Real-time automatic spinal level identification with ultrasound image processing

In this paper, we propose an ultrasound image processing procedure for fully automatic lumbar spine level identification. The image processing procedure starts with automatic sacrum identification, with feature selection and support vector machine (SVM) classification. After sacrum is detected, a panorama image stitching procedure is initiated to obtain the overall spinous processes structure. Throughout the image stitching and ultrasound probe movement, an image quality evaluation standard is utilized to select the suitable frames to be stitched onto the panorama image, so as to eliminate the accidental bad-quality frames in case the probe goes out-of-line during the scanning movement. In the meantime, the spinous levels are identified and counted on the panorama image, which are then reflected on the original ultrasound image in real time, so as to inform anesthetists where is the level L3-L4 and when to stop scanning. The processing result with off-line collected lumbar ultrasound videos shows a high accuracy on sacrum identification and spinous level identification.