• Subjects: Anisotropy; Bayesian analysis; Brain - diagnostic imaging; Deep Learning; Diffusion Magnetic Resonance Imaging - methods; Diffusion MRI; Diffusion Tensor Imaging; Glioma; Humans; Image enhancement; Image Enhancement - methods; Image Processing, Computer-Assisted; Image synthesis; Interpretability; Magnetic resonance imaging; Mathematical models; Medical imaging; Multiple sclerosis; Neural networks; Neuroimaging; Neuroimaging - methods; Noise; Risk assessment; Robustness; Safety; Super-resolution; Tractography; Uncertainty; Uncertainty quantification; Deep learning; Neuroimaging; Super-resolution; Image synthesis; Interpretability; Tractography; Uncertainty quantification; Diffusion MRI; Robustness; Safety; Image enhancement
• ISSN: 1053-8119; 1095-9572
• DOI: 10.1016/j.neuroimage.2020.117366

•Proposes methods for modelling different types of uncertainty that arise in deep learning (DL) applications for image enhancement problems.•Demonstrates in dMRI super-resolution tasks that modelling uncertainty enhances the safety of DL-based enhancement system by bringing two categories of practical benefits:(1) “performance improvement”: e.g., the generalisation to out-of-distribution data, robustness to noise and outliers (Section 4.3)(2) “reliability assessment of prediction”: e.g., certification of performance based on uncertainty-thresholding (Section 4.4.1); detection of unfamiliar structures and understanding the sources of uncertainty (Section 4.4.2).•Provide a comprehensive set of experiments in a diverse set of datasets, which vary in demographics, scanner types, acquisition protocols or pathology.•The methods are in theory applicable to many other imaging modalities and data enhancement applications.•Codes will be available on Github. Deep learning (DL) has shown great potential in medical image enhancement problems, such as super-resolution or image synthesis. However, to date, most existing approaches are based on deterministic models, neglecting the presence of different sources of uncertainty in such problems. Here we introduce methods to characterise different components of uncertainty, and demonstrate the ideas using diffusion MRI super-resolution. Specifically, we propose to account for intrinsic uncertainty through a heteroscedastic noise model and for parameter uncertainty through approximate Bayesian inference, and integrate the two to quantify predictive uncertainty over the output image. Moreover, we introduce a method to propagate the predictive uncertainty on a multi-channelled image to derived scalar parameters, and separately quantify the effects of intrinsic and parameter uncertainty therein. The methods are evaluated for super-resolution of two different signal representations of diffusion MR images—Diffusion Tensor images and Mean Apparent Propagator MRI—and their derived quantities such as mean diffusivity and fractional anisotropy, on multiple datasets of both healthy and pathological human brains. Results highlight three key potential benefits of modelling uncertainty for improving the safety of DL-based image enhancement systems. Firstly, modelling uncertainty improves the predictive performance even when test data departs from training data (“out-of-distribution” datasets). Secondly, the predictive uncertainty highly correlates with reconstruction errors, and is therefore capable of detecting predictive “failures”. Results on both healthy subjects and patients with brain glioma or multiple sclerosis demonstrate that such an uncertainty measure enables subject-specific and voxel-wise risk assessment of the super-resolved images that can be accounted for in subsequent analysis. Thirdly, we show that the method for decomposing predictive uncertainty into its independent sources provides high-level “explanations” for the model performance by separately quantifying how much uncertainty arises from the inherent difficulty of the task or the limited training examples. The introduced concepts of uncertainty modelling extend naturally to many other imaging modalities and data enhancement applications.