MACHINE LEARNING PREDICTION OF EYEGLASS LENS POWER USING SIMPLE MOBILE CAMERA

To develop a machine learning machine learning based model that can predict the dioptric power of lenses used in eye glasses using a simple mobile camera. 

 Department of Ophthalmology, The University of Jordan, Amman, Jordan.

Data acquisition performed through a pre-designed system to capture images using different lens powers and different settings. The system comprises a specifically designed chart printed on an A4 paper to be imaged, a trial lens and a camera holders that can be adjusted manually, and a mobile camera (iPhone X). We used the camera to image the chart through different trial lenses, where we first neutralized the autofocus function on mobile camera. Images were captured from different distances and in different illumination settings using spherical trial lenses. Spherical lenses used differed by 0.5 diopters. The generated images underwent a pre-processing step. After that, we designed a machine learning model using k-nearest neighbors algorithm (k-NN) and reported its accuracy and the error in diopteric power

A total of 175 images were captured, pre-processed and fed into the machine learning model. The dioptric power of lenses used ranged from +3 diopters to -8 diopters, with a median power of -3. The ground truth predicted was the power of trial lens used to capture the image. The model was able to predict the power of lenses used with 95% accuracy. The dioptric power for prediction error ranged from 0.5 diopters to 2 diopters.

This study represent the development of the first artificial intelligence model that can predict the power of spherical lenses using simple mobile camera. Such model will provide important information for ophthalmologists and optician performing a tele-medical consultation or consultation in remote areas where no lensometers available.