Artificial intelligence is already capable of analyzing medical images such as these histopathological images for cancer diagnosis. However, as soon as text information is integrated, its accuracy suffers significantly. This tells why.
Given that commercial AI models often do not yet achieve the accuracy required for clinical application, researchers in Dresden have discovered that if text is integrated into images, it can negatively influence the judgment of AI models. This was discovered by researchers from Mainz and Dresden. This can lead to problems, especially when analyzing medical image data.
See more an literature source here: https://blog.bimpress.de/medtech-moegliche-schwachstelle-in-beliebten-ki-modellen-gefunden/