Artificial intelligence capabilities in choosing surgical treatment for tetralogy of Fallot

Objective. To develop a machine learning (ML) model recommending surgical tactics for tetralogy of Fallot (TF). Material and methods. A single-center cohort study was conducted. We prepared a dataset based on retrospective pool of medical data from 300 patients with TF who underwent surgery between 2019 and 2022. The target feature was 4 surgeries in children with TF. Inclusion criteria: children who underwent primary surgery. Patients were divided into 4 groups (target variables) depending on surgery: group 1 — radical correction of TF, group 2 — right ventricular outflow tract reconstruction, group 3 — systemic-pulmonary anastomosis, group 4 — stenting/ballooning of the right ventricular outflow tract. Results. The LGBMClassifier (Light Gradient Boosted Machine) model had the best quality metrics. Satisfactory results were also obtained for LogiscticRegression (Logistic Regression in Machine Learning), RandomForestClassifier, XGBClassifier (eXtreme gradient boosting), CatBoostClassifier. Conclusion. Integration of AI algorithms into the treatment of patients with TF opens up significant prospects for improving care.