Use of MRI to assess the prediction of heart motion by stereo-tracking of markers on the body surface

We have developed a visual-tracking-system (VTS) which uses stereo-imaging to track the motion of markers on patients during cardiac SPECT imaging with the goal of using the tracked motion to correct for patient motion. The aim of this study is to determine using MRI in volunteers if the rigid-body-motion (RBM) model can be used to predict the motion of the heart within the chest from the motion of markers on the surface of the chest. Our methodology for investigating body-motion separate from the influence of respiration is to have the volunteers hold their breath during the acquisition of a sequence of 2 sets of EKG-triggered MRI sagittal slices covering the heart. The first set is acquired pre-motion, and the second post-motion. An analysis of the combined motion of the individual markers on the chest is used to obtain an estimate of the six-degree-of-freedom (6-DOF) RBM motion of the volunteers. The motion of the heart within the slices is estimated by semi-automatic 3D segmentation of the heart region in the second set of slices, and subsequent registration of this region to the first set of slices. The metric for judging agreement between the motion estimated by MRI and the VTS is the Average Error which is the average of the magnitudes of the difference in the vector displacements of all voxels in the heart region. Studies with the Data Spectrum Anthropomorphic Phantom and no-motion studies in which the volunteer did not intentionally move between the 2 acquisitions were used to establish a baseline for agreement between the 2 methods of motion estimation. For phantom acquisitions the Average Error when the motion was just translation was 0.8 mm. With complex motions which included a combination of rotations and translations the Average Error increased to 1.9 mm. In the human volunteers the Average Error averaged over all volunteer no motion acquisitions was 2.4 mm, which likely reflects the difficulty in humans maintaining their lung volume during breath-hold. For the case of translational motions which might be expected to be RBM the Average Error averaged over all volunteer translational studies was 2.1 mm, which is similar to that of the no-motion studies. For the case of exaggerated bends and twists larger than typically seen clinically which might be expected to be non-RBM for humans, the Average Error averaged over all such volunteer studies was 6.7 mm. Thus use of the RBM model with VTS predictions of heart motion during reconstruction should decrease the extent of artifacts for the types of patient motion studied, but less so for those which are better modeled as non-RBM. © 2010 IEEE