Аннотация:According to the World Health Organization cardiovascular diseases are the leading cause of death worldwide. Current research assumes that changes in blood microcirculation and microrheology, including the intrinsic properties of red blood cells and platelets, may play a significant role in the development of cardiovascular diseases (CVD): chronic heart failure, atrial fibrillation, etc. [1-2]. The aim of this study was to quantify and correlate changes in blood microrheology and capillary blood flow parameters measured in vivo and in vitro in patients with cardiovascular disease to identify diagnostically significant markers using machine learning.Totally 351 patients with various cardiovascular diseases and 25 practically healthy donors participated in the study. Fasting blood samples were drawn from the patient's cubital vein into 4 ml tubes containing EDTA K2 and 3.2% sodium citrate anticoagulants. The red blood cell aggregation and deformability were assessed in vitro using diffuse light scattering and laser ektacytometry methods on whole blood and diluted red blood cell suspensions respectively, using a RheoScan-AnD300 device (RheoMediTech, Republic of Korea). The platelet aggregation kinetics were assessed using the light scattering signal from the cell suspension using an ALAT-2 laser platelet aggregation analyzer (Biola, Russia) with platelet aggregation induced by adenosine diphosphate at concentrations of 1 and 5 μM in platelet-rich plasma. The microcirculation in vivo parameters were assessed using software processing of images and video fragments obtained using the Kapilyaroskan-1 optical capillaroscope (AET, Russia), equipped with a high-speed CCD camera, during visualization of the capillaries of the nail bed of the patient's fingers. A comprehensive machine learning approach was used to predict the presence of cardiovascular diseases. The study utilized a combination of various algorithms, including logistic regression, gradient boosting on CatBoost trees, k-nearest neighbors, and decision tree-based algorithms. The SHAP (SHapley Additive exPlanations) method was used to analyze in-depth the contribution of each laboratory parameter to the final model prediction and identify the most diagnostically significant markers. An analysis of the importance of machine learning model features using the SHAP method revealed that the most significant parameters for the random forest and catboost algorithms were capillary transition diameter measured using an optical capillaroscope, deformability, and erythrocyte aggregation. The K-nearest neighbors, logistic regression, and tree algorithms primarily relied on platelet and erythrocyte aggregation parameters. However, it is worth noting that the set of parameters that significantly influenced the model output was largely the same for each machine learning algorithm. Moreover, both complex and basic algorithms achieved 100% accuracy in the binary classification of healthy donors and patients with cardiovascular disease. Thus, machine learning was used to identify the most significant blood parameters obtained by optical methods and blood clinical tests that may influence the development and progression of cardiovascular disease and reflect them being markers of these diseases, both individually and together. This work was supported by grant No. 25-15-00172 of the Russian Science Foundation
