СРАВНЕНИЕ МЕТОДОВ КЛАССИФИКАЦИИ ДЛЯ ДИФФЕРЕНЦИАЛЬНОЙ ДИАГНОСТИКИ БОЛЕЗНИ ПАРКИНСОНА

Abstract

Background. Parkinson's disease is a common chronic neurodegenerative disease that impairs the quality of life. Currently, there are no drugs that can cure this disease. Early detection of pathology will improve the accuracy of diagnosis and prognosis, as well as start therapy at the stage when it can be most effective. Positron emission tomography with the radiopharmaceutical 18F-DOPA allows the detection of dopaminergic deficiency in patients with Parkinson's disease at the preclinical stage and differential diagnosis with essential tremor, in which dopamine-producing neurons are not affected. The purpose of this study is to determine the ability of various classification methods to differentiate patients with Parkinson's disease from other study groups. Materials and methods. The study involved 3 groups: healthy individuals (n = 33), patients with Parkinson's disease (n = 32) and patients with essential tremor (n = 29). The following classification methods were used in our work: naive Bayes classifier, k-nearest neighbors, random forest, logistic regression and artificial neural network. Results. All considered methods showed high quality of classification. The logistic regression model showed the highest results. The lowest values of sensitivity, specificity and accuracy were shown by the k-nearest neighbors’ method. Conclusion. Mathematical models will allow individual diagnosis of PD based on 18F-DOPA PET data with sensitivity, specificity and accuracy above 95%.