Daniel F. Leotta (1) and Roy W. Martin (2,3)
<BR>
<BR>
1 Surgery, University of Washington, Seattle, WA,
<BR>
2 Bioengineering, University of Washington, Seattle, WA,
<BR>
3 Anesthesiolgy, University of Washington, Seattle, WA
<BR>
<BR>
<B>Three-Dimensional Spatial Compounding of Ultrasound Scans with
Incidence Angle Weighting</B>
<BR>
<BR>

A 3D ultrasound imaging system has been used to compare methods for 
spatial compounding of images acquired with different scanhead positions 
and orientations. The most commonly cited techniques for the combination 
of overlapping data are the maximum and mean. However, these methods do 
not take into account the relative confidence in the data, which may vary 
with viewing geometry.  A compounding algorithm has been developed which 
assigns regional weights depending on the local incidence angle of the 
ultrasound beam. Since image quality is expected to be best at incidence 
angles close to zero degrees, voxels acquired with views closer to normal 
are assigned higher weights when compounded. The weights are derived from 
extracted surfaces and a knowledge of the beam direction during acquisition.
<BR>
<BR>
Scans from multiple windows were performed on two bones <I>in vitro</I> 
and the 
shoulder rotator cuffs of two subjects. Border measurements (peak value 
and width) along surfaces of interest were compiled as a function of 
ultrasound beam incidence angle and compared for single views (no 
compounding) and for maximum, mean and weighted mean compounding. 
<BR>
<BR>
All compounding algorithms show dramatic filling of surfaces, as 
demonstrated by a low variability of the peak values measured along the 
borders when compared with those of a single view. For a bone imaged <I>in 
vitro</I>, the border peak gray level was 84 +/- 18 for a single scan, 
101 +/- 4 for the maximum combination, 80 +/- 6 for the mean combination, 
and 93 +/- 3 for the weighted mean. The border full width at half maximum 
(in voxels) was 2.4 +/- 0.11 for a single scan, 2.6 +/- 0.11 for the 
maximum, 2.6 +/- 0.08 for the mean, and 2.3 +/- 0.06 for the weighted mean. 
<BR>
<BR>
The weighted mean produces less variability than that of the maximum 
and mean for both intensity and border width. While there is an overall 
increase in the measured border width for the maximum and mean, the 
weighted mean does not demonstrate degraded border resolution. Similar 
results were found for the borders of both the humerus and the supraspinatus 
tendon <I>in vivo</I>. Surfaces derived from the weighted reconstructions 
demonstrated fewer gaps and fewer spurious connections between surfaces, 
which could be of particular importance for automated image analysis.