Evaluation of model-based processing algorithms for averaged transmitral spectral Doppler images.

In an effort to characterize more fully diastolic function using Doppler echocardiography, we have previously developed an automated method of model-based image processing for spectral Doppler images of transmitral blood flow. In this method, maximum velocity envelopes (MVEs) extracted from individual Doppler images are aligned and averaged over several cardiac cycles. The averaged waveform is fit by the solution of a kinematic model of diastolic filling. The results are estimates of the model parameters. As expected, the mean and standard deviation of the model parameter estimates depend on many factors such as noise, the number of cardiac cycles averaged, beat-to-beat variation, waveform shape, observation time and the processing methods used, among others. A comprehensive evaluation of these effects has not been performed to date. A simulation was developed to evaluate the performance of three automated processing methods and to measure the influence of noise, beat-to-beat variation and observation time on the model parameter estimates. The simulation's design and a description and analysis of the three automated processing methods are presented. Of the three methods evaluated, using the inflection point in the acceleration portion of the velocity contour as the first data point to be fit was found to be the most robust method for processing averaged E-wave MVE waveforms. Using this method under nominal conditions, the average bias was measured to be < 3% for each of the model parameters. As expected, the biases and standard deviations of the estimates increased as a result of increased noise levels, increased beat-to-beat variation and decreased observation time. Another important finding was that the effects of noise, beat-to-beat variation and waveform observation time on the parameter estimates are dependent on the location in model parameter space.