Use of multivariable autoregressive model for detection of abnormalities in cardiac patients

Abstract

A multivariate autoregressive (MAR) model capable of determining the dynamic interactions between the electrocardiogram (ECG), the arterial blood pressure (ABP) and respiratory signals is presented. The model is able to quantify the cross-interactions among the signals. The use of the MAR model is then demonstrated using signals obtained from the MIT-BIH database for a case with respiratory failure due to cardiac problem. MAR spectral analysis is carried out to find the correlation between two signals, (viz, ECG and ABP, ECG and Respiration). It is found that a high coherence exists in the low frequency (LF) band. The coherence analysis is then applied to a few test cases of normal and abnormal signals. An index, called coherence index, is proposed for the assessment of abnormal condition in cardiac patients. Based on the limited testing, it is observed that the coherent index is lower for abnormal signals than for the normal signals and hence the method can be helpful in the detection of abnormality of cardiac patients. A continuous variation of coherence index for a long record of data has been obtained and the plot shows significant changes as the condition of the patient deteriorates in the ICU.