Statistical analysis
Continuous data were expressed as mean ± standard deviation (SD) or
median and interquartile range (IQR). Reproducibility within core lab
and between centers was compared by Bland-Altman plot and percent
difference was calculated for LV EDV, ESV, and SV as difference of the
two measures divided by the mean of the two measures. Time required to
complete fully automated and semi-automated contouring were compared
between centers using analysis of variance (ANOVA) and as a group using
student t-test. BSA was calculated using Haycock’s method7. To build the z-model of a parameter (i.e. ESV, DSV
and SV), we selected an optimum exponent, α, of the index parameter
(parameter/BSAα) such that: 1.) The index parameter
satisfactorily follows a normal distribution and 2.)The index parameter
does not depend upon BSA. Z-score was then calculated as
\(Z=\frac{[\ \left(\frac{\text{parameter}}{\text{BSA}^{\alpha}}\right)-\left(\text{mean\ value\ of\ indexed\ parameter}\right)]}{\text{SD\ of\ indexed\ parameter}}\).6Normality of an indexed parameter was evaluated using Shapiro-Wilk and
Kolmogorov-Smirnov tests, Q-Q plot, skewness and kurtosis. Dependence of
the indexed parameter on BSA was evaluated with a test of the slope of
the linear regression of the indexed parameters on BSA. We conducted
grid search with a 0.001 step size to find the optimum exponent, α, and
chose the one that maximized the sum of p-value for Shapiro-Wilk test
and the p-value of testing the slope of index parameter vs. BSA. During
the model development, diagnostic analysis were conducted using
leave-one-out method. Few data points with extreme values that
influences the distribution of indexed parameter were excluded from the
final z-model development. After the optimum has determined, association
of indexed parameter with age and gender were further examined with
respectively linear regression and Student t-test. Gender specific
z-scores model hence developed because there is difference between
genders in indexed parameter. A two-sided p-value <0.05 was
considered statistically significant. Statistical analysis was performed
using SAS version 9.4 (SAS institute, Cary, NC).