Purpose

To develop a low complexity highly-automated multimodal approach to segment vertebral structure and quantify mixed osteolytic/osteoblastic metastases in the rat spine using a combination of CT and MR imaging. We hypothesize that semi-automated multimodal analysis applied to 3D CT and MRI reconstructions will yield accurate and repeatable quantification of whole vertebrae affected by mixed metastases.

Purpose: To examine the effect of image resolution and structural model on quantifying architectural differences between healthy and metastatically involved vertebrae.

Method: Lumbar vertebrae of healthy(n=6) and meta-statically involved(n=6) rnu/rnu rats were utilized. Osteolytic vertebral metastases were developed via intracardiac injection of human MT1 breast cancer cells. μCT images of the vertebrae were acquired ex vivo at 14μ isotropic spatial resolution. The whole vertebrae were segmented using an automated atlas based demons deformable registration followed by level set curvature evolutions. A subsequent iteration of level set was used to yield a segmentation of the trabecular centrum. The individual trabecular network was further segmented using intensity based thresholding. Architectural parameters were computed from the segmented μCT images: Cortical Bone Volume(CBV), Trabecular Bone Volume(TBV), Trabecular Bone Surface Area and the degree of anisotropy based on Mean Intercept Length(MIL). From this, trabecular Thickness(TbTh), Trabecular Number(TbN) and Trabecular Separation(TbS) were calculated using the Parfitt Model (Parfitt, Bone & Mineral. 1987). TbTh was also calculated separately using the Hilderbrand model (Hilderbrand, J of Microscopy 1997). The degree of anisotropy was determined via Mean Intercept Length (MIL) measured utilizing a binary shift/subtraction approach. The measures of TbTh and MIL were compared for each image at 8.725(high), 17.45(medium) and 34.9(low) μm3 isotropic spatial resolutions.

Results: Parfitt’s plate model showed a significant decrease in TBV, TbN and CBV and a significant increase in TbS in the metastatic vertebrae in comparison to the healthy group at the highest resolution. In both Hilderbrand’s and Parfitt’s models at the highest resolution there was no significant difference in TbTh between the healthy and metastatic groups. In both models, TbTh and TbS values rose while TBV and TbN decreased as the resolution was lowered. Significant reductions were observed only in TbTh between the healthy and metastatic vertebrae at the medium and low resolutions. In all cases, the Hildebrand model yielded lower values of TbTh than the Parfitt model. However, achieving robust automated results using the Hildebrand method was limited in the final stage of the segmentation due to sensitivity to small islands of bone. Structural anisotropy remained consistent in all groups at all resolutions, with ~3x greater MIL in the superior/inferior direction. The degree of anisotropy was, however, consistent in both groups suggesting that the metastatic destruction does not have any directional preference.

Conclusion: The automated use of Parfitt’s plate model along with the MIL method can be used to yield quantitative analyses demonstrating differences in vertebral microstructure due to metastatic involvement. However the sensitivity of these architectural parameters to resolution motivates the need for high resolution scanning in future preclinical applications.

Purpose: The objective of this study is to develop and utilize a highly automated microCT based analysis tool to quantify microstructural differences in bone due to metastatic involvement in whole rat vertebrae.

Method: First and Third lumbar vertebrae from healthy (n=4) and metastatically involved (n=4) rnu/rnu rats were excised for analysis (total of 8 vertebrae). Lytic metastases were developed via intracardiac injection of MT1 human breast cancer cells. The specimens were scanned using microCT at 17.5 microns isotropic resolution. A highly automated algorithm was developed for whole vertebral segmentation based on the microCT data, including the posterior elements (AmiraDev3.1). This was accomplished using an atlas-based method incorporating demons deformable registration followed by refinement through level set curvature evolution. Volumetric concurrency was used to compare segmentations generated by the automated algorithm to manually refined segmentations. The segmentations were up-sampled by 4 and edge-enhanced and further segmented using a thresholding technique to have a clear segmentation of the individual trabeculae without advancing into the bone marrow(AmiraDev3.1). The cortical shell was removed automatically before analyzing the trabecular structure. Cortical bone volume(CBV) was calculated by subtracting the volume of the full segmentation from the segmentation with no cortical shell. The interior segmentation was then used to calculate Trabecular Bone Volume(TBV), Trabecular Thickness(TbTh), Trabecular Separation(TbSp), Trabecular Number(TbN) based on the expressions described by Parfitt, et al(1983). Finally mean intercept length(MIL) was used to calculate the anisotropy of the trabecular tissue. Analysis were carried out on both the healthy and metastatically involved vertebrae.

Results: The automated algorithm including the level set method refinement produced good tracking of the boundaries of entire rat vertebrae. Consistent results yielded significant reduction in TBV, slight reduction in TbN and TbTh, and significant increase in TbS in metastatic vertebrae compared to healthy. no significant differences were observed in CBV. The metastatic vertebrae was also found to be significantly more anisotropic than the healthy group.

Conclusion: The accuracy of the highly automated algorithm developed in this study to analyze microstructure in whole rat vertebrae make it a suitable tool for further analyzing the effects of existing and new treatments for spinal metastases at a preclinical level.