Detecting failure modes in image reconstructions with interval neural network uncertainty

Luis Oala,Maximilian März,Cosmas Heiß,Jan Macdonald,Wojciech Samek,Gitta Kutyniok

doi:10.1007/s11548-021-02482-2

Abstract

PurposeThe quantitative detection of failure modes is important for making deep neural networks reliable and usable at scale. We consider three examples for common failure modes in image reconstruction and demonstrate the potential of uncertainty quantification as a fine-grained alarm system.MethodsWe propose a deterministic, modular and lightweight approach called Interval Neural Network (INN) that produces fast and easy to interpret uncertainty scores for deep neural networks. Importantly, INNs can be constructed post hoc for already trained prediction networks. We compare it against state-of-the-art baseline methods (MCDrop, ProbOut).ResultsWe demonstrate on controlled, synthetic inverse problems the capacity of INNs to capture uncertainty due to noise as well as directional error information. On a real-world inverse problem with human CT scans, we can show that INNs produce uncertainty scores which improve the detection of all considered failure modes compared to the baseline methods.ConclusionInterval Neural Networks offer a promising tool to expose weaknesses of deep image reconstruction models and ultimately make them more reliable. The fact that they can be applied post hoc to equip already trained deep neural network models with uncertainty scores makes them particularly interesting for deployment.

Highlights

As we demonstrate, existing uncertainty quantification (UQ) approaches come with limitations regarding their capacity to detect failure modes or their post hoc applicability to trained deep learning models
U(z) should be correlated with the component-wise error |x − Φ(z)|. We evaluate this for three different failure modes [7] that can arise during inference (see “Experiment B (i): general prediction error detection”
Our main contributions can be summarized as follows: We present a deterministic, modular and fast UQ-method for deep neural networks (DNNs), called Interval Neural Networks (INN)

Summary

Introduction

The reconstruction of unknown signals from indirect measurements plays an important role in many applications, including medical imaging [2,14]. Such tasks are modeled as finite-dimensional linear inverse problems y = Ax + η, (1). Important examples include magnetic resonance imaging and computed tomography, where A is a subsampled discrete Fourier or Radon transform, respectively. Solving the inverse problem (1) requires computing an approximate reconstruction of x from the observed measurements y. E.g., based on sparse regularization models, constitute the state of the art for solving (1) in many cases and are backed by theoretical guarantees [8]. Data-driven deep learning methods are increasingly gaining attention and are repeatedly able to International Journal of Computer Assisted Radiology and Surgery outperform traditional solvers in terms of empirical reconstruction performance or speed, see for example [2]

Objectives

Methods

Conclusion