Novel Machine Learning Pipeline for Real-Time Oculometry

In modern medicine, there is a need for widely available and efficient oculometric systems. This is because oculometry provides a simple insight into the general motor status of the patient. A good example are neurodegenerative diseases (ND) such as Parkinson’s disease (PD) with bradykinesia as one of the effects causing a general slowing of movements, also visible in eye movements. PD develops secretly for many years before showing visible effects, but eye-tracking tests can reveal it sooner. That is why we decided to create a modern system for performing this type of tests based on neural network models, modern inference approach and algorithms of various types to help obtain reliable results. Our idea was to disconnect the software from the hardware and base the hardware requirements on consumer grade equipment, such as a web-camera. This step is necessary because modern telemedicine must necessarily be based on basic equipment available to patients in their households. Our results showed that even with 30 Hz frequency and standard reflexive-saccade (RS) test, our system was able to distinguish the results of healthy person from PD patient or young from old person This solution brings the ability to receive quantitative online data, in opposition to standard telemedicine allowing only for verbal and/or visual interactions with the patient. We hope that low hardware requirements will popularize automated oculometric tests in the context of age-related complications and ND diseases, which may contribute to earlier diagnoses and the collection of the appropriate amount of data to develop new means of preventing different diseases.