Algorithm for Calculating the Dioptric Power of an Optical Cylinder for the Optical Stage of Keratoprosthesis Implantation

Objective : to analyze the clinical and functional outcomes of the optical stage of keratoprosthesis implantation and to develop the machine learning model for calculating the optical cylinder power. Patients and methods. The study enrolled 90 patients (90 eyes) with burn injuries and terminal dystrophic corneal opacities who underwent the surgical treatment. The average age of the patients was 59.5 years (40.75; 70.5), of whom 65 were men and 25 were women. The patients underwent keratoprosthesis implantation in two stages. The first stage involved the implantation of the intralamellar plate into the corneal leucoma or corneal-prosthetic complex implantation. The second stage involved the implantation of the optical cylinder, which was selected considering axial length, total corneal leucoma thickness, the thickness of autograft covering and the presence or absence of the native lens/intraocular lens/lens-iris diaphragm. Outcomes. Considering the results for the quality metrics (MAE and R2) and the achieved predictability within ±0.50 D in 30.0 % of cases and ±1.00 D in 50 % of cases on the test sample, it was decided to develop the model based on two features: axial length and total leucoma thickness. The model with two features demonstrated better quality than the model with one feature on the test sample across all quality metrics (MAE and R2) and the predictability within ±0.50 D and ±1.00 D, which were 30.0 % and 70.0 % respectively. Conclusion. The linear regression model was developed to calculate the dioptric power of an optical cylinder based on the axial length and total leucoma thickness. It demonstrated predictability within ±1.00 D in 70.0 % of cases. Further observations are required to improve the model.