Accuracy of retrospective image fusion of PET and CT in comparison to PET/CT: A patient study

Abstract

669

Objectives: The localisation of PET lesions is often difficult due to the limited morphological information. The problem can be solved by coregistration with CT images. For coregistration either a hardware approach (PET/CT) or a retrospective image fusion between dedicated PET and CT can be used. For clinical usage, the quantitative accuracy of the image fusion is important. Main factors that influence the accuracy are the time distance between the single examinations, difference of the patient storages and body motions (e.g. respiration). A priori PET/CT reduces the influence of these factors best and the smallest coregistration error can be expected. Therefore in this study PET/CT was used as gold standard for determining the coregistration error of the retrospective image fusion of dedicated PET and CT which was performed under optimal examination conditions.

Methods: 60 patients were examined with PET/CT (Biograph, Siemens) and with dedicated PET (ECAT HR+, Siemens) using F-18-FDG, Ga-68-Dotatoc or I-124 as tracers. An uniformed patient storage and a breathing protocol for the CT were used. Afterwards the CT of the PET/CT and the dedicated PET were coregistrated with a mutual based coregistration algorithm (HERMES workstation). An optimized coregistration protocol was applied (i.e., only coregistration of restricted body areas, tracer optimized weighting factors for transmission and emission). The coregistration error was determined by measuring the deviation between geometric- (or activity-centroids) of corresponding lesions in the coregistrated images. A coregistration quality control was carried out using landmark overlap assessment. In case of an error above 10 mm a manual correction in z- direction was followed.

Results: The coregistration error of the abdomen and thorax area was found to be approximately 13 mm for all investigated tracers. The largest error was always determined in z-direction and in tissue with greater distance to rigid body structures like the spine. If the coregistration quality assessment was sufficient, the measured coregistration error was less than 10 mm. The manual correction used in case of large errors always improved the alignment. The coregistration including data set preparation needed in mean 4 minutes.

Conclusions: Retrospective coregistration is possible with an coregistration error of about 13 mm using a tracer-adapted coregistration protocol. It can be used, if this accuracy is sufficient for the clinical question. A qualitative assessment of landmarks overlap can ensure the coregistration quality. In case of insufficient accuracy a manual correction in z- direction should be tried out.