Purpose: Continuous curvilinear capsulorhexis (CCC), as a prerequisite for successful cataract surgery, is one of the most important and difficult steps in phacoemulsification. In clinical practice, the size and circularity of the capsular tear and eccentricity with the lens are often employed as indicators to evaluate the effect of CCC.
Methods: We present a neural network-based model to improve the efficiency and accuracy of evaluation for capsulorhexis results. The capsulorhexis results evaluation model consists of the detection network based on U-Net and the nonlinear fitter built from fully connected layers. The detection network is responsible for detecting the positions of the round capsular tear and lens margin, and the nonlinear fitter is utilized to fit the outputs of the detection network and to compute the capsulorhexis results evaluation indicators. We evaluate the proposed model on an artificial eye phantom and compare its performance with the medical evaluation method.
Results: The experimental results show that the average detection error of the proposed evaluation model is within 0.04 mm. Compared with the medical method (the average detection error is 0.28 mm), the detection accuracy of the proposed evaluation model is more accurate and stable.
Conclusion: We propose a neural network-based capsulorhexis results evaluation model to improve the accuracy of evaluation for capsulorhexis results. The results of the evaluation experiments show that the proposed results evaluation model evaluates of the effect of capsulorhexis better than the medical evaluation method.
Keywords: Capsulorhexis results evaluation; Continuous curvilinear capsulorhexis; Edge segmentation; Nonlinear fitting.
© 2023. CARS.