Abstract
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.
Method
Mixed spinal metastases were developed via intra-cardiac injection of canine Ace-1 luciferase transfected prostate cancer cells in a 3 week old rnu/rnu rat. Two sequential MR images of the L1-L3 vertebral motion segments were acquired using a 1H quadrature customized birdcage coil at 60 m isotropic voxel size followed by CT imaging at a 14m isotropic voxel size. The first MR image was T1 weighted to highlight the trabecular structure to ensure accurate registration with the CT image. The second MR image was T2 weighted to optimize differentiation between bone marrow and osteolytic tumour tissue. Samples were then processed for undecalcified histology and stained with Goldners Trichrome to identify mineralized bone and unmineralized new bone formation.
All images were resampled to 34.9 m and manually aligned to a global axis. This was followed by an affine registration using a Quasi Newton optimizer and a Normalized Mutual Information metric to ensure accurate registration. The whole individual vertebrae and their trabecular centrums were then segmented from the CT images using an extended version of a previously developed atlas based registration algorithm. An intensity-based thresholding method was used to segment the regions corresponding to osteoblastic tumor predominantly attached to the outside of the cortical shell. The whole vertebral segmentation from the CT was warped around the T2 weighted MR to define the bone boundaries. An intensity-based thresholding approach was then applied to the T2 weighted MR segment the osteolytic tumor.
Results
The customized MR coil acquired good quality images of both the bone and soft tissue structures in the spine. The CT based automated segmentation of the whole vertebrae and the trabecular centrums yielded high volumetric concurrency (∼90%) when compared to manually refined segmentations. Automated thresholding was even more robust in segmenting the individual trabecular networks and osteolytic tumours. The automation of the osteoblastic tumor segmentation was more challenging yielding concurrencies of ∼80% when compared to manually refined segmentations.
Conclusion
We successfully combined CT and μMR imaging to accurately segment mixed metastatic lesions within rat vertebrae using a highly-automated algorithm. These segmentations could readily be used for quantitative evaluation of new and existing treatments aimed at skeletal metastases or to generate finite element models to evaluate biomechanical behaviour or fracture risk.
