gms | German Medical Science

Objective: Sagittal balance describes a condition in which the body weight is located along a line slightly behind the rotation axis of the two femoral heads, and leads to a state of minimal energy consumption and optimal dispersion of weight.

Spine surgery, especially instrumentations, can affect the sagittal balance and might thereby cause negative or positive long-term side effects on neighboring segments. To avoid this, planning of surgery should be adjusted to the individual anatomy. Therefore, free handed standing full spine x-ray images are acquired and spinal parameters are calculated via manual placement of characteristic landmarks of the vertebrae or segmentation of each vertebra, which is a time-consuming process. In order to speed up the process of parameter calculation, a semi-automatic segmentation method is presented to outline lumbar vertebra contours where calculation of single parameters based on vertebra segmentation is already implemented in our in-house tool.

Method: The developed semi-automatic segmentation method consists of a preprocessing routine and the segmentation via Active Contour models (ACM). During preprocessing edges are enhanced via filtering procedures. The ACM is an energy minimization approach influenced by parameters like contour continuity, curvature and impact of intensities, as implemented in our routine. Initialization of the ACM segmentation is done performing one mouse-click roughly in the vertebra center and aligning a simple rectangle or patient related template (based on manual segmentations) to the clicked point.

The ACM was first trained using free handed standing full spine x-ray images of 5 patients, delivering optimized values for the parameters mentioned. Analysis of segmentation using a rectangle and template initialization was performed on the 5 data sets used for training and 3 further data sets. Segmentation quality is evaluated using the Dice coefficient (DC).

Results: Without training of the ACM segmentation, using a patient-related template yielded an average DC of 84.76% (SD: 2.38%), whereas ACM segmentation with rectangle shaped initialization delivered an average DC of 82.28% (SD 3.41%). After training of the ACM template-based segmentation yielded an average DC of 89.20% (SD 1.83%) and an average DC of 86.00% (SD 2.56%) for rectangle-shaped initialization.

Conclusions: The semi-automated tool presented offers an efficient way of achieving segmentation of single vertebra for further analysis in spinal parameter calculation.