Fig. 1: False positive and False negative examples after adding artefacts to images.
From: Understanding the robustness of vision-language models to medical image artefacts
a An example of a false positive case. Vision-Language models (VLMs) correctly classify normal cases when analysing original unaltered images but misclassify the same cases as diseased when random motion is introduced. b An example of a false negative case. VLMs correctly identify abnormal cases in original unaltered images but misclassify them as normal after random noise is introduced. More misdetection examples with full VLMs’ responses shown in Supplementary Data 1.
