Abstract
359
Objectives We compare different tagged MR-based motion correction methods for PET image reconstruction in simultaneous PET-MR.
Methods Breathing motion was estimated from tagged MRI using three motion estimation approaches: a harmonic phase (HARP) tracking algorithm with 8 respiratory gates, HARP with 32 respiratory gates, and intensity based 4D B-spline image registration with 8 gates (B-spline). A dynamic torso phantom that mimics respiratory breathing and contains three 1-cm abdominal lesions (2:1, 4:1, and 6:1 tumor-to-background contrast (TBR)) was built and scanned in our dedicated simultaneous PET-MRI. A complimentary spatial modulation of magnetization (CSPAMM) tagged MRI sequence was used with HARP and a SPAMM sequence used with B-spline along with an inflating-deflating trigger cycle of 1 sec. The motion field obtained with each approach was incorporated into the system matrix using a listmode OSEM formalism (16 subsets, 6 iterations).
Results HARP with 32 gates and B-spline with 8 gates yielded the best improvement in image quality due to motion compensation for TBR (18-74% for HARP32 and B-spline8) and SNR (25-85% for HARP32, 15-69% for B-spline8). HARP with 8 gates yielded significantly worse performance (2-9% for TBR, -12-38% for SNR) because motion between adjacent gates was too large and violated the HARP conditions. CSPAMM for HARP required double the data acquisition time compared to SPAMM for B-spline. Less gates also yields faster MRI acquisition for 3D volume.
Conclusions HARP-based motion correction (32 respiratory gates) and 4D B-spline image registration (8 gates) yielded comparable substantial improvements in PET image quality compared to no correction. 4D B-spline is a promising approach for routine simultaneous PET-MR studies as it requires a reasonable number of respiratory gates consistent with the clinical setting