RU-2861547-C1 - METHOD FOR PREDICTING RISK OF RECURRENT ISCHAEMIC STROKE USING MACHINE LEARNING AND GENERATIVE ADVERSARIAL NEURAL NETWORKS

RU2861547C1RU 2861547 C1RU2861547 C1RU 2861547C1RU-2861547-C1

Abstract

FIELD: medicine; predictive medicine. SUBSTANCE: invention can be used for predicting the risk of recurrent ischaemic stroke. Venous blood is collected. Spontaneous platelet aggregation is determined. Induced platelet aggregation testing is performed using ristocetin, adrenaline, adenosine diphosphate, arachidonic acid, and collagen on days 1-2, 10-12, and after 6 months from hospitalisation. Markers ITGB3, GPIba, PLA2G7, ITGA2, ADRA2A, HMOX1, PTGS1, PTGS2, ABCB1, TBXA2R, PEAR1, and 9p21.3 are analysed. Synthetic patient profiles are generated using a generative adversarial neural network of the WGAN-GP architecture. The generated synthetic data is combined with real patient data. Said data set is used for automated construction of a machine learning model for predicting low, moderate, or high risk of recurrent ischaemic stroke. EFFECT: increasing the reliability of the prediction and the possibility of individual modelling of preventive and therapeutic scenarios by analysing clinical and genetic data using machine learning and generative adversarial neural networks. 3 cl, 3 dwg, 3 ex

Inventors

Anisimova Anastasiya Vyacheslavovna
Vorobev Igor Vladimirovich
GALKIN SERGEJ SERGEEVICH
Gunchenko Anastasiya Sergeevna
Anisimov Kirill Vladimirovich
GUSEV EVGENIJ IVANOVICH

Dates

Publication Date: 20260505
Application Date: 20251030

Claims (3)

1. A method for predicting the risk of recurrent ischemic stroke, characterized by collecting venous blood, determining spontaneous platelet aggregation, testing induced platelet aggregation in the blood using inducers ristomycin, adrenaline, adenosine diphosphate, arachidonic acid and collagen on days 1-2, 10-12, 6 months after hospitalization, analyzing genetic markers using hydrogel biochips containing markers ITGB3, GPIba, PLA2G7, ITGA2, ADRA2A, HMOX1, PTGS1, PTGS2, ABCB1, TBXA2R, PEAR1 and a marker of the intergenic region 9p21.3; training a GAN model; Synthetic patient profiles are generated using a generative adversarial neural network (WGAN-GP) architecture, the generated synthetic data is combined with real patient data, and this data set is then used to automatically build a machine learning model to predict low, moderate, or high risk of recurrent ischemic stroke.
2. The method according to claim 1, in which the machine learning model is implemented on the basis of a computing system containing an interactive interface for providing the physician with a prognosis for the development of a recurrent ischemic stroke, visualization of risk factors and modeling of therapeutic scenarios.
3. The method according to paragraph 1, in which continuous training of the machine learning model is carried out by incrementally updating its parameters upon receipt of new clinical and genetic data and automatic replacement of the model version when the forecast quality decreases below a specified threshold.

Description

The invention relates to medicine and information technology in medicine, in particular to methods for predicting and assessing the risk of developing a recurrent ischemic stroke based on the analysis of clinical and genetic data using machine learning and generative adversarial neural networks. The incidence of ischemic stroke remains consistently high both in Russia and worldwide. There are various clinical variants of stroke, among which ischemic strokes predominate (80-85%) [Gusev E.I., Skvortsova V.I. Brain Ischemia. Moscow: Meditsina, 2001. P. 13]. In most cases, ischemic stroke is a multifactorial disease, the development of which depends on both external factors and the individual's genetic characteristics. Clearly, the problem of preventing both primary and recurrent ischemic strokes remains relevant. Literary data indicate a fairly high percentage of recurrent ischemic strokes despite secondary stroke prevention. One of the problems is the inconsistency of genetic research results. This can be explained by the presence of concomitant comorbidities in the patient and the characteristics of different ethnic groups [Ferreira, M.; Freitas-Silva, M.; Assis, J.; Pinto, R.; Nunes, J.P.; Medeiros, R. The emergent phenomenon of aspirin resistance: Insights from genetic association studies. Pharmacogenomics 2020]. Another problem is the many influencing factors that determine the ultimate success or failure of secondary prevention. The clinical features of the disease, comorbidities, concomitant medications and non-modifiable risk factors such as age should be taken into account [Ferreira, M.; Freitas-Silva, M.; Assis, J.; Pinto, R.; Nunes, J.P.; Medeiros, R. The emergent phenomenon of aspirin resistance: Insights from genetic association studies. Pharmacogenomics 2020]. In addition, the interaction of genetic polymorphisms, as well as clinical factors, can influence the predisposition to the development of ischemic stroke [Cai, H.; Cai, B.; Sun, L.; Zhang, H.; Zhou, S.; Cao, L.; Guo, H.; Sun, W.; Yan, B.; Davis, S.M.; et al. Association between PTGS1 polymorphisms and functional outcomes in Chinese patients with stroke during aspirin therapy: Interaction with smoking. J. Neurol. Sci. 2017]. Over the past few years, machine learning (ML) models have been shown to be capable of solving various problems in biology and medicine, including pharmacogenetics [Wei, W.; Zhao, J.; Roden, D.M.; Peterson, J.F. Machine Learning Challenges in Pharmacogenomic Research. Clin. Pharmacol. Ther.2021, doi:10.1002/cpt.2329]. One of the key advantages of ML approaches is their ability to find subtle relationships and draw conclusions from complex data. The ability to accurately predict the course and outcome of ischemic stroke early is crucial for optimizing treatment strategies, setting realistic therapeutic goals, and planning rehabilitation measures. Early short-term predictors of the outcome and course of ischemic stroke are essential for planning therapeutic strategies within the first few days of stroke onset. A known method for predicting the development of recurrent acute cerebrovascular accident in patients who have suffered an ischemic stroke involves conducting a clinical and electrocardiographic examination of the patient 30 days after the acute cerebrovascular accident (ACVA). A biochemical blood test is performed, the presence/absence of atrial fibrillation or flutter, diabetes mellitus are determined, and the prescribed medications are clarified. It is also clarified which medications the patient is receiving regularly for the prevention of recurrent ACVA. Points are assigned for the treatment performed/not performed. The recurrent cerebrovascular accident prediction index (RCI) is calculated. With an RCI of 80.0-100%, the risk of recurrent ACVA is considered low. With an RCI of 60.0-75.0%, the risk of recurrent ACVA is considered average. With an RCI of 0-50.0%, the risk of recurrent ACVA is considered high. The method allows for an objective and accurate prediction of the development of recurrent stroke and timely correction of drug therapy through a quantitative assessment of the drug therapy used [RU 2593350 C2, Federal State Budgetary Institution "State Research Center for Preventive Medicine" of the Ministry of Health of the Russian Federation, 10.08.2016]. A known method for predicting the risk of ischemic stroke in patients with coronary artery disease and persistent atrial fibrillation involves examining the following parameters: the onset time of adenosine diphosphate-induced platelet aggregation in seconds, the concentration of thrombin-activated fibrinolysis inhibitor in percent, the peak ejection velocity from the left atrial appendage in centimeters per second, the concentration of soluble fibrin-monomer complexes in milligram percent, the left ventricular ejection fraction in percent, and the presence or absence of left atrial appendage thrombosis. The S index is calculated using the given formula,