An algorithm has been created for diagnosing a tendency to the development of cardiovascular diseases, using the results of ophthalmoscopy - examination of the fundus.
Experts from the research organization Verily Life Sciences have developed an algorithm that can assess the likelihood of developing cardiovascular disease by analyzing fundus images. Images of the eyes of 284,335 patients were used to train the program. The development article was published in the journal Nature Biomedical Engineering.
The fundus of the eye is examined using an ophthalmoscope - a device with special magnifying lenses. Thus, it is possible to assess the condition of the retina and many vessels of the fundus. Ophthalmoscopy is one of the most important methods of eye examination; it is used not only for visual impairments, but also in the diagnosis of diabetes and hypertension. Based on the examination data, conclusions can be drawn about the patient's age, blood pressure indicators, and a tendency to bad habits. All these signs are important in the diagnosis of predisposition to cardiovascular diseases.
The creators of the new algorithm used deep learning technologies. At the disposal of the program were images of the fundus of 284 335 patients, whose average age was about 55 years. The algorithm also had access to basic information about patients: their age, gender, ethnicity, body mass index (ratio of height to weight), blood pressure and glycated hemoglobin indicators - the latter indicator is usually used to assess the effectiveness of diabetes treatment. In addition, patients indicated whether they smoked and reported diagnosed cardiovascular disease.
To assess the accuracy of the method, two datasets with eye images of 12,026 and 999 patients, respectively, were used. The algorithm was supposed to determine if a person developed cardiovascular disease within five years after the ophthalmologist took the x-ray. The program was able to answer correctly in 70% of cases. This result is comparable to the effectiveness of the popular SCORE (Systematic COronary Risk Evaluation) scale, according to which the risk of disease is detected on average in 72% of cases. However, SCORE requires the patient's blood test results.
The new development must go through several more stages of effectiveness verification. The creators of the algorithm emphasize that artificial intelligence will not replace a doctor, but will become an additional diagnostic tool.