Chapter 1.Demonstrating the Validity of Methods for Studying and Evaluating the Results of Spectral Analysis of Arterial Oscillograms Recorded During Blood Pressure Measurement and AI Algorithms for Detecting Risks of Mental Diseases

Abstract

This study presents an innovative application of blood pressure monitoring with developed software methods for evaluating arterial oscillograms to diagnose the state of the autonomic nervous system. Despite the complexity of the physical interpretation of the origin of ultraslow-frequency waves (ULF) recorded during arterial oscillography, the results of the spectral analysis of ULF allowed the authors to substantiate their informativeness as an indicator of the level of activity of the CNS control signal. The results also confirmed the conclusions from the literature that the power of ultraslow waves increases significantly when the body's regulatory systems are depleted. To ensure the availability and convenience of arterial oscillography methods and analysis of its results, the authors first developed the Oranta-AO information system (a web environment, mobile application, cloud computing kernel). The data obtained were used to study the dynamics and assess the adaptive capabilities of the organism. The artificial intelligence algorithms and machine learning methods developed in this process, which are part of the expert system, provide additional information for specialists to make the necessary decisions. The results obtained make it possible to use these methods to diagnose premorbid conditions, evaluate treatment outcomes, predict various pathological, mental and functional conditions, and apply them in tele-rehabilitation. The main scientific contributions from the perspective of complex systems analysis and classifier development presented in this research are as follows: obtaining a base feature set from a general data set where the number of features significantly exceeds the number of data rows, using clustering in a reduced dimensional space to find objectively separable classes; taking into account correlations between features specific to one or only a few (but not all) classes and using the products of such features as input parameters to increase classification accuracy; developing a methodology for base feature set expansion by searching for their class-specific correlations with additional features and selecting those that most increase classification accuracy. This study explores the clustering of arterial oscillogram (AO) data among a sample of patients, focusing on ultra-low-frequency (ULF) indicators and their relationship with depression levels. Through dimensionality reduction using UMAP, two distinct classes emerged, categorized as milder and more severe cases. Using machine learning methods, an automated classifier was developed based on correlated ULF indicators, leading to improved classification accuracy. By incorporating ULF parameters as well as products of correlated parameters and additional measured factors, the classifier achieved high reliability in estimating depression levels. In particular, the nearest neighbors method yielded accuracies of up to 0.9792. This research enables the creation of an automated diagnostic classification AI service capable of reliably estimating at least four levels of depression based on AO analysis. The last part of the article presents the results of the implementation of arterial oscillogram analysis methods and AI algorithms in the Oranta-AO information system. The developed Oranta-AO information system has been put into medical practice in a number of leading clinics in Ukraine.