Will your future be dark?
It might be. Unless pattern recognition saves your eyes.
Pattern recognition might sound pretty nerdy, but you come across it more often than you think. Shazam created an app to find out which song is playing in the club, Spotify predicts which other artists you might like and Google created an algorithm to identify cats in YouTube videos. But while it is cool that a computer can recognize a cat video, there might be more important possibilities for pattern recognition as well, such as in the medical sector.
The CHCF, the California HealthCare Foundation, is one of those organizations that decided to use pattern recognition for more important things. They created an application to detect a premature medical implication: diabetic retinopathy, a long-term complication of diabetes. The implication is caused by damage to the tiny blood vessels that support the retina. If left untreated, you will lose all your vision. This complication actually affects 80% of all patients who have had diabetes for 10 years or more (Kertes et al., 2007).
So how did they figure out how to detect this issue?
The CHCF organized a competition – with a $100,000 prize – on a website called Kaggle. This website has all sorts of competitions for some of the smartest people on the globe – statisticians and data scientists. The CHCF gave participants a database with thousands of images of healthy and affected retinas and let them figure out a solution.
In the end, one smart bloke called Benjamin Graham – who worked as a statistician at the University of Warwick – came up with an algorithm that identifies signs of diabetic retinopathy from an eye scan.
The big advantage of the algorithm is that it is faster, cheaper and more accurate than real doctors. Images are analysed instantly, instead of first having to be sent to a lab. This gives the advantage that there is less work involved, lowering the medical expenses involved. And while normally doctors only agree 84% of the time with each other on a diabetic retinopathy diagnosis, the algorithm agrees with a doctor’s opinion 85% of the time (The Economist, 2015) – so the algorithm can actually be more accurate than a human doctor. Jorge Cuadros, the CEO of Eyepacs, a company interested in using the algorithm, is intrigued by the high correlation between the algorithm and human experts. Even more so when there is a disagreement, sometimes the algorithm proves to be right, not the human doctor (Farr, 2015).
So does this mean the diagnosis will be conducted by computers now?
Even though the algorithm offers so many advantages, it will still take a long time before it has taken a place in clinical practice. Currently the solution is being held back by regulations, such as those from the FDA, the Food and Drug Administration. Fear is another obstacle that needs to be overcome, because who is going to take the blame when something goes wrong?
But in the end it will probably all work out. As pattern recognition software applied in medicine becomes better, institutions will have more incentives to bring the algorithms into the clinic.
The Economist,. (2015). Now there’s an app for that. Retrieved 6 October 2015, from http://www.economist.com/news/science-and-technology/21664943-computers-can-recognise-complication-diabetes-can-lead-blindness-now
Farr, C. (2015). This Robo Eye Doctor May Help Patients With Diabetes Keep Sight. KQED Future of You. Retrieved 6 October 2015, from http://ww2.kqed.org/futureofyou/2015/08/20/this-robo-eye-doctor-may-help-patients-with-diabetes-keep-sight/
Kertes PJ, Johnson TM, ed. (2007). Evidence Based Eye Care. Philadelphia, PA: Lippincott Williams & Wilkins.