Intelligent, Portable, and Cost-Effective Eye-Disease early warning system based on a developed Pupillary Dynamics Dataset with Machine Learning

Human pupil’s response dynamics to variable illumination stimuli may lead to the prediction of cephalic autonomous nervous system integrity. In this paper, the sensitivity of the multiple discrete wavelengths (red: 610 nm, green: 560 nm, blue: 450 nm, and ambient: $5000^{\circ}\mathrm{K})$ on the human subject’s eye has been measured as the input light was focused (transient state). Consequently, an exclusive pupillary dynamics dataset having a 15-second pupil’s dynamic response video has been developed for 66 Asian human subjects (47 males and 19 females) with an age variation from 18 to 48 years. The time has been recorded from the transient state of the pupil’s diameter until it is stabilized under a steady state. This unprocessed dataset has been efficiently enhanced in quality to detect the circumference of the pupil using video processing algorithms with binarization, adaptive thresholding, morphological processes, Hough transform, and Gaussian blur techniques. The dataset was used to train and test the machine learning (ML) model with the logistic regression (LR) approach. A prediction accuracy of 94% has been achieved to identify a healthy and normal functional eye. A repeatability of 90% has been obtained to predict the true-positive and truenegative test cases. Consequently, an intelligent, cost-effective, and portable eye-disease early warning system has been prototyped and presented in this work.