In June, 2019, the research paper of B&VIIT AI R&D was published in the npj Digital Medicine under the name of Nature Medicine for the first time in the ophthalmology field in Korea.
Recently, it has become more important to screen candidates that undergo corneal refractive surgery to prevent complications. Until now, there is still no definitive screening method to confront the possibility of a misdiagnosis. We evaluate the possibilities of machine learning as a clinical decision support to determine the suitability to corneal refractive surgery. A machine learning architecture was built with the aim of identifying candidates combining the large multi-instrument data from patients and clinical decisions of highly experienced experts. Five heterogeneous algorithms were used to predict candidates for surgery. Subsequently, an ensemble classifier was developed to improve the performance. Training (10,561 subjects) and internal validation (2640 subjects) were conducted using subjects who had visited between 2016 and 2017. External validation (5279 subjects) was performed using subjects who had visited in 2018. The best model, i.e., the ensemble classifier, had a high prediction performance with the area under the receiver operating characteristic curves of 0.983 (95% CI, 0.977–0.987) and 0.972 (95% CI, 0.967–0.976) when tested in the internal and external validation set, respectively. The machine learning models were statistically superior to classic methods including the percentage of tissue ablated and the Randleman ectatic score. Our model was able to correctly reclassify a patient with postoperative ectasia as an ectasia-risk group. Machine learning algorithms using a wide range of preoperative information achieved a comparable performance to screen candidates for corneal refractive surgery. An automated machine learning analysis of preoperative data can provide a safe and reliable clinical decision for refractive surgery.
Recently, refractive surgery has produced excellent visual outcomes, and the number of refractive surgeries has grown.1 It has now become more important for the refractive surgeon to select candidates to undergo corneal refractive surgery in order to avoid complications.2 In order to minimize complications after surgery, the surgeon has to accurately examine the patient’s eyes to preoperatively identify cases with a likely poor outcome.
There are complicated relationships between optical parameters such as myopic level, pupil size, corneal radius, and ablation zone.3 When a clinician considers the optical parameters to improve visual quality, the preoperative corneal radius and sphericity were used in a calculation formula to obtain the postoperative corneal curvature.4 Age and refraction should also be considered as predictors of refractive stability after surgery.5 Because surgeons may find it hard to calculate all nonlinear relationships of optical variables to minimize the complication of each patient, the clinical decision was made based on the surgeon’s experience.
Ocular imaging technology has evolved in recent years to address candidacy issues in the corneal refractive surgery.6 A complete preoperative examination has to be performed, and the refractive surgeon should review all examination results before recommending a procedure. This can be a time-consuming process, and it is possible to overlook a sign of surgery contraindications. This is even more likely given the increasing workload for the refractive surgeon with the rise in population seeking refractive surgery. Up to now, there is still no definitive screening method to confront the possibility of a misdiagnosis.
Machine learning, which is an area of artificial intelligence research, has become popular in clinical medicine because of its ability to handle big data and to classify cases with high accuracy.7 Support vector machines (SVM), random forests (RF), artificial neural networks (ANN), AdaBoost, and least absolute shrinkage and selection operator (LASSO) are widely used approaches in machine learning.8 These techniques have been applied to many tasks in medicine and bioinformatics to select informative variables and predicting diagnoses more accurately.9 The current machine learning technique classified Pentacam-based corneal data with good performance for keratoconus diagnosis.10 A random forest model using Pentacam measurement data showed the good diagnostic accuracy to classify patients into stable cases and clinical ectasia after refractive surgery.11 Random forest was also used to combine the corneal biomechanical factors from Corvis ST (Oculus, Wetzlar, Germany).12 However, to our knowledge, the diagnostic value of the combination of all preoperative data has not been previously emphasized in the literature investigating patient selection using machine learning.
In our experience, surgeons or medical centers have slightly different criteria for corneal refractive surgery. Based on a clinical decision, refractive surgery can be performed in selected patients having a condition of relative contraindications, such as young adults, unstable refraction, large pupil size, dry eye, and diabetes. Moreover, surgeons should consider the nonlinear relationships of optical parameters to minimize the complication of each patient. Since all patient data and ocular measurements have been digitalized, the current technology can analyze the database to help refractive surgeons. For this study, we have built a machine learning architecture with the aim of identifying candidates for corneal refractive surgery to support clinical decision making (Fig. 1). The machine learning model was trained using clinical decisions of highly experienced experts. The employed architecture was based on large-sized preoperative clinical and ophthalmometric data and validated in a large Korean population indicated for refractive surgery.