The Achilles heel of Artificial Intelligence (AI) applied to medicine is not algorithms, but access to structured and homogeneous data.
A little more than 5 years after the arrival of Xiaoyi, the first robot capable of passing the medical qualification exam in China, AI applied to medicine is still “limping” and the cause, observes Andrea Cavalli, scientific deputy director of the Italian Institute of Technology (Iit), is the lack of a common platform between the medical and IT worlds.
The theme on the unfulfilled objectives of the application of AI in the medical field is taken up again by two recent articles published in the British Medical Journal (Bmj), in which we recall how in 2016 one of the fathers of deep learning techniques, Geoffrey Hinton, stated: “We should stop training radiologists. It is clear that within five years deep learning will do better than radiologists”.
Over the years, AIs have made great progress and made important contributions in the medical field, but today they are tools that certainly cannot replace human activity.
“They are tools that help a lot, which offer great added value, but certainly Hinton’s phrase is currently not reflected in medical practice”, commented Cavalli.
It has not been technological improvements that have limited the growth of AI in recent years, but bureaucratic issues.
AIs are very powerful tools capable of analyzing and extracting important information from data, but if these are scarce or difficult to analyze, AIs cannot work: if they have “junk” data at stake, they produce “junk” results.
“To date, health data is the Achilles’ heel – continued Cavalli – just look at the electronic health records. The data is poorly structured and above all too heterogeneous, we have a healthcare system which should be the source of the data too fragmented. But this is not just an Italian problem: it is the same in the rest of the world”.
Without data, AIs have no fuel, therefore, to achieve real progress there is only one solution: “Data, data, data”, concluded Cavalli.
[email protected]
© Copyright 2023 Freedom Editorial