Image registration, Computer Applications-3D, Ultrasound, Breast
J. C. H. Chiu1, A. Choate2, J. Ecanow2, G. Spear3, S. Mondello 4, Y. Lei5, C. Segarceanu5; 1Evanston, IL/US, 2Evanston/US, 3Evanston /US, 4Messina/IT, 5Chicago/US
The 3D and 2D MDs positively correlated with PCD,
PAD and BRD,
independent from the other probe-position associated variables (ρ = 0.49,
p<0.0001; ρ = 0.30,
p<0.0001; and ρ = 0.28,
The 3D average MD with the probe and body positions differences in close range,
PCD <10 mm,
BRD <10 degrees and PAD <20 degrees was 7.9 mm (standard deviation,
The smallest average 3D localization error (7.6mm) was obtained with the scan planes parallel between operators,
PCD <10 mm,
PAD <20 degrees with the body in supine position on the exam table (sd=2.2 mm,
n=14) (Fig 2).
When the scan planes between operators were perpendicular and the other conditions unchanged,
the average error was 8.5 mm (SD=4.2mm,
n=12) (Fig 3).
Larger average localization errors,
ranging from 10 to 37 mm,
were calculated when at least one of the angle difference or the PCD was increased.
When images were taken in the same body position or different body positions,
the MD or localization error increased when increasing the PAD and PCD,
as shown in Fig.
The localization error with manual measurements showed a wide range (5 mm to 41 mm) (Fig.4).
The distance from lesion to nipple showed small changes between images of same target obtained with different probe angles and subject in same body position,
for intra operator measurements,
(sd range 2.9 mm to 7.4 mm) and inter operator measurements,
(difference of average measurements 1.7 mm to 7.3 mm).
The average time to annotate an image with the automated system was 7 seconds while the time to manually annotate an image was between 20 to 60 seconds.