Automated non-linear motion detection and estimation for MPI SPECT

Uncontrolled motion in myocardial perfusion imaging (MPI) may significantly influence the diagnostic value of a Single-Photon Computed Tomography (SPECT) study. The main goal is to develop advanced algorithms resulting in the most realistic and motion free datasets with high efficiency as opposed to data-driven solutions. Due to the motion phenomena the characterization - quantitative measurements - total perfusion deficit (TPD) or ejection fraction (EF) of the left ventricle (LV) will suffer significant distortion. An automatic method has been developed in attempt to solve the correction as a result of multiple artifacts obfuscated by unexpected patient motions. Nevertheless the SPECT heart images will suffer from additional artifacts, attenuation from intestines, uncontrolled imaging environments and parallel collimation. Moreover during the acquisition, different non-linear motion effects interfere the final measurements. A fully automatic method is invented, working in the complete detector field-of-view (FOV) in attempt to eliminate the linear and non-linear motions. The algorithm (non-linear mcARM) is based on the projection-reprojection scenario, with noise reduction by automatic heart mask estimation and optical flow to achieve the desired results. The performance is evaluated on mathematical phantom and patient database showing an increased accuracy in motion detection compared to the state of the art MoCo (Cedars-Sinai). The proposed method showed significant improvement over MoCo based on our findings on both phantom and patient studies calculating the TPD scores and by visual interpretation. Additionally to the improvement the method was able to detect non-linear movements in the datasets as well. Conclusion: A fully automatic complete FOV motion correction algorithm with non-linear motion detectability is introduced, with fast running times (2-4 seconds). Evaluation on patient and phantom studies showed promising results for the future.