Get started with AI right away: simple, practical, affordable
In healthcare it is about patients, not demos. So here is what research on AI in diagnosis and screening shows, where things are still uncertain and where AI already helps without putting patients at risk.
Imaging: lung cancer
In 2019 a team including researchers from Google published a model in Nature Medicine that estimated the risk of lung cancer from low-dose CT scans. Without a previous scan for comparison, the model did better than six radiologists, with fewer false positive and fewer false negative results. With a previous scan, it performed on a par with them. Read the summary of the study.
This was research on existing scans. It shows what is possible, but it does not prove that patients benefit in everyday practice.
Breast cancer screening
In Sweden, the MASAI study compared AI-supported reading of mammograms with standard double reading by two radiologists in about 80,000 women. With AI, 244 cancers were found against 203 with the standard approach, and radiologists' reading workload fell by 44.3 per cent. The researchers conclude that AI-supported screening is safe. See the article in The Lancet Oncology.
That was an initial safety analysis. Whether women ultimately benefit had yet to be shown. An accompanying comment in the same journal spoke of breathtaking results and a word of caution.
A warning: IBM Watson for Oncology
Not every promise comes true. MD Anderson Cancer Center in Texas worked with IBM Watson on a system for cancer treatment advice. The project cost about 62 million dollars and ended in 2016. According to an audit, the system was not in clinical use at that point. See this overview from Becker's Hospital Review. IBM sold its Watson Health business to Francisco Partners in 2022 (see Merative).
The lesson: a well-known name and a polished demo are not evidence. Ask for research on real patients, carried out by people other than the supplier.
Administration and records
A large share of care time goes on administration. Tools that summarise conversations or make records searchable can ease that without making decisions about the patient. The care provider checks the result and stays responsible.
Health data is special category personal data under the GDPR. Before you start, know where the data ends up and who can access it.
Rules
AI that counts as a medical device falls under the European rules for medical devices (MDR). The AI Act comes on top of that. A legal explanation is in this article from the University of Amsterdam.
Ready to explore AI applications in your own field? Take the first step by joining one of our trainings and workshops or in-depth sessions to go deeper into AI. See our upcoming events for more information.