Figure 1.
Overview of the proposed segmentation method.
Given a SPECT dataset (a), the skeleton is extracted from the SPECT dataset (b). Next, the atlas skeleton (c) and the extracted skeleton (b) are registered to each other (d, e) using an anatomically realistic kinematic model. After the registration, the segmented data is reformatted into segments corresponding to the mouse atlas and thus mapping the data to a standardized atlas space (f, g). The data is now ready for an easy, fast and intuitive side-by-side exploration (multi-modal, follow-up or cross-sectional data) (h).
Figure 2.
The MOBY mouse atlas skeleton.
As originally included in the atlas (top), after segmenting the individual bones (middle), and a detail of the kinematic constraints and the DoFs of the femur/tibia-fibula bone complex (bottom). The colors indicate the different labels of each bone.
Table 1.
Resolution of each SPECT and correspondent CT dataset.
Figure 3.
Examples of SPECT skeleton isosurfaces with the corresponding CT skeleton isosurfaces after the pre-processing step.
The figure shows the positioning differences of the mouse in the scanner, SPECT (left) and the correspondent CT (right). The SPECT skeletons are incomplete, with several parts missing: especially in the case of front, hind limbs and the skull with large holes (blue arrows); also remnants of non-relevant objects such as lungs, kidneys and bladder are present (red arrows). In the bottom dataset the right femur and part of the spine are missing (green arrows) due to incomplete acquisition. The CT skeletons are complete and clean after the pre-processing step and are used in the validation of the proposed approach to calculate the Euclidean point to surface distance between the registered atlas and the skeleton surface. Black arrows indicate examples of regions where over and underestimation of the bone thickness occurred during the skeleton estimation in the data pre-processing step.
Figure 4.
Hierarchical anatomical tree followed during the registration process.
* indicates where user input is necessary: to pin-point the spine location where the vertebra connects the spine to the pelvis.
Figure 5.
APR layout of the segmented mouse data.
(a) - global articulated planar reformatted visualization of the atlas. (b), (c), (d) and (e) show the different data visualization options after applying the proposed approach. One can choose to visualize simultaneously and side-by-side a particular region of interest in cross-sectional studies for CT, SPECT or the combination of both. (b) - side-by-side visualization of the CT femur bone of 3 subjects, (c) - side-by-side visualization of the SPECT pelvic bone of 3 subjects, (d) - side-by-side visualization of the CT skull data fused with the correspondent SPECT data for 3 subjects, (e) - side-by-side visualization of the skull data of one particular subject: CT, SPECT and a combination of both. Follow-up data visualization was demonstrated in [10] for longitudinal CT mouse data.
Table 2.
Mean Euclidian point to surface distance between the SPECT and CT skeletons after the pre-processing step.
Table 3.
Quantitative results of the MOBY atlas-to-skeleton registration for 6 mouse SPECT datasets and [1].
Figure 6.
Top and side views of the segmented SPECT skeleton initially presented in Figure 1a).
The registered MOBY atlas is represented in green.