Google medical research

Google also uses AI (artificial intelligence) for medical research. A spectacular example of this is the 'detection of new systemic biomarkers in eye images'. In this study, a deep learning system is being trained to analyze external eye images for diabetes and elevated levels of glycated hemoglobin. Previously, photographs of eyes were not known to contain clues to these conditions. The great thing is that such photos can be taken with smartphones, which reduces the need for specialized equipment. Other possible biomarkers that indicate the status of various organs (e.g. kidney, blood and liver) on pictures of the external eye are still being sought.
The results of the study have even been published in the renowned journal Lancet Digital Health. The research is still in its early stages and can be compared to calculating the risk of some diseases using a questionnaire, but it has great potential.

Model development and training

To train the AI model, eye images were compared to lab test results for nine promising predictors of diseases including: liver disease (i.e., liver damage or bile duct obstruction), chronic kidney disease, anemia, and leucopenia, which affects the ability of the body to fight infections.
The AI model performed statistically better than the base model on eight of the nine diseases. And while the level of accuracy is probably still insufficient for diagnostic applications, it is comparable to other screening tools, such as mammography and diabetes prescreening. These tests are used to identify individuals who may benefit from additional research.

The eye photos

The photos of the eye were taken with a table camera with a headrest. But even at low resolution, the model outperforms the base model, even when the images are scaled down to 150x150 pixels, which is much lower than the resolution of the average smartphone camera.

The research not only shows the potential of eye images, it also shows that biomarkers on these images can provide a good prediction of a person's health status. Nevertheless, many steps still need to be taken to see if this technology can also help patients in the real world.
But the first, promising start is there. And ultimately this could be a godsend for patients who live far from a hospital.