The use of a continuous left ventricular diameter signal as a monitor of cardiac performance
Description
It was demonstrated that there is information contained in a continuous left ventricular diameter signal which can be used to distinguish between changes in the different determinants of cardiac function. A technique was presented to identify informative features in the dynamic pattern of the waveform. This is intended for use as a monitor of cardiac performance during and immediately after cardiac surgery. In fifteen dogs, a series of hemodynamic interventions of short duration were administered: blood withdrawal, brachiocephalic artery constriction, and infusion of thiamylal sodium, calcium chloride, dextran, angiotensin and nitroprusside. The diameter signals were recorded before and after each intervention. Each signal was dignitized and reduced to its frequency components. The magnitudes were obtained using a Fast Fourier Transform and a window which results in accuracy of magnitude calculation within.1%. The phases were calculated with an iterative technique using the magnitudes, fundamental frequency, and the original waveform. The range of each phase angle was halved with each iteration. This resulted in phase accuracy to within.35 degrees. The phase of the first harmonic was set to zero, and the phases of the higher harmonics were adjusted accordingly. For each intervention, the reconstructed 'before' waveform was subtracted from the 'after' waveform. Closely spaced points in the subtraction waveform were compared. This yielded features which consistently distinguished between two different interventions. An algorithm was created for classifying an unknown case based on the features of the subtraction waveform, including the maximum, minimum and average diameter, and the power of the diameter waveform. The algorithm consisted of a series of nearest-neighbor pattern recognition tests to reduce the possible categories of intervention to one. The algorithm was tested on the dataset. It correctly identified 46 out of 74 cases or 62 percent: 89 percent of the blood withdrawal cases were correctly identified, as were 80 percent of the angiotensin cases. Of the categories classified as calcium chloride infusion, 86% actually were. Of those classified as nitroprusside, 78% actually were. A monitor of cardiac performance was proposed using the extracted features