BEST IN PHYSICS (IMAGING): 3D Assessment of Complex Spine Deformity Using Linear Tomosynthesis and Statistical Shape Model Registration
C Zhao1*, M Herbst2, T Weber2, S Vogt2, L Ritschl2, S Kappler2, A Margalit1, A Jain1, J Siewerdsen1, W Zbijewski1, (1) Johns Hopkins University, Baltimore, MD, (2) Siemens Healthineers, Forchheim, Germany
Presentations
TU-A-TRACK 2-1 (Tuesday, 7/27/2021) 10:30 AM - 11:30 AM [Eastern Time (GMT-4)]
Purpose: Spine deformities (e.g. pediatric scoliosis and kyphosis) are conventionally evaluated using anteroposterior (AP) and lateral (LAT) long-view radiography or slot-scans. However, the evaluation of vertebral axial rotation is not possible in two-view planar imaging. We investigate the feasibility of comprehensive 3D assessment of spine deformity using linear unidirectional tomosynthesis (AP only) augmented with 3D-2D registration of a vertebral statistical shape model (SSM).
Methods: We consider an imaging protocol involving a longitudinal linear scan (1cm translation/view) of the x-ray source and flat-panel detector (varied collimation: 5x40 cm to 20x40 cm, 0.3 mm pixel) to acquire AP projections of the entire spine (180 - 60 views depending on collimation). Such scans can be implemented using a variety of robotic x-ray systems. To obtain 3D pose of an individual vertebra, its SSM was 3D-2D registered to the tomosynthesis projections. Feasibility studies utilized a previously validated high-fidelity x-ray system simulation framework. A digital pediatric anthropomorphic phantom with the capability to introduce spine deformations was developed. The accuracy of the 3D-2D SSM registration was tested on 120 instances with randomized initialization errors (±10 mm in translations and ±10° in rotations, 60 instances each for normal and deformed spines). Radiation dose was varied 1 – 5 mAs/view.
Results: The 20x40 cm collimation (60 views covering 60 cm length) provided sufficient depth information to accurately estimate 3D vertebral pose. For both normal and deformed spine, coronal plane translations and rotations were measured with less than ±1 mm and ±2° errors. Vertebral axial rotation errors were <±3°; AP translation errors were within ±5mm. The registration was robust down to 2 mAs/view exposure, corresponding to ~0.11 mGy mean dose to the spine (estimated by Monte Carlo simulation).
Conclusion: 3D-2D SSM registration using AP linear tomosynthesis achieves reliable 3D quantification of lumbar spine deformity with modest radiation dose.
Funding Support, Disclosures, and Conflict of Interest: Work supported by academic-industry collaboration with Siemens Healthineers, XP Division
Handouts
Keywords
Tomosynthesis, Registration, X Rays
Taxonomy
Contact Email