Практическое применение сервиса с искусственным интеллектом в диагностике болезни Паркинсона

Abstract

In the article we discuss the features of Parkinson’s disease (PD) diagnostics: the generally accepted clinical criteria, their sensitivity and specificity compared to modern diagnostic methods like a voice diagnostic service for PD based on artificial intelligence technologies. The purpose was to clinically test a prototype of a voice-based PD diagnostic service developed as part of the BRAINPHONE project. Material and methods. 100 patients (a balanced sample) who applied to the Republic Consultative and Diagnostic Center for Extrapyramidal Pathology (Kazan) were included in the unselected cohort clinical study. All patients were first examined by two experienced specialists in Parkinson’s disease and then by a trained neural network. The neural network conclusions options were «Parkinson’s disease cannot be excluded» or «No signs of Parkinson’s disease» (decision threshold 0.5). Results. As a result of testing the service, the following data were obtained: sensitivity — 80%, specificity — 92%. Compared with the generally accepted clinical criteria for diagnosing PD, voice diagnosis using a trained neural network developed within the BRAINPHONE project in this study showed comparable results in sensitivity and better results in specificity.