Aims Lumbar facet joint injections (LFJIs) are the standard of care in diagnosing the sources of and treating low back pain. The current gold standard for LFJIs is manually performed under fluoroscopic guidance. However, this approach has certain disadvantages. It is only accessible in large hospital settings, relatively expensive to take, and exposes clinicians and patients to radiation hazards. Ultrasound guidance is a potential alternative as it is accessible to non-hospital settings, relatively cheap, and poses no known risk to humans. However, LFJIs performed under ultrasound guidance have not yet been established in clinical routine due to the difficulty of interpreting ultrasound images. We propose a novel algorithm and framework to reconstruct the patient’s lumbar spine with only ultrasound images that combine, for the first time, a statistical shape model with Neural Networks, aiming to pave the way toward LFJIs under pure ultrasound guidance.
Methods We propose to create a statistical shape model of the lumbar spine from 12 patients’ CT scans and train a neural network system capable of segmenting bone features from the noisy Ultrasound images. We then register the 3D statistical shape model to the segmented 2D images from the neural network system to get the desired patient-specific lumbar spine, which can be seen as a model fitting problem. Experiments are conducted to measure the variation of the statistical shape model and the accuracy of the neural network system.
Results Experiment results show that the statistical shape model created has suitable shape variation that could be used for 3D shape analysis. More CT scans could be added to increase the variation further. The neural network system successfully segments the ultrasound images. The output images could be used as the registration target for the statistical shape model. The further step is to perform non-rigid registration to fit the statistical shape model to the target images.
Conclusion The proposed approach could be the first step in replacing fluoroscopic imaging with ultrasound guidance in LFJIs.