On instabilities of deep learning in image reconstruction - Does AI come at a cost?
This addresses a critical safety issue for medical imaging and regulatory bodies like the FDA, highlighting a fundamental flaw in applying deep learning to reconstruction tasks.
The paper demonstrates that deep learning methods for image reconstruction are often unstable, where tiny perturbations or structural changes can cause severe artifacts or missed features, and more samples can paradoxically degrade performance.
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 unstablemethods for image reconstruction. The instabilities usually occur in several forms: (1) 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 tumour, may not be captured in the reconstructed image and (3) (a counterintuitive type of instability) more samples may yield poorer performance. Our new 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.