Abstract
654
Objectives: Simultaneous PET/MR system enables synchronized acquisition of anatomical and functional information of human body. Conventional MR acquisition in PET/MR procedure usually consists of both attenuation-estimation sequence and diagnostic sequence. To reduce motion artifacts in MR, MR navigator is integrated with the MR acquisition. The introduction of MR navigator enables accurate measurement of diaphragm location at high sampling rate, and hence convenient respiratory phase identification of both MR and PET data during the scan. Although methods have been proposed to use MR data derived respiratory motion vector field (MVF) to correct PET data, most of the proposed methods rely on additional non-diagnostic 4D MR acquisitions for MR based image registration and MVF determination, which leads to prolonged acquisition duration and/or reduced diagnostic MR sequence. The aim of this study is to propose a motion correction scheme for PET/MR data with only routine diagnostic sequence and attenuation sequence. Methods: Patients with liver and pancreatic cancer were recruited and imaged at Zhongshang Hospital on a newly developed simultaneous PET/MR system. For data acquisition, two 3D MR images were obtained at known respiratory phases with help of MR navigator. The first 3D image was a water-only image obtained using the Dixon method to provide attenuation coefficient map for PET data. The other was a diagnostic MR image acquired by T2-weighted single short fast spin echo sequence (SSFSE). These two acquisitions were designed to locate at two end respiratory phases—end-inspiration and end-expiration phase with the help of MR navigator. This protocol is a small modification of the routine MR protocol in PET/MR imaging where all MR sequences are triggered at the same respiratory phase. PET data were gated into several equal-count respiratory phases using MR navigator. The two MR images of different sequences were registered using a B-spline image registration algorithm with mutual information based cost function. Additional regularization was used during registration to reduce influence of noise. Interpolation of the MVF was achieved with motion information from navigators to generate MVFs for middle respiratory phases. The generated 4D MVF was then used for motion compensation of the PET data. Attenuation maps at different respiratory phases were generated by transforming the Dixon MR image using properly interpolated MVFs and gated PET reconstructions were transform to the same desired phase to improve image statistics.
Results: Application of our method on clinical data sets suggested that our proposed image registration was successful for the registration of two MR images acquired using two different MR sequences. Attenuation maps for different respiratory phases were successfully generated based on the estimated MVFs. Motion correction of gated PET images using MR-generated MVFs significantly improved image quality.
Conclusions: In summary, we proposed a motion correction method for PET/MR data with minimal modification of routine data acquisition protocol. MR-derived respiratory MVF was obtained from routine MR images of different sequences and was used to generate phase-matched attenuation maps and to correct respiratory motion of PET data. This method has advantage of using the whole PET dataset while in some tradition data processing methods, only PET data acquired at a quiescent period is used for minimal blurring at the cost of reduced photon counts. Compared with PET-based motion compensation method, this method provides more accurate MVF with the use of MR images. Different from conventional PET/MR motion correction methods, no extra MR sequence is required for our