Comparison of Correspondence Models of Internal and External Respiratory Motion using 4D MRI

To achieve high resolution nuclear medicine imaging, respiratory motion has to be compensated, due to the image acquisition process occurring over several respiratory cycles. Motion affected imaging can either be compensated indirectly by gating or directly by applying some motion correction method. All motion correction methods rely on an estimated or assumption of motion. This paper follows the approach of using an external source of information to invert the deformation caused by respiratory motion. The proposed external source of information or surrogate is a stereo camera observation of the anterior surface of the torso. Previous approaches use linear maps to estimate internal motion from the anterior surface. In this paper, linear maps are compared to non-linear kernel models. Using a 4D MRI dataset, in evaluations intended for patient-specific estimation, it is found that non-linear kernel models on average produce more accurate estimates of internal motion compared to linear maps.