Artificial intelligence (AI) has great potential to improve medical thinking, and AI methods often show more comparable or superior performance than reviews by medical professionals. However, the presentation at the Society of Nuclear Medicine and Molecular Imaging (SNMMI) Annual Meeting highlighted significant risks to AI models, which engineers and users should consider. In particular, presenters highlighted data threats and data fraud.
The authors conducted a review of threats and mitigation strategies to highlight the importance of data security in AI clinical thinking efforts. Among the concerns are productive adversarial networks (GANs): “unregulated neural networks, competing to produce new models in a given training sample.”
In the case of photography, the photographs produced are inseparable from the original model. GANs used to create in-depth fake photos and videos. Whether intentionally or not, the process can be manipulated.
Such deception can have far-reaching consequences, with proponents claiming, such as forged images, delays in obtaining problematic images because they appear to be genuine, improper treatment effects if the decision is based on altered images, and financial implications.
They provided a few suggestions for reducing these risks, including image authentication in addition to AI verification, verified image detection watermarks, and encryption across all categories (as images move between scans, storage, workspaces, and more). They also encourage AI developers to educate AI users about issues of trust and trust.
Researchers beleive that effective data security strategies are needed to prevent data compromise while promoting the development of medical imaging technology and patient care through AI methods. It is important that health care professionals are informed of the risks in order to balance patient care obligations with the obligation to control data integrity.
