Published online before print January 26, 2006, 10.1148/radiol.2383050456
Automated Spine Survey Iterative Scan Technique1
Kenneth L. Weiss, MD,
Judd M. Storrs, BS and
Richard B. Banto
1 From the Center for Imaging Research (K.L.W., J.M.S., R.B.B.), Department of Radiology, Section of Neuroradiology (K.L.W.), Department of Biomedical Engineering (K.L.W., J.M.S., R.B.B.), and Department of Psychiatry (K.L.W.), University of Cincinnati College of Medicine, PO Box 670762, 234 Goodman St, ML 0762, Cincinnati, OH 45267-0762; and the Neuroscience Institute, University Hospital, Cincinnati, Ohio (K.L.W.). Received March 16, 2005; revision requested May 6; revision received June 2; accepted June 22; final version accepted August 12.
Address correspondence to K.L.W. (e-mail: kenneth.weiss{at}uc.edu).
View larger version (36K):
[in a new window]
|
Figure 1a: (a) Abbreviated schematic diagram of the computer algorithm and (b) remaining iterations, as detailed in the Appendix.
|
|
View larger version (37K):
[in a new window]
|
Figure 1b: (a) Abbreviated schematic diagram of the computer algorithm and (b) remaining iterations, as detailed in the Appendix.
|
|
View larger version (56K):
[in a new window]
|
Figure 2: Seven-section sagittal MR imaging projected volumes; the 35-cm FOV top and bottom image halves of the spine illustrate typical search regions (in green). Voxels that exceed the signal intensity threshold are depicted in orange. Those that meet additional disk constraints are depicted in white. Yellow lines connect the centroids of these putative disks (white) in the top and bottom halves.
|
|
View larger version (57K):
[in a new window]
|
Figure 3: Combined sagittal MR image depicts search paths parallel to the (yellow) line connecting the centroid of C2-3 with the longest disk chains from the top and bottom half images from Figure 2. Dots correspond to filtered local maxima along these paths.
|
|
View larger version (65K):
[in a new window]
|
Figure 4: Sagittal MR image shows entire spine with autolabeled intervertebral disks (yellow). The red dot corresponds to the manual assignment of the C2-3 centroid. The vertebrae themselves may also be autolabeled (labeling omitted here for clarity). Additionally, given the three-dimensional coordinates generated by the algorithm, disks or vertebral bodies can also be labeled in any subsequent imaging plane, provided no gross interimaging patient movement has occurred.
|
|
View larger version (67K):
[in a new window]
|
Figure 5: Sagittal MR image in a patient with 23 mobile presacral vertebrae (ie, those cephalic to the first sacralized vertebra), rather than 24, demonstrates neuroradiologist concordant autolabeling of the first 22 intervertebral disks (yellow). ASSIST software terms the first seven vertebrae C1 through C7, the next 12 vertebrae T1 through T12, and the subsequent five vertebrae L1 through L5, regardless of rib bearing status or possible sacralization. Alternatively, in this patient with only 23 mobile presacral vertebrae, the 22nd disk may be labeled L4-S1 and the unlabeled diminutive disk immediately caudal, S1-2 (23rd disk in this patient).
|
|
View larger version (68K):
[in a new window]
|
Figure 6: Sagittal MR image in a patient with 25 potentially mobile presacral vertebrae demonstrates neuroradiologist concordant autolabeling of the first 23 intervertebral disks (yellow). The 23rd disk is labeled L5-S1, with apparent lumbarization noted on this and adjacent sagittal images. The unlabeled 24th disk is prominent and may be termed lumbarized S1-S2 or simply S1-2.
|
|
View larger version (59K):
[in a new window]
|
Figure 7: Sagittal image in a patient with surgically fused L4-5 interspace and associated artifact from metallic cage demonstrates neuroradiologist concordant labeling of all 23 interspaces (yellow), including a good approximation of the L4-5 disk.
|
|
View larger version (74K):
[in a new window]
|
Figure 8a: Sagittal MR images in a patient with vertebral planus of T10. (a) Image mislabeled by using the initial algorithm (in first batch of 27 studies). Note the initial algorithm required two seed points (in red). (b) Neuroradiologist concordant autolabeled image after algorithm modifications, including adjustment of Gaussian filters. The modified algorithm required only a single seed point (in red).
|
|
View larger version (73K):
[in a new window]
|
Figure 8b: Sagittal MR images in a patient with vertebral planus of T10. (a) Image mislabeled by using the initial algorithm (in first batch of 27 studies). Note the initial algorithm required two seed points (in red). (b) Neuroradiologist concordant autolabeled image after algorithm modifications, including adjustment of Gaussian filters. The modified algorithm required only a single seed point (in red).
|
|
Copyright © 2006 by the Radiological Society of North America.
