On instabilities of deep learning in image reconstruction and the potential costs of AI

Abstract

Deep learning, due to its unprecedented success in tasks such as image classification, has emerged as a new tool in image reconstruction with potential to change the field. In this paper, we demonstrate a crucial phenomenon: Deep learning typically yields unstable methods for image reconstruction. The instabilities usually occur in several forms: 1) Certain tiny, almost undetectable perturbations, both in the image and sampling domain, may result in severe artefacts in the reconstruction; 2) a small structural change, for example, a tumor, may not be captured in the reconstructed image; and 3) (a counterintuitive type of instability) more samples may yield poorer performance. Our stability test with algorithms and easy-to-use software detects the instability phenomena. The test is aimed at researchers, to test their networks for instabilities, and for government agencies, such as the Food and Drug Administration (FDA), to secure safe use of deep learning methods.

Editorial for “Combined Denoising and Suppression of Transient Artifacts in Arterial Spin LabelingMRIUsing Deep Learning”

“…This concerning effect has also been noted in a recently published article 9 and could be indicative of misclassification of small pathological features as artifact, in addition to the inherent blurring of filter‐based techniques. Future work should be encouraged to adopt stability tests 9 to address this phenomenon in detail.…”

Solving inverse problems using data-driven models

“…It is known that adding white noise to the input helps against adversarial attacks for classifiers (Cohen, Rosenfeld and Kolter 2019). Moreover, in inverse problems one always has noisy data, so it remains unclear whether the computed perturbation in Antun et al (2019) actually acts as an adversarial example when noise is added.…”

Universal adversarial attacks on deep neural networks for medical image classification

“…Furthermore, properties inherent to image processing may cause misclassifications. For instance, DNNbased image reconstructions are often performed for purifying adversarial examples [28]; however, it causes image artifacts, resulting in misclassifications by DNNs [29]. It may be difficult to completely avoid security concerns caused by adversarial attacks.…”