Scientists from Ural Federal University (UrFU), together with an international group of researchers, have developed a method for early diagnosis of neurological diseases using retinal images. The technology is based on artificial intelligence and has already shown high efficiency in identifying signs of ADHD, autism, and Parkinson's disease.
The method is based on the analysis of electroretinography, a non-invasive test that records the retina's response to light. The algorithm, created by specialists from the "Artificial Intelligence" center of UrFU, is capable not only of classifying results as "normal" or "pathology" but also of highlighting specific areas of the signal associated with abnormalities. This is achieved using explainable AI - an approach that allows the doctor to understand the logic of the model and use it as a decision support tool.
Our algorithms are computationally simpler than neural networks; they require less data and resources but provide good accuracy. This approach helps doctors quickly and inexpensively check the probability of diseases
The study was conducted on a database collected by an international consortium led by Professor Paul Constable of Flinders University (Australia). The scientists applied time series classification methods and the SHAP library, based on game theory, to explain the models' predictions. This increased the transparency of the algorithm and trust in the results.
In the future, the developers plan to adapt the technology for diagnosing retinal diseases, including glaucoma and congenital night blindness, as well as for an expanded range of neurodegenerative disorders.
Early detection of Parkinson's disease or autism significantly increases the chances of successful therapy and treatment adjustment. For Russian clinics, the method could become an accessible tool that reduces the burden on doctors and speeds up diagnosis.