Automatic Refractive Error Estimation Using Deep Learning-Based Analysis of Red Reflex Images

Abstract

Purpose/Background: We evaluate how a deep learning model can be applied to extract refractive error metrics from pupillary red reflex images taken by a low-cost handheld fundus camera. This could potentially provide a rapid and economical vision-screening method, allowing for early intervention to prevent myopic progression and reduce the socioeconomic burden associated with vision impairment in the later stages of life. Methods: Infrared and color images of pupillary crescents were extracted from eccentric photorefraction images of participants from Choithram Hospital in India and Dargaville Medical Center in New Zealand. The pre-processed images were then used to train different convolutional neural networks to predict refractive error in terms of spherical power and cylindrical power metrics. Results: The best-performing trained model achieved an overall accuracy of 75% for predicting spherical power using infrared images and a multiclass classifier. Conclusions: Even though the model's performance is not superior, the proposed method showed good usability of using red reflex images in estimating refractive error. Such an approach has never been experimented with before and can help guide researchers, especially when the future of eye care is moving towards highly portable and smartphone-based devices.

Keywords: fundus imaging; myopia; red reflex; refractive error.

Figure 1

IR and colour red reflex image capture using nun IR.

Figure 2

Images of pupils cropped from colour and IR images. Column 1 is the original image. Column 2 is cropped version.

Figure 3

nun IR illumination path [50].

Figure 4

Red reflex color images taken by nun IR using a fake eye, for which spherical power can be adjusted.

Figure 5

IR images with corresponding crescent types.

Figure 6

Pupils cropped automatically using MedicMind-AI portal.

Figure 7

Inception-v3 architecture.

Figure 8

EfficientNet architecture.

Figure 9

Spherical power distribution in the Dargaville dataset.

Figure 10

Crescent type categories (A–D).

Figure 11

Sphere vs. crescent type in the Choithram dataset.

Figure 12

Cylinder vs. crescent type.

Figure 13

Distribution of spherical power for 288 IR red reflex images.

Figure 14

Distribution after removing median 30% of IR images.

Figure 15

IR image before (left) and after (right) increasing contrast.

Figure 16

Combined color and IR images.

Figure 17

Myopic IR images.

Figure 18

Normal (left image) and myopic (second and third image) IR images.

Figure 19

Accuracy of techniques for IR images using EfficientNet (≥70% dark, ≥60% medium otherwise light).