MRI-Based Attenuation Correction for PET/MRI Using Multiphase Level-Set Method

Abstract

Inaccuracy in MR image-based attenuation correction (MR-AC) leads to errors in quantification and the misinterpretation of lesions in brain PET/MRI studies. To resolve this problem, we proposed an improved ultrashort echo time MR-AC method that was based on a multiphase level-set algorithm with main magnetic field (B 0 ) inhomogeneity correction. We also assessed the feasibility of this levelset-based MR-AC method (MR-AC level ), compared with CT-AC and MR-AC provided by the manufacturer of the PET/MRI scanner (MR-AC mMR ). Methods: Ten healthy volunteers and 20 Parkinson disease patients underwent 18 F-FDG and 18 F-fluorinated-N-3-fluoropropyl-2-β-carboxymethoxy-3-β-(4-iodophenyl)nortropane ( 18 F-FP-CIT) PET scans, respectively, using both PET/MRI and PET/CT scanners. The level-set-based segmentation algorithm automatically delimited air, bone, and soft tissue from the ultrashort echo time MR images. For the comparison, MR-AC maps were coregistered to reference CT. PET sinogram data obtained from PET/CT studies were then reconstructed using the CT-AC, MR-AC mMR , and MR-AC level maps. The accuracies of SUV, SUVr (SUV and its ratio to the cerebellum), and specific-to-nonspecific binding ratios obtained using MR-AC level and MR-AC mMR were compared with CT-AC using region-ofinterest-and voxel-based analyses. Results: There was remarkable improvement in the segmentation of air cavities and bones and the quantitative accuracy of PET measurement using the level set. Although the striatal and cerebellar activities in 18 F-FP-CIT PET and frontal activity in 18 F-FDG PET were significantly underestimated by the MR-AC mMR , the MR-AC level provided PET images almost equivalent to the CT-AC images. PET quantification error was reduced by a factor of 3 using MR-AC level ( The development of tomographic imaging technologies has made dramatic progress in recent decades. Among the modern medical imaging systems, PET and MRI have greatly contributed to understanding normal and abnormal brain functions and evaluating various neurologic disorders (1-4). Although PET is the most sensitive medical imaging device, providing both functional and biochemical information, it has limited spatial resolution, signalto-noise ratio, and anatomic information. Conversely, MRI offers detailed anatomic information about the brain along with excellent soft-tissue contrast and various types of hemodynamic information (i.e., perfusion and diffusion). Accordingly, the combination of PET and MRI can provide a 1-stop shop for clinical examination and new methodology for exploring the brain with multiparametric and complementary imaging information (5,6).In addition, fully integrated PET/MRI scanners based on semiconductor photosensors, such as avalanche photodiodes and silicon photomultipliers, allow the simultaneous acquisition of both image datasets, which possess several distinct advantages over the sequential scan in conventional PET/CT examinations (7-11). Accurate spatiotemporal correlation of PET/MRI signals permits the studies to demonstrat...

Generation of PET Attenuation Map for Whole-Body Time-of-Flight 18F-FDG PET/MRI Using a Deep Neural Network Trained with Simultaneously Reconstructed Activity and Attenuation Maps

“…A voxelwise scatterplot was plotted for comparison, and the correlation coefficients were calculated. The similarity of segmented regions from m-CNN and m-MLAA attenuation maps was compared using the Dice similarity coefficient (28,29). For the segmentation of m-CNN and m-MLAA, the sample thresholds used for generating m-segment were applied (0.015, 0.04, and 0.09 between air, lung, fat, and water).…”

Adaptive template generation for amyloid PET using a deep learning approach

“…For attenuation correction of PET, MR images were acquired simultaneously with PET using a dual‐echo UTE sequence (TE = 0.07 and 2.46 ms, TR = 11.9 ms, flip angle = 10°). The UTE images were reconstructed into a 192 × 192 × 192 matrix with an isotropic voxel size of 1.33 mm (An et al, ). A T1‐weighted 3D ultrafast gradient echo sequence was also acquired in a 208 × 256 × 256 matrix with voxel sizes of 1.0 × 0.98 × 0.98 mm.…”

[11C]-(R)-PK11195 positron emission tomography in patients with complex regional pain syndrome

“…A total of 27 dynamic PET frames (12 frames × 5 seconds, 5 × 60, 3 × 180, 5 × 300, and 2 × 600) were reconstructed using filtered back projection followed by 4-mm FWHM Gaussian postfiltering, with routine corrections for physical effects such as radioactive decay and attenuation (using an ultrashort-echo-time MR-based attenuation map). [29] The reconstructed individual frames consisted of 127 transaxial slices with a matrix size of 344 × 344, a pixel size of 1.0431 × 1.0431 mm 2 , and a slice thickness of 2.0313 mm.…”