Online Appendix for the following JACC: Cardiovascular Imaging article
TITLE: New Method of Evaluating Left Ventricular Diastolic Function from Color
M-mode Echocardiography
AUTHORS: Kelley C. Stewart, MS, Rahul Kumar, MD, John J. Charonko, PHD, Takahiro
Ohara, MD, PHD, Pavlos P. Vlachos, PHD, William C. Little, MD
APPENDIX
Appendix A: Detailed Methodology
Description of the Automated Algorithm
An automated data analysis algorithm was developed to determine the Vp from CMM
echocardiograph images. Original CMM images were analyzed in MATLAB (The Mathworks,
Natick, MA) using in-house developed image processing algorithms. From the original CMM
image (Error! Reference source not found.A), the velocity color scale region was selected and
a single beat region of interest was selected from the CMM image shown in Error! Reference
source not found.B. From this ROI, the E-wave, A-wave, and the vertical position of the mitral
plane were selected for orientation of the algorithm. The grayscale values from the CMM image
background were removed, thus leaving only the velocity color scale image (Error! Reference
source not found.C). Color scale cubic spline values were then used to create a point-by-point
velocity reconstruction of the remaining image.
A de-aliasing technique similar to the techniques used by Thomas et al. and Rovner et al.
(25,26) was used to reconstruct the image shown in Error! Reference source not found.D.
Through the use of image processing tools, the E-wave velocity field was reduced to a series of
twenty-seven isovelocity contours evenly spaced between 45% and 55% of the peak E-wave
transmitral velocity shown in Error! Reference source not found.E. The reconstructed velocity
contours are shown in Error! Reference source not found.F with the 45% to 55% isovelocity
contours shown from light to dark.
Ensemble Contour Methodology
A smoothing spline was fit to a series of isovelocity contours and is referred to as the
ensemble contour in the remaining analysis. Error analysis using 25 representative patients
spanning the five categories considered in this study was performed to determine the optimum
number of ensemble contours required for the reconstruction. This analysis showed that a
minimum of 27 velocity waveforms should be used for the optimal reconstruction and ensured
that, in all representative cases, there was no more than a 1% change in r2 values with an
increasing number of waveforms.
Change-Point Methodology
Preliminary analysis of the isovelocity contours revealed the presence of an apparent
decrease in the slope of the velocity ensemble contours indicating deceleration of the filling
wave. This is consistent with previous observations of a change in slope or curvilinear
isovelocity contour (17-19). We used a statistical change-point analysis method (27, 28) on the
derivative of the ensemble contour to determine the deceleration point. The method is based on
a cumulative sum of the difference between the value of interest ( xi ) and the mean value ( x ).
Equation 1 displays the cumulative sum equation:
Cumulative Sumi  Cumulative Sumi 1  ( xi  x)
The waveform produced by the output of the cumulative sum equation was plotted to
determine the significance of the change throughout the signal. The peaks within this cumulative
sum waveform were sorted according to their magnitude. The peak with the highest magnitude
signifies the most statistically significant change and was labeled as the “deceleration point.”
Appendix B: Intra and Interobserver variability Analysis
Seventeen patients were analyzed three times by three different observers. With this data
we have completed inter-observer and intra-observer variability analysis. The table below
displays the average of the mean percent error of the three observations for all of the patients for
the intraobserver variability. The interobserver variability is the average of the mean percent
error of the mean value of each patient per observer. The mean percent error is shown in
equation 1. Where n is the number of observation pairs analyzed
mean % error 
1 n observationt  observationt 1

n t 1  observationt  observationt 1 


2


Development Cohort Reproducibility
Vs
Deceleration
Initial Vp
Terminal Vp
Point
Intraobserver Variability (%)
9.9 ± 13.2
6.4 ± 10.4
9.2 ± 15.1
9.4 ± 15.8
Interobserver Variability (%)
13.8 ± 11.5
7.1 ± 9.2
4.6 ± 14.3
9.1 ± 21.5