Have you noticed a whitish discolouration of your child’s pupil in dim lighting conditions, from certain angles, or in place of a redeye in flash photos? The child might have leukocoria, which can be caused by one of over a dozen kinds of eye diseases, including congenital cataract (60%), retinoblastoma (18%), retinal detachment (4.2%), persistent fetal vasculature (4.2%), and Coats disease (4.2%). In children under the age of five, one of the most common causes is retinoblastoma, universally fatal if left untreated. The incidence of this alone is 1 in 15,000 live births and leukocoria shows up in more than half the cases.

We built an app to help with prescreening at home. Our metrics are based on the results of research reported in this paper.

App Screenshot

This app works totally offline (except for a donation button to charity). There are no ads and it doesn’t phone home, so your data is completely private.

What we did

  1. Slice out the centers of the pupils in the leukocoria section of Figure 3 (Examples of Cropped Leukocoric and Non-Leukocoric Pupils) and batch process them to give each one an average color over the area.
  2. Assign each slice a value based on proximity of its average color to the area of positives as plotted in graphs conveniently given in Figure 10 (Saturation-Value Scale for Quantifying Leukocoria in Photographs of Children with Retinoblastoma.).
  3. Plotting them showed that they were fairly bimodal. We developed a metric based on this.
  4. Built an Android app that implemented our metric.
  5. Tested images from Google Image Search.

Current state

We are happy to report that this app has accurately categorized images that were inconclusive to humans. And by humans, I mean our team and concerned parents on the internet who posted photos before getting a formal diagnosis. In that regard, this weekend hackathon project was a success. There is still much to do, however.

Note also that the PayPal Donate button is in sandbox mode and thus will not operate.

Future work

  • We need more data. It is unclear how accurate this is and what the incidence of false positive/negative is.
  • We need to be able to batch process a lot of photos, which means using something like OpenCV to select the pupil area automatically for scanning.
  • We can build a service that will scan photos on social networking sites and alert people of possible cause for seeing an eye doctor. This would be opt-in only.

For more information

  1. Abdolvahabi A, Taylor BW, Holden RL, Shaw EV, Kentsis A, et al. (2013) Colorimetric and Longitudinal Analysis of Leukocoria in Recreational Photographs of Children with Retinoblastoma. PLoS ONE 8(10): e76677. doi:10.1371/journal.pone.0076677
  2. Haider S, Qureshi W, Ali A. Leukocoria in children. J Pediatr Ophthalmol Strabismus. 2008;45(3):179–80.
  3. Seregard S, Lundell G, Svedberg H, Kivelä T. Incidence of retinoblastoma from 1958 to 1998 in Northern Europe: advantages of birth cohort analysis. Ophthalmology. 2004;111(6):1228–32.
  4. Abramson DH, Frank CM, Susman M, Whalen MP, Dunkel IJ, Boyd NW., 3rd Presenting signs of retinoblastoma. J Pediatr. 1998;132(3 Pt 1):505–8.
  5. Know the Glow(Children’s Hospital, Los Angeles)

Acknowledgements

The white eye icon is derived from this icon by Liz Aragon and is redistributed under the Creative Commons—Attribution-NonCommercial-ShareAlike license.

The figure we used to derive our metric is taken from doi:10.1371/journal.pone.0076677 and is redistributed under the Creative Commons—Attribution license.

The image selection and crop functionality is based on this tutorial and its sample code.

Categories: Projects

Leave a Reply