AUTOMATED
EDGE-DETECTION FOR MEASUREMENT OF BRACHIAL REACTIVITY: COMPARISON OF INTER- AND
INTRA- OBSERVER CONCORDANCE WITH MANUAL MEASUREMENT. B
Haluska*, A Sutherland, C Case, R Kennedy, TH Marwick. Princess
Alexandra Hospital, University of Queensland, Brisbane, Qld. The
reproducibility of brachial artery reactivity (BAR) relies on the ability to
measure very small changes in lumen diameter between rest, hyperemia and after
nitrate administration. We sought to compare automated measurement with edge-tracking
software with manual measurement and also determine which method had the
highest concordance between observers. Methods:
Two observers measured images from 120 pts who had undergone assessment of BAR.
All images were digital cineloops acquired using an ATL HDI 5000 with a 12MHz
vascular probe. Images were measured
off-line using HDILab (ATL Corp, Bothell WA).
Four images per patient were measured: rest, hyperemia, rest2, and
GTN. Each image was measured using
HDILab automated measurement (auto) which measures the leading edge of anterior
and posterior endothelium and tracks this in sequential frames, and also
manually (man), where a set of point to point measurements was averaged.
Absolute measurements as well as percent change were compared and then compared
against the mean for the observers. Results:
Mean artery size was 4.2±1mm at rest. The BAR response varied from 36% to -32%
for observer 1 and 33% to -55% for observer 2. Correlation between LDOT and
distance for observers 1 and 2, and inter-observer correlation for automated
and manual methods were all p<0.01; Measurement Man
vs auto 1 Man
vs auto 2 Man
1 vs 2 Auto
1 vs 2 Rest 4.2
± 1.0 0.98 0.96 0.95 0.94 Hyper 4.4
± 1.0 0.98 0.97 0.95 0.93 %Chg 4.0
± 1.0 0.87 0.78 0.60 0.57 GTN 4.6
± 1.0 0.99 0.97 0.97 0.96 %Chg 8.0
± 1.0 0.89 0.81 0.66 0.58 The
standard deviation from the mean was 5% for observer 1 and 7% for observer 2.
The equivalent deviation between observers was 8% for manual and 10% for
automated measurement. Conclusion:
BAR measurements appear concordant between observers but both manual and
automated methods show deviation from the mean.
