|
Clinical Evaluation
of
Automated Technique
to
Reorient
Left-Ventricular
Myocardium in
Cardiac SPECT
Rakesh Mullick
Norberto F. Ezquerra
C. D. Cooke
R. D. Folks
E. V. Garcia
|
|
|
| |
Abstract
Clinical diagnostic interpretation of cardiac SPECT data requires
reorientation of transaxial slices of the left ventricle (LV) into
oblique (short, vertical and horizontal) slices. In order to generate
these oblique slices manual and semi-auto matic techniques have been
used in the past to reorient the volume data set. These techniques
are subjective, cumbersome and time-consuming. An automatic approach
to determine the pose of the LV and to delineate the long-axis has
been developed. The developed methodology is composed of three main
steps:
(i) Segmentation - Automatic identification of voxels
corresponding to the LV;
(ii) Topological Model Creation - Using the segmented
data to generate a 3D polygonal representation of the LV structure;
and
(iii) Topological Goniometry - Geometric and graphical
analysis of the topology to determine LV long axis.
In this report, we present a clinical evaluation of this methodology.
This approach was applied to 124 consecutive Tc-99m (50) and Tl-201
(74) cardiac SPECT datasets to automatically determine the LV orientation.
The orientation of the LV was defined using the horizontal (a) and
vertical (b) angles. The angles reported by the automatic approach
were then compared to those manually determined by experts for use
in the clinical evaluation. The results of our analysis is tabulated
below:
| Angle |
Mean
Absolute Deviation (deg.) |
| Technetium-99m |
Thallium-201 |
| horizontal
|
3.51 ±
3.48 |
6.19 ±
6.46 |
| vertical |
4.70 ±
3.81 |
6.62 ±
5.62 |
| % successful |
100% (50/50) |
90.54%
(65/74) |
Good correlation was observed between the manual and automatically
determined angles. Mean angular deviation reported corresponds to
less than a 2 voxel offset. The analysis failed for only 7 of the
124 datasets due to significantly lower counts in the data. Average
processing time per dataset was <30 sec. using modest computing power.
Conclusion: These results indicate that this objective,
standardized, technique to automatically determine the LV long axis
for reorientation is fast, accurate, robust and ready for clinical
implementation.
|
|
|
|