Machine learning predicting myopic regression after corneal refractive surgery using preoperative data and fundus photography
Description
The aim of this study is to develop deep learning models that can identify high-risk patients for refractive regression based on preoperative data and fundus photography. This retrospective study was approved by the Institutional Review Board of Korean National Institute for Bioethics Policy (KoNIBP, 2019-1685-003), which waived the requirement for informed consent because the data was fully de-identified to protect patient confidentiality. The collection of external validation samples in each cohort was approved by the ethics board at the Aerospace Medical Center. This study adhered to the tenets of the Declaration of Helsinki. This study was performed using data from the B&VIIT Eye Center in Seoul, South Korea. From 2015 to 2016, a total of 12,912 patients aged 18-45 years, who underwent vision-correction, were screened to identify corneal refractive surgery cohorts for analysis. The study subjects were required to undergo LASIK, LASEK, or SMILE at least one eye and to have the ocular measurements in the fourth year after surgery (for a period of more than 48 months and less than 60 months). Of these, 11,821 patients were excluded due to no follow-up visit, other surgery such as phakic intraocular lens implantation, and histories of corneal disorder, retinal disorder, and glaucoma. Finally, the study population consisted of 2,009 eyes of 1,091 patients. The datasets utilized in this study are not publicly available because of reasonable privacy and security concerns. Instead, a sample of anonymized fundus photography data and clinical measurements were released. Note that this was not the exact data used in the research, but was a cleaned-up reproduction of our study's key insight. The sample imaging and clinical data used in this study have been deposited in this repository (randomly sampled). In this repository, there are 150 fundus photos with myopic progression after refractive surgury and 180 fundus photos with no myopic progression (myopic regression.zip file). And there are a total of 165 ocular measurements (55 LASIK, 55 LASEK, and 55 SMILE, sample.cvs file) . The two datasets are not coupled for security reasons. The ResNet50file_5foldcv.zip file contains the ResNet50 models trained in the 5-fold cross-validation process to predict myopic regression after refractive surgery.
Files
Steps to reproduce
The detailed underlying architecture is copyrighted by the B&VIIT Eye Center and will not be available to public.