Analysis of the circadian rhythm of cardiovascular signals and their prognostic use in decision support systems

The focus of my research activity has been on the processing of cardiovascular signals in order to be able to use them as a support tool for doctors in their clinical decision making. Although the analysis of these cardiovascular signals has mainly been based on the punctual estimation of blood pressure and heart rate parameters, it is well known that the outpatient information, obtained from the 24h ambulatory monitoring, can provide prognostic support. Therefore I have tried to examine in detail how the values of the parameters related to blood pressure and heart rate over 24h change and how the relationship between them varies. Since the cardiovascular risk factors alter the trend of these biological signals, I have performed an analysis of the effects of each single risk factors on the circadian trend of the two signals and their relationship. Since, in recent years, mathematical approaches have been developed for the construction of clinical decision support systems applied, in the cardiovascular field, only to the classification of single heart beats of different etiologies; I have developed decision support systems to identify subjects with or without cardiovascular diseases. The pathologies examined were ischemic heart (IHD) and dilated cardiomyopathy (DCM). The described problems have been addressed using linear and nonlinear methods of signal processing and applying artificial intelligence algorithms. The average circadian trends of pressure and heart rate and their relationship on different categories of subjects were obtained. The linear and non-linear parameters were calculated from the heart rate variability signal and machine learning techniques were developed, the Artificial Neural Network (ANN) and Classification and Regression Tree (CART), applied to the previous parameters in addition to age, gender and to a specific clinical parameter. The results showed that the cardiovascular signals over 24h show a characteristic linear circadian rhythm divisible into four time intervals for the pressure signal (three intervals for the heart rate) in both normotensive and hypertensive subjects highlighting the importance of taking into account the time of day in which the signal is measured. The relationship between these two signals evaluated over 24h could be useful for understanding the control mechanism of the autonomic nervous system. The examination of the effects of risk factors such as smoking, obesity and dyslipidemia on cardiovascular signals showed that each factor modifies the physiological signals. The investigation of the influence of age and gender on cardiovascular signals highlighted an inversion of the trend in linear and non-linear parameters of heart rate variability in subjects>60 years of age and a gender differentiation only during the night. Finally, the results obtained by developing decision support systems based on machine learning techniques applied to various combinations of parameters, selected through principal component analysis, stepwise regression or correlated for less than 90%, showed that the ANN technique identify normal subjects and IHD with an accuracy of 80% and that the CART algorithm classify DCM patients with an accuracy of 97%. The latter technique was also able to distinguish these two etiologies from each other and from normal subjects with an accuracy of 81%. The results of my PhD activity highlight the importance of circadian analysis of cardiovascular signals, suggesting that particular attention should be paid to the time in which the measurements are performed providing useful information for the evaluation of the mechanisms that regulate the physiological control of the examined signals. Furthermore, the use of decision support systems based on machine learning techniques applied to parameters obtained in a non-invasive way from the processing of the heart rate variability is useful for diagnosing various cardiovascular diseases.