6533b85efe1ef96bd12bfab5

RESEARCH PRODUCT

Determination of projection geometry from quantitative assessment of the distortion of spherical references in single-view projection radiography

Felix RoederD. D. BruellmannBernd D'hoedtRalf Schulze

subject

Plane (geometry)business.industryImage processingGeometryGeneral MedicineIterative reconstructionRadiusSubpixel renderingOpticsProjection (mathematics)DistortionShadowbusinessMathematics

description

A method is introduced, inferring the three-dimensional (3-D) location from the 2-D radiographic shadow of an opaque spherical reference body of known radius by considering its elliptical distortion, the 2-D shadow location and a known source-to-receptor distance. Three noncollinear spheres fixed to a rigid object constitute all possible degrees of freedom, i.e., the entire 3-D imaging geometry. The method may be used (a) to determine the 3-D imaging geometry from a single 2-D view and (b) to correct for foreshortening of object distances coplanar with the plane defined by the sphere triplet. Apart from the mathematical background the article describes a small feasibility experiment, performed with four different sphere diameters and a commercial dental ccd-receptor system (pixel length: 0.0195 mm). The mouse-cursor based image evaluation revealed an average underestimation of the critical depth- (x-) coordinate decreasing with increasing radius (-30.3% for r=0.5 mm to 2.8% for r=2.5 mm). Intraobserver reliability (the standard deviation between three single cursor-based assessments) ranged between 0% and 8% of the actual true depth. The main source of the input error is associated with the assessment of the amount of elliptical distortion, where subpixel accuracy is demanded. Consequently, software-based automated image evaluation is required using availablemore » methods for pattern recognition and point-spread correction. Provided sufficient accuracy, the method provides an important tool for foreshortening correction, depth assessment, motion analysis, and 3-D reconstruction from two or more 2-D views.« less

https://doi.org/10.1118/1.1796951