Article Text

Download PDFPDF
Prediction of visual outcomes by an artificial neural network following intravitreal injection and laser therapy for retinopathy of prematurity
  1. Ching-Yen Huang1,
  2. Ren-Jieh Kuo2,
  3. Cheng-Han Li2,
  4. Daniel S Ting3,
  5. Eugene Yu-Chuan Kang1,
  6. Chi-Chun Lai1,
  7. Hsiao-Jung Tseng4,
  8. Lan-Yan Yang4,
  9. Wei-Chi Wu1
  1. 1Department of Ophthalmology, Chang Gung Memorial Hospital and College of Medicine, Chang Gung University, Taoyuan, Taiwan
  2. 2Department of Industrial Management, National Taiwan University of Science and Technology, Taipei, Taiwan
  3. 3Vitreo-Retinal Department, Singapore National Eye Center, Singapore, Singapore
  4. 4Biostatistics Unit, Clinical Trial Center, Chang Gung Memorial Hospital, Taoyuan, Taiwan
  1. Correspondence to Dr Wei-Chi Wu, Department of Ophthalmology, Chang Gung Memorial Hospital, Taoyuan 33305, Taiwan; weichi666{at}


Aims To construct a program to predict the visual acuity (VA), best corrected VA (BCVA) and spherical equivalent (SE) of patients with retinopathy of prematurity (ROP) from 3 to 12 years old after intravitreal injection (IVI) of anti-vascular endothelial growth factor and/or laser photocoagulation treatment.

Methods This retrospective study employed a feedforward artificial neural network with an error backpropagation learning algorithm to predict visual outcomes based on patient birth data, treatment received and age at follow-up. Patients were divided into two groups based on prior treatments. The main outcome measures were the difference between the predicted and actual values of visual outcomes. These were analysed using the normalised root mean square error (RMSE). Two-way repeated measures analysis of variance was used to compare the predictive accuracy by this algorithm.

Results A total of 60 ROP infants with prior treatments were included. In the IVI group, the normalised average RMSE for VA, BCVA, and SE was 0.272, 0.185 and 0.131, respectively. In the laser group, the normalised average RMSE for VA, BCVA and SE was 0.190, 0.250 and 0.104, respectively. This result shows that better predictive power was obtained for SE than for VA or BCVA in both the IVI and laser groups (p<0.001). In addition, the algorithm performed slightly better in predicting visual outcomes in the laser group (p<0.001).

Conclusions This algorithm offers acceptable power for predicting visual outcomes in patients with ROP with prior treatment. Predictions of SE were more precise than predictions of for VA and BCVA in both groups.

  • retina
  • vision

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


  • Contributors Study concept and design: CYH and WCW. Construction of the neural network: CHL and RJK. Statistical analysis: HJT and LYY. Drafting of the manuscript: CYH. Critical review of the manuscript: DST, EYCK, CCL and WCW. WCW had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

  • Funding This study was supported by the Chang Gung Memorial Hospital Research Grants (CMRPG3I0071-3 and CMRPG3G30581-3) and the Ministry of Science and Technology Research Grants (MOST 106-2314-B-182A-040-MY3).

  • Disclaimer The sponsors had no role in the design or conduct of this research.

  • Competing interests None declared.

  • Patient consent for publication Not required.

  • Ethics approval This study was approved by the Institutional Review Board (IRB) of Chang Gung Memorial Hospital, Linkou, Taiwan (IRB number: 201801801B0C601).

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data availability statement Data are available upon request.