Data-driven motion estimation in Single Photon Emission Computed Tomography (SPECT)

Abstract

Single Photon Emission Computed Tomography (SPECT) is one of the most important imaging modalities in nuclear medicine used for imaging different body organs for diagnostic purposes. It is well known that patient movement during the acquisition time do introduce artifacts in the reconstructed SPECT images. The projection frames become out of alignment due to the patient's movements, degrading the reconstructed image and do introducing artifacts. The accuracy of the diagnosis could be considerably impacted by these motion artifacts. Getting consistent projection data from the acquisition requires motion correction. Therefore, the motion correction of patients in tomographic images is very essential for more accurate diagnosis of diseases. The aim of this project is to develop and evaluate a data driven approach to motion estimation and correction of SPECT data without necessity for motion tracking system. The objectives can be summarized as the different groups of experiment were carried out on several set of data to validate the motion estimation and correction procedure. A fully three-dimensional (3-D) algorithm is proposed that estimates the patient motion based on the projection of motion-corrupted data and updated the image and finally correct the motion. At first, initial studies have been performed using a digital version of the Hoffman brain phantom. Movement was simulated by constructing a mixed set of projections in discrete positions of the phantom. Several number of motion induced projections set were generated from the physical phantom. The single or multiple movement have been applied to the physical phantom and the data were acquired. The translational movement, angular location and rotational motion (all 6 dof parameters) were analyzed. Motion estimation and correction algorithms were applied to the acquired data as well as simulated motion induced physical phantom data. Clinical data with/without motion were collected and motion correction algorithms were applied to estimate motion and reconstructed image were improved. The proposed method iteratively estimate and compensate the motion during image reconstruction. In every iteration, the rigid motion was assessed view-by-view and then used to update the system matrix. A first rough motion estimate can be produced using the initial 1 reconstructed image, which is motion contaminated. In order to provide a motion-corrected image at the initial iteration, this motion is taken into consideration during the reconstruction process. To increase the likelihood, the motion estimate and the motion-corrected image are then updated alternately. The iterations are stopped when it appears that the updated motion has converged. There are two steps in the algorithm: (1) combined image and motion estimation, and (2) final reconstruction (motion compensation). Each iteration consists of 2 steps: a motion update and an image update. The OSEM algorithm is run through several iterations to update the image. With the last motion estimate, a final iterative reconstruction was carried out. The method was evaluated on physical phantom study, simulations phantom study and patient scans. In the physical phantom studies, the motion correction algorithms were applied to the phantom’s both stationary data and motion induced data. Most of motion blurring in the reconstructed images disappeared after the compensation on the motion induced data. Various simulated motion-induced data sets were produced from the simulation phantom data and calculated the Mean Square Difference (MSD) value. The average MSD values between two different data sets were found 933.25 and 1247.5. With the simulated motion-induced data, motion correction algorithms were applied. The quality of the reconstructed images was improved significantly after the compensation. In the physical phantom studies, the motion estimation for translational movements were found 7.6 mm in X axis, 7.1 mm in Y axis and 8.2 mm in Z axis. And it was also found that the heighest rotational motions were 12.3⁰ in X axis, 16.6⁰ in Y axis and -14.1⁰ in Z axis. The motion correction algorithms were applied to the clinical patient’s data also. And it was clearly seen that the motion corrected reconstructed images were significantly improved after the compensation. The proposed method effectively eliminated motion-induced artifacts in mentioned three types of SPECT data. The proposed method can be used retroactively on SPECT scans affected by motion because only measured raw data are required for motion estimate and compensation in this method. 2 The contributions of this research are summarized as the development of an innovative technique that can correct the six DOF motion in SPECT using the acquired data only. The methodology for six DOF data-driven motion estimation and correction in SPECT was developed and validated for using the digital and physical phantom studies as well as in clinical applications including human subjects. The developed technique was made automated and did not require any external equipment. It was not necessary to have any prior knowledge of motion. It can be readily adjusted to various types of collimation and multiple detector geometries.