Non linear and Dynamic Time Warping classification of morphological patterns identified from Plethysmographic observations in the radial pulse

Impedance Plethysmographic (IP) observations from the arterial pulse forms a powerful tool for deciphering various cardiovascular diseases. However, a major bottleneck for applying this technique effectively is the assignment of variable waveform morphology to its respective diseases. Rationale of this work is to investigate Non linear and Dynamic Time Warping (DTW) approaches for classifying Impedance Plethysmographic (IP) waveforms. Our adaptation is two fold, firstly, where we establish the IP waveforms as a dynamic system using embedding dimension. Analysis of IP waveforms reveal the existence of eight different patterns corresponding to normal subjects and those suffering from diseases of heart, liver and lungs. Secondly we take up the task of classification of these patterns with template matching using DTW technique. Parametric evaluations reveal that IP waveforms exhibit dynamical nature in a three dimensional attractor. This observation is further validated by exhibition of determinism in reccurrence plots and by parameters determined using recurrence quantification analysis. Classification of the IP patterns by DTW on 1200 samples selected meticulously from 5628 patterns shows an average accuracy of 94%, 95% sensitivity, high statistical predictive values and kappa value of 0.9314 for the eight different classes.