Estimation of lesion position in computed tomography

In this simulation study, we investigate application of an objective estimation task to evaluate the influence of a varying pitch length in determining the position of a lesion in helical cone-beam computed tomography (HCBCT). Four different objects are scanned in the study, each being a three- dimensional Shepp-Logan head phantom with a lesion at a different position: the first has a spherical lesion on center, the second a spherical lesion off-center, the third an ellipsoidal lesion on center, and the fourth an ellipsoidal lesion off-center. The objects are scanned using 128 views over 2pi radians with a 100times300 pixel (20 cm times 60 cm) detector with source to detector distance of 80 cm and helical radius 50 cm. Two different pitch lengths are used, one of 10 cm and one of 15 cm. A Markov chain Monte Carlo technique is used to obtain maximum likelihood estimates of the position coordinates x, y, and z directly from the sinogram data. Bias and root mean square error in the estimates are determined by comparing the results with the known true positions. We find that while biases are consistently small, the 10 cm pitch length gives smaller root mean square errors for all 24 cases considered (four objects, two pitch lengths, and three parameters for each). This study is the first use of the estimation task formalism in HCBCT, and can be extended to image space and more realistic backgrounds and noise models, as well as additional parameters in future work.