Evaluation of a 4D PET image reconstruction method with respiratory motion compensation in a patient study

Objectives The objective is to evaluate a 4D PET image reconstruction method with respiratory motion compensation for improved lesion detection in a patient study.

Methods In the 4D image reconstruction method, we estimated the patient’s respiratory motion by a group-wise B-spline non-rigid image registration method applied to respiratory-gated PET images. The estimated motions were used to transform the respiratory-gated PET images reconstructed with CT attenuation map for attenuation compensation (AC) to a reference frame and then averaged. In this study, we applied this method to the respiratory-gated PET data from the delayed 15-minutes PET scans of 3 patients with detectable lesions in the liver and/or lungs. We investigated motion estimations based on three types of respiratory-gated PET images reconstructed without AC, with uniform AC and with CT-AC to assess the effect of AC methods. For each patient, the motion-compensated PET image reconstructed by our method was compared to the image reconstructed the 3D OS-EM algorithm from the patient data without gating. The maximum standard uptake values (SUVmax) of the lesions were used as an index for evaluation.

Results A total of 15 small to medium sized lesions from the 3 patients were included in the data analysis. The SUVmax for these lesions in the PET images obtained from our 4D image reconstruction method modeling the motion estimation based on gated PET images without AC, with uniform AC and CT-AC, showed an average of 21%, 24% 17% improvement respectively over that of the conventional 3D method. Meanwhile the noise levels of the images from the two methods were about the same for the same patient.

Conclusions This patient study demonstrated that the 4D PET image reconstruction method with respiratory motion compensation improved the PET image quality in terms of lesion detectability. The highest improvements were observed for the 4D method modeling the motion estimations based on the respiratory-gated PET images with uniform AC, which was consistent with our previous simulation studies.

Research Support NIH grant R01 EB 00016