Enhancing Predicted Visual Acuity After SmartSight Lenticule Extraction: Identifying Key Factors With Machine Learning.

PURPOSE: To develop a predictive model aimed at assessing the likelihood of improvement in corrected distance visual acuity (CDVA) for patients undergoing lenticule extraction using the SmartSight system from SCHWIND eye-tech-solutions. This model evaluates the effectiveness and weight of various clinical and procedural parameters in predicting enhancements in visual acuity.

METHODS

Data from 1,262 eyes treated with the SmartSight system, encompassing 86 features, were analyzed. Regression and classification techniques were employed to estimate the probability of CDVA gain, ensuring robust results by comparing different methods. The dataset was divided into training (70%, 883 treatments) and testing (30%, 379 treatments) subsets to ensure comprehensive model evaluation using machine learning methods. Feature significance was determined via analysis of variance and principal components analysis to identify impactful parameters.

RESULTS

Three principal components accounting for more than 70% of the data variance were identified. The first component was primarily linked to system settings, whereas the second and third components were associated with patient gender and laterality. Factors significantly influencing CDVA improvement included a higher spot-to-track distance ratio, tighter track distance, lower pulse energy, lower average laser power, larger spot distance, greater cap thickness, and lower dosage. These variables were ranked based on their impact on CDVA enhancement.

CONCLUSIONS

Using low-energy asymmetric spacing for lenticule extraction with the SmartSight system is safe and effective. This approach improves visual outcomes for patients undergoing treatment for myopic astigmatism, offering a reliable method for predicting CDVA improvements. [J Refract Surg. 2024;40(12):e974-e984.].