Abstract
Purpose :
To develop an automated pupilometry system (SAP) to diagnosis of type 2 diabetes based on the pupillary reflex.
Methods :
A pupilometer with a lighting system (0 to 250 lux) was used, positioned 3 cm away from one eye and external lighting sealing. While the RGB LED lighting system (R: 623 G: 517 B: 466) offers a solution for a pupil response, an infrared camera captures it as images. A camera (Point Gray Firefly MV 0.3 MP Mono USB 2.0) operates at a wavelength of 850 nm. To evaluate the direct pupil reflex, the pupilometer was used to record videos during stimuli with red wavelengths (623) and blue (466) wavelengths, with a luminance of 250 cd / m2 and 1 second of duration after the patient was adapted to the dark for 10 minutes. The interval between stimuli was of 59 seconds. After a data capture, a data processing phase, data return declaration and data normalization were applied. In the last phase, a learning machine algorithm, called Random Forest, was applied to create the classification model of patients. The patients were classified in groups: Group 1 – Without Type 2 Diabetes, Group 2 – with Typpe 2 Diabetes. All patients underwent complete ophthalmologic consultation and macular Cirrus HD-OCT. Thus, the patients were according to the diagnosis of the type2 diabetes based on the American Diabetes Association. The study was approved by the Institutional Review Board CAAE: 23723213.0.0000.5083.All patients signed a written informed consent form for this research.
Results :
SAP was able to record, induce, and extract 96 pupil features. 31 volunteers were analyzed (16 in Group 1, 15 in Group 2), of which 22 were female volunteers (70.97%) and 9 were male volunteers (29.03%). A mean age of 60 year. As a result of the automated classification, Random Forest presented a result of 94.0% accuracy in the identification of diabetics type II was obtained.
Conclusions :
The proposal proved to be promising, noninvasive, objective and portable method of identifying the Type 2 Diabetes. Finally, the work reveals that pupillary reflex.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.