Optimization of Ultrasound Image Quality for Automatic, Real-Time Segmentation of Spinal Bone Surfaces
L MacLean1*, D China1, K Ding2, J Siewerdsen1, N Johnson3, N Crawford3, A Uneri1, (1) Dept. of Biomedical Engineering, Johns Hopkins University, Baltimore, MD, USA, (2) Dept. of Radiation Oncology and Molecular Radiation Sciences, Johns Hopkins University, Baltimore, MD, USA, (3) Globus Medical Inc., Audubon, PA, USA
Presentations
SU-J-207-6 (Sunday, 7/10/2022) 4:00 PM - 5:00 PM [Eastern Time (GMT-4)]
Room 207
Purpose: Current approaches for navigated spine procedures suffer from accuracy loss due to anatomical changes during surgery. Intraoperative ultrasound (US) enables real-time tracking of intervertebral motion and can correct tracker registration both precisely to the local target and robustly against global deformations. This work experimentally assesses and optimizes US image quality (IQ) for application in image-based tracking of vertebrae.
Methods: The work comprises: (1) assessing IQ with respect to imaging parameters; and (2) optimizing the segmentation of the vertebral cortex. Initial work measured spatial resolution using a phantom (fishing lines) with respect to gain (G), dynamic range (DR), and focal depth (FD). Scan parameters were varied with respect to the probe type, frequency, power, and depth-to-target (with nominal settings: G=40, DR=74dB, and FD=15mm). Spatial resolution was quantified as the point-spread function (PSF), calculated as the standard deviation of a 2D-Gaussian fit to fishing line images. A U-Net segmented the vertebrae with a VGG11 encoder pretrained on ImageNet. Segmentation accuracy was quantified as the Euclidean distance between the predicted and ground-truth surface, and measured against the acquisition parameters and PSF.
Results: Resolution improved at higher frequencies, lower DR, lower gain, and at FD similar to the depth-to-target. Power was not significant. Based on these measurements, a linear vascular probe at higher frequencies (11.5MHz) was chosen. Studies in a spine phantom demonstrated strong correlation between segmentation accuracy and spatial resolution. Optimal selection of gain and DR achieves segmentation with 3.0mm median segmentation error (interquartile range 1–5.3mm).
Conclusion: Accurate segmentation of vertebral surfaces from US benefits from optimization of imaging parameters with respect to IQ. Phantom studies demonstrate segmenting the posterior vertebral cortex is feasible, which can estimate and correct for intervertebral motion in navigated procedures. Future work will extend IQ assessment and evaluate segmentation in cadaveric models.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by research collaboration with Globus Medical Inc.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email