Purpose: The objective of this study was to establish an automated and objective method to quantitatively characterize the extent, spatial distribution, and temporal progression of metastatic disease in the bony spine.

Methods: Serial patient CT scans from GE Light-speed Plus CT Scanners were standardized to 120kVp, 1.25mm/2.5mm slice interval/ thickness, standard reconstruction, and 0.468mm/0.468mm pixel spacing. From 3D reconstructed CT images, trabecular regions within vertebral bodies (VBs) were segmented through atlas-based deformable registration (ITK, NLM, Bethesda). Voxel intensity histograms (voxel counts vs. Hounsfield Units) were used to characterize 32 healthy and 11 metastatically involved vertebrae (T5 to L5). Healthy histograms were fitted to Gaussian regression curves and compared using one-way repeated measures ANOVA (p< 0.05). Tumours were segmented as connected areas with voxel intensities between specified thresholds (Amira 3.1.1, TGS, Berlin).

Results: Histograms of healthy vertebrae were found to be Gaussian distributions (avg. RMSD = 30 voxel counts). The Gaussian mean & #956; ranged from 120 to 290HU, presumably due to inter-patient differences in age and activity. However, the histogram data sets were not significantly different (p> 0.8) across intra-patient vertebral levels T5-L5. Consequently, the Gaussian parameters, & #956; and standard deviation & #963;, determined from fitted healthy histograms could be used in adjacent metastatic levels to define patient-specific lytic and blastic thresholds for tumor segmentation. The ideal lytic and blastic segmentation thresholds were determined to be & #956;−& #963; and & #956;+2& #963; respectively: i.e. while histograms of metastatic VBs were non-Gaussian (RMSD of 56 voxels), subtracting from them the tumourous regions segmented accordingly restored the Gaussian nature of the distributions (RMSD of 24 voxels). Metastatic involvement can then be quantified from histograms of metastases in terms of: (1) lytic/ blastic volumes from areas under the curves; (2) severity of the pathologic involvement from the distribution and range; (3) tumor progression over time or treatment effects by taking the difference between sequential scans.

Conclusions: This proposed histogram-based method for characterizing spinal metastases shows great potential in extending the quantitative capacity of CT-based radiographic evaluations, especially in tracking meta-static progression and treatment effectiveness in clinical research applications. Funding: Other Education Grant Funding Parties: NSERC and CBCRA

Using serial CT scans, this project aims to develop a clinical research tool that analyzes changes in vertebral density in spines involved with metastatic disease. Tracking of total vertebral body and tumor volume alone was investigated. A program was developed to semi-automate the segmentation of the region of interest followed by image registration to superimpose the segmentation onto spatially aligned serial scans. Based on analysis of a simulated metastatic vertebra, generating a voxel distribution histogram from the vertebral body best quantified density in serial scans. This quantification method may improve clinical decision-making and treatment options for patients with vertebral metastases.

To develop a clinical research tool to serially track tumor involvement in vertebrae with metastatic disease by quantifying changes in CT attenuation.

Segmentation of the vertebral body and analysis of the voxel distribution within the region provides the most accurate method of quantifying changes in tumor involvement for the metastatic spine.

A quantitative method to assess the progression or regression of disease may improve clinical decision–making and treatment options for patients with spinal metastases.

The vertebral body segmentation was more accurate at tracking tumor involvement (voxel distribution histogram: 96.8% +/− 0.75% accuracy, mean density error: 4.7% +/− 0.8%) than segmenting the tumor volume alone (voxel distribution histogram: 86.1% +/− 3.6% accuracy, mean density error: 14.1% +/− 4.8%).

A program was developed to semi-automatically segment the total vertebral body and tumor volume alone from CT scans of metastatically involved vertebrae. Image registration through user-defined landmarks and surface matching was used to spatially align serial scans, and the initial segmentation was superimposed with the aligned scans. Changes within the segmentation between CT scans were tracked using mean density and a voxel distribution histogram. A cadaveric vertebra with a simulated tumor was scanned at five orientations with 20° offsets to determine the accuracy of the methods. Error primarily resulted from unavoidable re-sampling during alignment of the scans.