Artificial intelligence – gamer glasses detect tumors or strokes – News


contents

A spin-off from the Inselspital in Bern uses virtual reality glasses as a medical diagnostic tool.

It has been known for a long time that a brain disease affects the movements of the eyes or the reactions of the pupils, says Dominik Senn, CEO of the startup MachineMD. But for doctors or experts, observing, measuring and evaluating the eyes is time-consuming and error-prone.

VR glasses: from the computer game to the clinic

In the future, this task will therefore be performed by virtual reality glasses and software that MachineMD is developing together with specialists from the Swiss Center for Electronics and Microtechnology (CSEM).

The VR glasses are high-quality commercial devices developed for computer games. Two screens show a visual stimulus while cameras monitor the reactions of the eyes and pupils.

In one of a total of eight tests, one eye is blinded by a bright area while the other eye remains dark. Cameras and software register the reactions of the pupils to the stimulus – possibly an indication that something is wrong with the signal processing in the brain. Other tests deal with the field of vision, how fast the eyes move or where they look.

Artificial intelligence analyzes line of sight

Understanding where someone is looking is a major challenge, explains Sareh Saeedi from the Swiss CSEM. She has been working on the digital analysis of viewing directions for a long time and has worked on the software for the glasses.

Unlike with a camera, you can hardly tell from the outside what the eye is really looking at – the physiological differences are too great. To solve the problem, the scientists at CSEM relied on artificial intelligence, so-called machine learning. They trained special software with thousands of images of eyes and information on which point in space the focus is on. In this way, the software independently learned to interpret the gaze correctly.

Synthetic data instead of human recordings

The preparation of the training recordings by hand is time-consuming and error-prone, says Andrea Dunbar, a specialist in artificial intelligence at CSEM. Instead of recordings of human eyes, images generated by a computer were used for the training, so-called synthetic data.

In addition to efficiency and precision, this procedure has other advantages: The computer can also generate any number of images of eyes with rare diseases and time-consuming clarifications relating to the protection of privacy are no longer necessary.

The first version of the VR glasses is scheduled to be launched in early 2024. The software developed by MachineMD can then generate a report that gives experts or doctors an indication if something is wrong. Later, the software should also be able to diagnose diseases such as multiple sclerosis or Alzheimer’s.

source site-72