Efficient triple output network for vertebral segmentation and identification

Precise vertebral segmentation provides the basis for spinal image analyses and interventions, such as vertebral compression fracture detection and other abnormalities. Deep learning is a popular and useful paradigm for medical image process. In this paper, we proposed an iterative vertebrae instance segmentation model, which has good generalization ability for segmenting all types of vertebrae, including cervical, thoracic, and lumbar vertebrae. In experimental results, our model not only used 17% less memory but also achieves better performance on vertebrae segmentation compared to existing methods. The existing method provides only two output for segmentation and classification respectively. However, with more memory available, our model is capable of providing third output for accurate anatomical prediction under the same amount of memory.