By Amruth Chinnappa
Google, along with its biotech subsidiary company Verily, has made cardiovascular diagnoses easier. They have developed an Artificial Intelligence software which can predict heart diseases by conducting a retinal scan. The task involves the application of Deep Neural Networks (DNNs) and is a non-invasive method to determine a persons likelihood of experiencing a heart attack.
A window to the soul
The study was published in Nature Biomedical Engineering on the conceptual linkage between retinal vessels and the risk of cardiovascular ailments. Google has managed to quantify this likelihood with its software. Blood vessels reveal a lot about the body and are usually accessible through invasive practices. The vessels present on the retina, however, are an exception to this as they are visible by a scan through the pupils. The rear interior wall of the eye, the Fundus, is full of blood vessels and they reflect the bodys overall health. Medical practitioners are not newcomers to this technique as it has been used to check for Glaucoma, cholesterol, and other ailments.
Deep Neural Networks and heat maps
“Pattern recognition and making use of images is one of the best areas for AI right now”, says Harlan M. Krumholz, a professor of medicine (cardiology) and director of Yale’s Center for Outcomes Research and Evaluation. The technique was used to generate a heat map which graphically represented data in an image correlating to a particular risk factor. Professor Krumholz further believes that pattern-recognition could generate information from sensors and other devices to improve physical examination, the understanding of diseases and their treatment methods. The researchers in this study trained a neural network with retinal scan data from 284,335 patients. The algorithms involved in the neural network were programmed to detect minuscule differences in the blood vessels. This was then used to make an estimate of factors like age, smoking and blood pressure, which could influence a heart attack or stroke.
Testing positive
The information resulting from the data set was validated on two different groups of patients numbering 12,026 and 999 people, part of whom had heart attacks within five years of their retinal screening. When the images of people who had a heart attack were shown alongside those who did not, the algorithm identified the patient at a higher risk nearly 70 percent of the time. This was a bit less accurate compared to the regularly used, invasive SCORE method which is correct 72 percent of the time. According to the researchers, the algorithm is slated to improve with a higher data set as an input. It was also claimed to have tremendous potential in the future, acting as the first line of diagnostic care.
“This may be a rapid way for people to screen for risk,” Dr Harlan Krumholz, a cardiologist at Yale University.
A different perspective
AI can use the data for a particular clinical reason and is deriving much more information out of it. This could point toward an entirely new train of studies, different from existing ones. Artificial Intelligence and the use of Deep Neural Networks has crept into use everywhere, from using facial features to determine homosexuality to nearly deciphering a persons thoughts. Although clinical use of the technology by Google is quite a long way from happening, the next decade is sure to change perspectives about humankinds future.
Featured Image Source: Pixabay