Purpose: This study aimed to develop a machine learning (ML) model using intraoperative optical coherence tomography (iOCT) data to predict postoperative vault and guide real-time intraocular collamer lens (ICL) orientation adjustments for optimal vault outcomes. Setting: Shaanxi Eye Hospital, Xi'an People's Hospital (Xi'an Fourth Hospital), Affiliated People's Hospital of Northwest University, Xi'an, Shaanxi Province, China Design: prospective case series Methods: We prospectively collected data from patients undergoing ICL implantation surgery. Intraoperative vault measurements were acquired in real time using iOCT immediately following ICL placement. Postoperative vault dimensions were then assessed using anterior segment OCT (AS-OCT) at 1-day, 1-week, and 1-month follow-up visits. These measurements were used to train and validate ML models for predicting postoperative vault based on intraoperative iOCT data. Results: The study included 106 patients (158 eyes). Among the evaluated ML models, the Random Forest algorithm demonstrated superior performance, yielding a mean absolute error (MAE) of 106.48 ± 19.23 μm and root mean square error (RMSE) of 141.37 ± 24.87 μm. Prediction accuracy was high across incremental error thresholds: 62.9% (±50 μm), 85.9% (±100 μm), 94.9% (±150 μm), 97.2% (±200 μm), and 98.9% (±250 μm). All models showed strong agreement between predicted and observed values, with the Random Forest model exhibiting minimal systematic bias. Conclusions: Our findings demonstrate that iOCT enables accurate real-time intraoperative vault measurement, which reliably predicts postoperative vault dimensions. The developed ML model provides an objective, data-driven approach to optimize ICL orientation during surgery, potentially improving clinical outcomes.
Du et al. (Mon,) studied this question.