Machine Learning Model for Predicting Visual Acuity Improvement After Intrastromal Corneal Ring Surgery in Patients...

Keratoconus is a progressive, degenerative corneal disease that can lead to significant visual impairment. The intrastromal ring segment implantation procedure is effective in reshaping the cornea and improving vision. However, vision does not improve postoperatively in all operated eyes, and the results vary widely among patients, making it challenging to predict postoperative visual gain.

This study investigated the potential of machine learning in predicting postoperative visual acuity in keratoconus patients undergoing intrastromal ring segment implantation with the aim of enhancing surgical decision-making.

This retrospective study analyzed 120 eyes of 102 patients with keratoconus who underwent ring segment implantation (1 symmetric or asymmetric segment, 150-300 μm thick, 150 degrees, or 160 degrees-arc). Preoperative and postoperative refraction, corneal topography, and tomographic data were collected. Various models were trained to predict postoperative visual acuity improvements.

The models demonstrated excellent performance, with XGBoost achieving perfect results in predicting whether vision will improve after surgery (R 2 = 1.0, Youden Index = 1.0; all test observations being correctly classified). The CatBoost model achieved an R 2 of 0.59 [0.7-line mean absolute error (MAE)] for predicting postoperative visual acuity, an R 2 of 0.76 (MAE, 1.08 D) for predicting keratometry, and an R 2 of 0.54 (MAE, 0.29) for predicting corneal asphericity. Key features for accurate predictions included preoperative keratometry values (K1, K2, Kmax), corneal asphericity, and visual acuity, whereas segment characteristics featured low importance.

This study shows the strong potential of machine learning for selecting candidates for surgery and predicting postoperative visual improvements after ring segment implantation in keratoconus eyes.