Predicting the success of cyclophotocoagulation in neovascular glaucoma using a neural network model

Abstract

Aim: To investigate the diagnostic significance of inflammation and intraocular blood circulation indicators in a neural network model predicting the effectiveness of transscleral cyclophotocoagulation treatment in patients with NVG of diabetic origin. Method: This study included 127 patients (127 eyes; aged Ме 65.0 years) with diabetic NVG and 20 healthy individuals as an immunological control. All patients underwent TSC СPC with a diode laser λ=810 nm. Treatment success was defined as achieving an IOP level of ≤ 21 mmHg and maintaining or improving BCVA after 12 months of observation. Preoperative systemic immune-inflammation index (SII) and systemic inflammation response index (SIRI) were calculated. We assessed the values of volumetric pulse blood filling, determined by the rheographic coefficient (RQ, ‰), using the rheoophthalmography method. Multiple regression analysis was used to conclude the significance of treatment efficacy based on initial clinical and laboratory indicators, followed by constructing a prediction model in the neural network. Results: The development of the neural network model identified the most significant “input” parameters: SIRI (100%), RQ (85.7%), and SII (80.7%), which significantly influenced treatment success. The sensitivity of the neural network model was 100%, specificity was 30%, and the percentage of correctly predicted events during testing on the control group was 92.9%. Conclusions: Neural network-based prediction of transscleral cyclophotocoagulation effectiveness for patients with diabetic neovascular glaucoma allows for a sufficiently accurate forecast of treatment success with a probability of 92.9%. We believe the in-time correction of systemic inflammation and intraocular blood circulation can significantly reduce intraocular pressure, preserve visual acuity, and improve the quality of life in patients with diabetic NVG after TSC CPC. Further research is required to support these findings.