A spline model for RV myocardium sampling from cardiac PET images

Objectives To develop a highly automated RV segmentation, sampling, and analysis tool for human and small animal positron emission tomography (PET) imaging of molecular and pumping function.

Methods A spline-based model was developed to measure tracer activity in the mid-RV myocardium from PET images. Model fitting was automated by optimizing a constrained cost function, and operator intervention was implemented using a graphical user interface (GUI). Inter- and intra-operator variability of tracer activity measurements was evaluated using FDG PET images of 7 confirmed and 12 unconfirmed PH human subjects. Images were processed twice by each of two operators; a novice and an expert. Accuracy of RV cavity volumes and ejection fraction (RVEF) measurements from cardiac-gated PET images were evaluated by comparing with cardiac magnetic resonance imaging (CMR) in a subset of 5 PH patients.

Results 50% of images required operator intervention to sample the RV tracer activity. Intra-operator variability measured as the reproducibility coefficient (RPC) was 5.6% and 6.4% for novice and expert respectively (p=n.s.). The sampling locations between repeated measurements varied by 2.2±0.6 and 1.9±0.6 mm for novice and expert respectively (p=n.s.). Inter-operator RPC was 8.2% for tracer activity and sampling locations varied by 2.9±0.6 mm between repeated measurements. RV cavity volumes and RVEF agreed closely with CMR results (r²=0.954, n=10 and r²=0.965, n=5 respectively).

Conclusions This semi-automatic sampling of the RV myocardium in PET has performance characteristics comparable with established left ventricle analysis tools and is suitable for current research investigating RV molecular and cardiac function. Additional work is required to improve automation and evaluate molecular function quantification with dynamic imaging.