Google AI predicted cancer diagnosis better than doctors
Tue 19 Jun 2018
A Google algorithm gave a better assessment of a cancer patient’s prognosis than doctors and hospital computers in a recent case, according to Bloomberg.
A female patient with late-stage breast cancer was admitted to a hospital, where she received a radiology scan and was seen by two doctors. The hospital’s systems, after reading her vital signs, decided that she had a 9.3% chance of death.
A few days later, the woman passed away. Doctors may have been better off paying more attention to Google’s prediction, which predicted a 19.9% chance of death.
It did this by using neural networks. These types of algorithms, which fall broadly under the umbrella of artificial intelligence, have been lauded for their healthcare uses for some time.
We’ve written about AI in cancer assessments on The Stack before, and a common thread seems to be that the ability of artificially intelligent systems to bring in masses of data and quickly assess it may have huge ramifications for patients.
And that’s what doctors and medical experts noted here. The algorithm was able to find ‘hard-to-reach’ data such as notes in PDFs or written on charts. By reading all of this data and quickly turning it into a prediction, these types of systems can assess survival chances, give prognoses and estimating the length of time a patient may have to stay in hospital.
These all have massive implications for healthcare, both in terms of improving a patient’s odds and experience, but also in administrative and organisational benefits, which could free up cash for better patient care.
Google’s AI advantages
According to Bloomberg, Google is well ahead of the game in terms of medical AI. Other notable commercially led AI projects, such as IBM Watson, have had success in areas like sport, but not in the high-stakes and complex world of medicine.
A big part of the reason why Google is so far ahead is because of its machine learning led approach. Most algorithms used in healthcare are written by hand, researchers told Bloomberg, but not Google’s. This means they are able to make much quicker progress.
With this sort of development, there is also a clear path to commercialisation. Once these algorithms can reliably be used in clinics and hospitals, Google’s dominance in this area could be consolidated.
It is, however, still “really early on” according to Google researchers. Patients with life-threatening diseases may hold out hope, though, if it continues to progress at this pace.