Prediction of Postcataract Surgery Visual Acuity Using Machine Learning Algorithms in a Camp-based Setting

Résumé fourni par la source

Background: Predicting visual outcomes after cataract surgery is critical, especially in resource-limited, camp-based settings. Machine learning (ML) techniques are promising tools for accurate forecasting based on complex clinical data. Aim: This study aimed to develop and evaluate supervised ML algorithms for predicting postoperative visual acuity (VA) on day 7 after cataract surgery based on demographic, clinical, and surgical parameters. Materials and Methods: This retrospective study included 170 patients who underwent cataract surgery at a camp in the Andaman and Nicobar Islands. A total of 21 preoperative and intraoperative variables were used to predict VA outcomes categorized as good (VA >6/18), moderate (6/18–6/60), or poor (VA <6/60). Eight ML algorithms – multinomial logistic regression (LR), support vector machine, k-nearest neighbors, Naïve Bayes, classification tree, random forest (RF), artificial neural network (ANN), and AdaBoost – were implemented through the Orange 3.24.1 platform. Model performance was evaluated through 10-fold cross-validation, with metrics including classification accuracy (CA), area under the curve (AUC), precision, recall, and F1 score. Results: The ANN achieved the highest CA (74.7%) and AUC (0.758), followed by LR and RF analyses. Principal component analysis and scatter plots highlighted key predictors such as age, nuclear opalescence, and preoperative VA. The confusion matrix and receiver-operating characteristic analysis further confirmed the robustness of the ANN and LR models in classifying outcomes. Conclusion: ML algorithms, particularly ANN and LR, effectively predict visual outcomes after cataract surgery in an outreach setting. These tools can enhance clinical decision-making, patient counseling, and resource optimization in underserved populations.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.