With both internal and external validation, the results were very good. The machine managed to capture what are those unknown factors that help to differentiate the eyes of men and women. There was just some trouble getting the AI to distinguish the sexes if the eye image belonged to someone with foveal pathology. However, even with this small limitation, it is shown that deep learning can go even further than the complex rational mind of humans. This has multiple applications and implications; but, focusing on the health of the eye, it is undoubtedly a very positive thing.
Male or female eye
Physiologically, there are many differences between people who are biologically male or female. Some are inconsequential, while others must be taken into account both for the diagnosis of diseases and for the decision of which are the best treatments. For example, when choosing a drug dosage, it is important to keep in mind that men and women may not tolerate the same amount.
In the case of the eyes there are also differences. In fact, sex hormones play a fundamental role in mechanisms such as eye hydration and circulation. In addition, the development of diseases such as glaucoma, cataracts or uveitis can also be influenced by biological sex.
Unsplash
'Deep learning' to improve the diagnosis of ophthalmological diseases
The retina is the only tissue in which both neural and vascular tissue can be visualized simultaneously and non-invasively.
This allows detecting a multitude of pathologies of both types, only with the analysis of images of the fundus of the eye. For example, vascular tortuosity and arterial narrowing make it possible to detect possible cardiovascular diseases. Instead, alterations in the retinal cell layer help diagnose neurological disorders.
The retina is the only tissue in which both neural and vascular tissue can be visualized simultaneously and non-invasively.
That is why another team of researchers, whose results were published in Nature in 2021, carried out this new deep learning training with more than 80,000 images from the UK Biobank. They managed to replicate the same thing as in 2018, so it did not seem like a simple coincidence. They were facing a new application of artificial intelligence.
