DeepMind Optometrist

Google DeepMind artificial intelligence will learn to diagnose diseases by analyzing 1 million retinal images

DailyTechInfo based on Ars Technica: Google's DeepMind AI to use 1 million NHS eye scans to spot diseases earlier

Google DeepMind, a division of the company engaged in the development of artificial intelligence systems, announced an agreement with the ophthalmological clinic NHS Moorfields Eye Hospital on the training of artificial intelligence systems for the early diagnosis of diseases. This program is designed for five years, during which the machine will be "fed" a total of about a million eye images, which are in the database of the Moorfields clinic. And the ultimate goal of this program is to develop an express diagnostic system that will be able to quickly analyze retinal images and diagnose various diseases at the earliest stage of their occurrence based on the data obtained.

The main causes of vision loss are diabetic retinopathy and age-related degeneration (yellow spot), in 98 percent of cases of diabetes, vision loss can be prevented if the disease is detected at an early stage, leaving doctors time to carry out proper treatment.

The Google DeepMind project will process two types of eye images, traditional fundus images, which are essentially images of the retina, and images obtained using optical coherence tomography (optical coherence tomography, OCT), which give images of a kind of "cross-section" of the retina and which are more informative from a medical point of view vision.

Retinal OCT (drawing from Wikimedia Commons – VM)

However, OCT images are very complex and their analysis can only be performed by medical personnel who have undergone specialized training. As a result, there are delays between the receipt of images, their analysis and diagnosis. And, despite a lot of attempts, the most powerful computer analytical tools cannot solve the problem of automating the analysis of OCT images.

The Google DeepMind artificial intelligence system, built on the principles of deep learning and self-learning and implemented in the form of artificial neural networks, will be able to independently learn to find signs of emerging diseases, primarily diabetic retinopathy. To do this, she, as usual, will need some initial push, a process of preliminary training, which will be conducted by qualified doctors. Further, the system will continue to train independently, and doctors will only monitor the work of artificial intelligence, making suggestions and making adjustments in case of such a need.

Naturally, at the beginning of the work, the artificial intelligence system will make a lot of mistakes. But further self-study, supported by help from doctors, will lead to a constant decrease in the percentage of erroneous diagnoses. And after a while, the system will catch up with the quality of work, and then surpass even the most experienced ophthalmologists.

It should be noted that this program is not the first time Google DeepMind and the National Health Service of Great Britain (NHS) have collaborated. But their last joint project ended in a loud scandal, because within the framework of the project, access to the complete medical histories of 1.6 million patients was opened, which is a gross violation of the principle of medical secrecy. During the new project, no such mistake will be made, all the images that will be processed by the artificial intelligence system will be completely anonymous, except for those that the patients themselves will consent to use.

Portal "Eternal youth" http://vechnayamolodost.ru  18.07.2016