Diagnostic Efficacy of CNN in Predicting Intraoperative Complications and Postoperative Outcomes in SMILE

To evaluate the diagnostic efficiency of the neural network in predicting complications of Small Incision Lenticule Extraction in a multi-center cross-sectional study.

The primary cause of global visual impairment currently is refractive error, and Small Incision Lenticule Extraction (SMILE) using femtosecond laser for corneal stromal lenticule extraction can alter the refractive power. However, complications such as opaque bubble layer (OBL), negative pressure detachment, and black spots may arise during the SMILE laser scanning process due to individual differences in corneal characteristics, significantly affecting the normal course of surgery and postoperative recovery. Experienced docters can often predict intraoperative complications based on scan images, patient cooperation, and other factors, but the learning curve is relatively long. At present, artificial intelligence has achieved the accuracy comparable to human physicians in the interpretation of medical imaging of many different diseases.Previously, we have trained a deep convolutional neural network for predicting intraoperative complications in SMILE procedures. The current multi-center study is designed to evaluate the efficacy of the convolutional neural network based algorithm in predicting intraoperative complications and to assess its utility in the real world.

Inclusion Criteria: * A condition in which the spherical equivalent refractive error of an eye is ≤-0.50 D when ocular accommodation is relaxed; * Age ≥18 years; * Spherical equivalent (SE) ≥-10.0D; * Corrected distance visual acuity (CDVA) ≥16/20; * Stable myopia for at least 2 years; * No contact lenses wearing for at least 2 weeks. Exclusion Criteria: * The presence or history of eye conditions other than myopia and astigmatism, such as keratoconus or external eye injury; * A history of eye surgery; * The presence or history of systemic diseases.