Revolution in medicine: AI improves histopathological diagnoses!

Revolution in medicine: AI improves histopathological diagnoses!

Researchers at the University Medical Center Mainz and the Technical University of Dresden have identified a potential vulnerability in popular AI models used in medical image processing. The study, titled “Incidental Prompt Injections on Vision–Language Models in Real-Life Histopathology,” examines the influence of text information on the analysis of medical image data. The work was published in NEJM AI, 2(6), AIcs2500078, and is carried out by a group led by the authors Clusmann, J., Schulz, S. J. K. and other researchers. These results could have significant implications for the use of AI in histopathological diagnostics.

The study's unique methodology describes how textual input can be implanted into image analysis systems to influence their results. This could potentially jeopardize medical decision-making. The contact for further information leads to PD Dr. Sebastian Försch from the University Medical Center Mainz and Prof. Dr. Jakob Nikolas Kather from TU Dresden, both of whom are considered leading experts in this area.

Clinical relevance of AI in pathology

As part of AI research, various approaches for use in medical image processing were presented. Huang et al. (2023) how AI identifies features in histopathology images associated with responses to neoadjuvant chemotherapy in breast cancer. This complements the findings of Campanella et al. (2019) who demonstrated computational pathology using weakly supervised deep learning methods.

These developments make it clear that AI is becoming increasingly important in digital pathology. The ALBRT system for predicting cell composition in histology images shows how AI services can be used to improve diagnostics. Such technologies could significantly increase diagnostic efficiency and patient safety, a topic addressed by Singh and Graber (2015).

The influence of data networking on medicine

Digitalization leads to enormous amounts of data, the effective use of which is already visible in the so-called Industry 4.0. In medicine, taking medical and non-medical data into account can optimize decision-making processes and individualize therapies. Big data and artificial intelligence are key terms here. These technologies not only facilitate clinical decision making, but also chronic disease monitoring and hospital data management.

In medicine, the rapid analysis of large amounts of data using AI opens up new perspectives. Applications range from medical image processing and diagnostics to robot-assisted surgery. It is clear that innovative approaches to the trustworthiness and validation of AI methods are essential, especially in the sensitive domain of medical diagnostics.

In summary, it can be said that the research at Mainz University Medicine and TU Dresden not only shows the fundamental challenges of AI in pathology, but also outlines its immense potential for the future of medical diagnostics. In this context, continuous discussion and further development of technologies is essential to ensure the safety and effectiveness of healthcare. Press contacts Barbara Reinke from the Mainz University Medical Center and Anja Stübner from the TU Dresden are available for further information.