18F-FDG-PET and MRI in patients with malignancies of the liver and pancreas. Accuracy of retrospective multimodality image registration by using the CT-component of PET/CT.

PURPOSE To evaluate the accuracy of retrospective rigid image registration and fusion between F-18 fluorodeoxyglucose positron emission tomography (FDG-PET) and magnetic resonance imaging (MRI) of the upper abdomen. PATIENTS, MATERIAL, METHODS Image fusion of PET and MRI was performed in 30 patients with suspected malignancy of the liver or pancreas. Using a commercially available image fusion tool capable of rigid manual point-based registration, PET-Images were retrospectively registered and fused by matching eight homologous points in the 3D spoiled gradient echo (GRE) MRI sequences acquired in portal venous phase and in the CT-component of PET/CT. Two separate observers (R1, R2) assessed accuracy of image registration by determining the distances in the x-, y- and z-axis as well as the absolute distance between anatomical landmarks which differed from the landmarks chosen for registration. Quality of fusion was graded using a three point grading scale (1 poorly fused; 2 satisfactory fused; 3 correctly fused) and compared to hybrid PET/CT fusion. RESULTS Mean time of registration per patient was less than 2 minutes. Objective registration assessment showed errors between 2.4-6.3 mm in x-axis: mean 3.6 mm (R1); 4.6 mm (R2), 2.3-9.3 mm in y-axis (mean 5.1 mm; 5.5 mm) and 3.3-12.0 mm in z-axis (mean 5.9 mm; 5.9 mm.) The mean error in absolute distance between points was 6.0-16.8 mm (mean 9.9 mm; 10.6 mm). In visual assessment, most fusions were graded to be satisfactory or correctly fused: R1, R2: grade 3, 11/30 (36.7%), 22/30 (73.3%); grade 2, 13/30 (43.3%), 8/30 (26.7%); grade 1, 6/30 (20%), 0/30 (0%). Fusions were mostly comparable to hybrid PET/CT fusions. All of the fusions were defined as diagnostically relevant by both observers. CONCLUSION Retrospective rigid image fusion of FDG-PET and MRI of the upper abdomen using the CT-component of PET/CT for registration is feasible without adaptation in image acquisition protocols and shows sub-centimeter registration errors in most cases.