Full Program »
One-Pixel Attacks Against Medical Imaging: A Conceptual Framework
This paper explores the applicability of one-pixel attacks against medical imaging. Successful attacks are threats that could cause mistrust towards artificial intelligence solutions and the healthcare system in general. Nowadays it is common to build artificial intelligence models to classify medical imaging modalities as either normal or as having indications of disease. One-pixel attack is made using an adversarial example, in which only one pixel of an image is changed so that it fools the classifying artificial intelligence model. We introduce the general idea of threats against medical systems, describe a conceptual framework that shows the idea of one-pixel attack applied to the medical imaging domain, and discuss the ramifications of this attack with future research topics.