How many vertebral labels should I use for registration?¶
If you plan on using the labeled spinal cord to perform registration to the template (using sct_register_to_template), you don’t necessarily have to label every vertebrae. Instead, you can choose to work with a subset of labels:
- 3+ labels:
This is the most accurate method for matching the vertebral levels of the anatomical image to the levels of the template. However, there is a key downside to this approach: Because more than two points must be matched, the level-matching transformation cannot be affine. As a result, the output warping field will be undefined for regions above the top label and below the bottom label.
- 2 labels:
In practice, the difference in accuracy between using 3+ labels and 2 labels is often negligible. Using 2 labels also has the added benefit of allowing for an affine level-matching transformation, which means the template-to-image warping field will be defined for the entire image.
- 1 label:
If your image covers only 1 vertebrae, you can still provide a single label. Note that the transformation in this case will be limited to a Z-axis translation, as an affine transformation can’t be determined for a single point.
Due to its benefits, we recommend starting with 2 labels for your registration when possible. The next page of this tutorial will show you how to extract 2 labels from the full set of labels produced by sct_label_vertebrae.