Spine surface detection from local phase‐symmetry enhanced ridges in ultrasound images
- Additional Document Info
- View All
PURPOSE: Improper administration of epidural anesthesia can result in nerve complications. This problem is exacerbated for obese patients whose vertebrae cannot be palpated. Ultrasound (US) has recently emerged as an attractive imaging modality for accurate epidural placement. However, anesthesiologists untrained in US have difficulty interpreting the anatomy in noisy spinal US images. Furthermore, the complex geometry in spinal US images is characterized by a discontinuous intensity profile because the transducer is often not perpendicularly oriented to spine surface regions such as laminae, articular and transverse processes. This makes the interpretation of spinal images more challenging than typical long bone surface images. In this article, we propose a new method to segment the spine anatomy in US images obtained in both the transverse and paramedian planes. METHODS: A set of 108 B-mode images were randomly chosen from 35 cine loops obtained from scanning the lumbar and thoracic vertebrae of 17 healthy volunteers with a BMI ranging from 19.5 to 27.9. A local phase-symmetry technique was applied to the B-mode images for enhancement of bone-like ridges, and the spine blobs were subsequently classified. The segmented spine surface from the blobs was compared against a radiologist's manual delineation of the spine surface. RESULTS: For the performance of the spine blob classifier, we obtain a Matthews Correlation Coefficient (MCC) of 0.77 and a geometric mean (G-mean) of 0.96. The mean absolute error between the manual delineation of the laminae by the radiologist and the automatic laminae segmentation is found to be 0.26 mm with a maximum possible absolute error of 2.01 mm for spinal US images of 70 mm depth. CONCLUSIONS: Our proposed technique successfully performs automatic segmentation of the spine surface - specifically the laminae, ligamentum flava, spinous, transverse, and articular processes - and can be extended to any bone anatomy present in an US image. This has implications for 3D visualization of bone surfaces and, without loss of generality, the vertebral column.
author list (cited authors)
Shajudeen, P., & Righetti, R.