Automatic labeling and segmentation of vertebrae in CT images

Detection, Labeling, and segmentation of the spinal column from CT images is a pre-processing step for a range of image-guided interventions. State-of-the art techniques have focused either on image feature extraction or template matching for labeling of the vertebrae followed by segmentation of each vertebra. Recently, statistical multi-object models have been introduced to extract common statistical characteristics among several anatomies. These models have also been used for joint labeling and segmentation of the lumbar spine and were shown to be robust, accurate, and computationally tractable. In this paper, I reconstruct a statistical multi-vertebrae pose+shape model and utilize it in a novel framework for labeling of an arbitrary vertebra in a CT image. I also use the model for segmentation of the entire vertebral column. I validate my technique in terms of accuracy of the labeling and segmentation of CT images acquired from 61 subjects. The vertebral column is correctly labeled in 97% of the subjects and mean distance error achieved for the segmentation is 2.1±0.7 mm.