BoneXpert – A mature AI product
In April, Kicky van Leeuwen et al presented the paper Artificial intelligence in radiology: 100 commercially available products and their scientific evidence. It got quite some attention, because it documents what many have suspected, namely, that most AI products lack sufficient publications.
The first pie chart below, adapted from the paper, shows that only 36 of the 100 AI systems had publications.
These 36 products are then explored in the next three pie charts (so in these, 10 degrees represent one product), and we see that very few products have what could be considered a sufficient degree of maturation, e.g.:
- Only 12 of the 36 products were validated in more than 3 scanner/modality manufacturers
- Only 9 were validated in more than 20 centres
- Only 11 were validated in more than 5 countries
This suggests that only about 10% of the AI products are mature. BoneXpert is contained in all these categories (in accordance with Table S2 in the supplementary information).
Interestingly, the paper offers an explanation of this state of affairs. It is due to the fact that most of these products were developed quite recently and then swiftly put on the market (i.e. CE-marked). Also, the companies behind the products were founded not long ago. But it takes time to mature a product, and the field seems be haunted by impatience with getting on to market. Below is an excerpt from a large figure from the paper, showing the timelines of the eleven musculoskeletal products.