Correction for partial volume effect at different ratios between spill-in and spill-out

Objectives In the majority of clinical situations, the investigated organ or tumour (region of interest, ROI) is surrounded by a background with non-uniform activity. Moreover, some part of this background may have higher activity than the ROI. We evaluated the performance of the template-based method in correcting for partial volume (spill-in/spill-out) effects (PVE) in situations where ROI was adjacent to high-activity area.

Methods We simulated SPECT scans where two 40mL containers with different activities were placed side-by-side of-centre in a large cylindrical phantom. By changing their activities, we varied the amount of spill-in/spill-out through the boundary between containers while activity in the phantom was always kept lower than in any of the containers. In total, seven ROI with different levels of spill-in/spill-out were investigated. Our iterative template-based algorithm updates initially reconstructed (uncorrected for PVE) image by repeating the following steps: (1) generation of a template image T with uniform ROI activity and background determined in the previous reconstruction; (2) projection and reconstruction of T, (3) correction for PVE inside ROI; (4) projection of ROI sub-image and update of the background using OSEM with ROI-projections added to the forward step.

Results In the initial image, some parts of ROI were affected by spill-in (activity exceeded the true one) and others - by spill-out. Our method successfully resolved these problems for six out of seven considered containers. The activity distributions in ROI became more accurate (errors of total activities decreased from 12-24% to 1-7%) and uniform (standard deviations decreased from 25-30% to 7-9%). In the remaining case, spill-in prevailed over spill-out and the method failed after creating an improper template.

Conclusions Our iterative template-based method can correct for PVE even in situations where part of the surrounding activity exceeds up to 4 times the activity in ROI, thus it can be used to improve the accuracy of oncology scans