Technology progress in medicine is providing more and more data, which is great. At the same time these data need to be processed and … understood. Take the Ocular Coherence Tomography (OCT). It allows an ophthalmologist to look at a patient’s retina with micro-metric resolution, both the surface of the retina and its structure, its layers. It takes a lot of experience to interpret the OCT results. Moorfield hospital in UK, one of the largest eye hospital in the world, every day handles around 1,000 OCT of patients referred by primary care doctors. This is overwhelming the capability of the hospital.
Researchers at Google have trained DeepMind having it (an algorithm) looking at 14,884 OCT retinal scans. The results are impressive: the DeepMind algorithm resulted in 94.5% correct referrals, that is au pair with the very best ophthalmologists (and there are not so many of them). More than that. It takes DeepMind 30″ to analyse an OCT and provide a referral (urgent, semi-urgent, routine, observation). The nice thing is that any primary doctor could connect to DeepMind and submit the OCT for analyses. Based on the response the patient can be referred to Moorfields if a specific follow up is required.
According to the researchers the result obtained can be compared to the development of a prototype that works really well. The next step is transforming the prototype into an industrial product that can be applied to the mass market.
This is just another example where AI can flank humans as a most effective tool but it is not intended to replace them.