Assessment of bone segmentation quality of CT scanners using laser scanning

Abstract

We present a protocol for the evaluation of the geometric accuracy of bone segmentation algorithms in multi-slice and conebeam computed tomography (CT) Three important issues are resolved in this protocol: construction of a skull phantom with an accurate geometrical description serving as gold standard; registration between this geometric model and the 3D CT images; and a quality measure to evaluate the segmentation accuracy. A 3-D model of a dry skull phantom is obtained using a high resolution laser scanner. CT images are acquired of the phantom immersed in a water-containing head mould to mimic the presence of soft tissues. The geometric accuracy of bone objects derived by segmentation of the CT images using optimal thresholding is evaluated by comparison with the ground-truth provided by the laser scanned model. Anatomical correspondences between the laser model and the CT images are automatically established by image registration, such that bone thickness measurements in both modalities can be compared at a large number of anatomically corresponding sites distributed along the bone surface. The protocol provides a standardized methodology to evaluate the image quality of bone models for various CT scanners.