Attenuation correction in 4D‐PET using a single‐phase attenuation map and rigidity‐adaptive deformable registration

Abstract

Purpose Four-dimensional positron emission tomography (4D-PET) imaging is a potential solution to the respiratory motion effect in the thoracic region. Computed tomography (CT)-based attenuation correction (AC) is an essential step toward quantitative imaging for PET. However, due to the temporal difference between 4D-PET and a single attenuation map from CT, typically available in routine clinical scanning, motion artifacts are observed in the attenuation-corrected PET images, leading to errors in tumor shape and uptake. We introduced a practical method to align single-phase CT with all other 4D-PET phases for AC. Methods A penalized non-rigid Demons registration between individual 4D-PET frames without AC provides the motion vectors to be used for warping single-phase attenuation map. The non-rigid Demons registration was used to derive deformation vector fields (DVFs) between PET matched with the CT phase and other 4D-PET images. While attenuated PET images provide useful data for organ borders such as those of the lung and the liver, tumors cannot be distinguished from the background due to loss of contrast. To preserve the tumor shape in different phases, an ROI-covering tumor was excluded from non-rigid transformation. Instead the mean DVF of the central region of the tumor was assigned to all voxels in the ROI. This process mimics a rigid transformation of the tumor along with a non-rigid transformation of other organs. A 4D-XCAT phantom with spherical lung tumors, with diameters ranging from 10 to 40 mm, was used to evaluate the algorithm. The performance of the proposed hybrid method for attenuation map estimation was compared to 1) the Demons non-rigid registration only and 2) a single attenuation map based on quantitative parameters in individual PET frames. Results Motion-related artifacts were significantly reduced in the attenuation-corrected 4D-PET images. When a single attenuation map was used for all individual PET frames, the normalized root mean square error (NRMSE) values in tumor region were 49.3% (STD: 8.3%), 50.5% (STD: 9.3%), 51.8% (STD: 10.8%) and 51.5% (STD: 12.1%) for 10-mm, 20-mm, 30-mm and 40-mm tumors respectively. These errors were reduced to 11.9% (STD: 2.9%), 13.6% (STD: 3.9%), 13.8% (STD: 4.8%), and 16.7% (STD: 9.3%) by our proposed method for deforming the attenuation map. The relative errors in total lesion glycolysis (TLG) values were −0.25% (STD: 2.87%) and 3.19% (STD: 2.35%) for 30-mm and 40-mm tumors respectively in proposed method. The corresponding values for Demons method were 25.22% (STD: 14.79%) and 18.42% (STD: 7.06%). Our proposed hybrid method outperforms the Demons method especially for larger tumors. For tumors smaller than 20 mm, non-rigid transformation could also provide quantitative results. Conclusion Although non-AC 4D-PET frames include insignificant anatomical information, they are still useful to estimate the DVFs to align the attenuation map for accurate AC. The proposed hybrid method can recover the AC-related artifacts and provide quantitative ...

Correcting CT misregistration in data‐driven gated (DDG) PET with PET self‐gating and deformable image registration

“…9 Specifically, there are research efforts to register PET and CT using deformable image registration (DIR). 10 However,direct,non-rigid image registration between CT images with high resolution and low noise, and PET images with low resolution and high noise, remains challenging. A recent alternative approach proposed was to register the CT to a reference phase selected from multiple DDG PET phases.…”

“…Respiratory motion models based on image registration algorithms have been previously proposed and applied to medical images for motion correction 9 . Specifically, there are research efforts to register PET and CT using deformable image registration (DIR) 10 . However, direct, non‐rigid image registration between CT images with high resolution and low noise, and PET images with low resolution and high noise, remains challenging.…”

Penalized PET/CT Reconstruction Algorithms with Automatic Realignment for Anatomical Priors

“…To adapt to irregular breathing patterns, another strategy applies individual motion model to deform the input attenuation map [23]. The model can be derived from other imaging data, such as dynamic CT and MR [24], as well as the nonattenuation corrected PET [25]. However, the former approach has the potential problem of propagating the error in the model estimation to the final reconstructed image, while the latter method's performance depends on the tracer distribution and data statistics.…”

Issues in quantification of registered respiratory gated PET/CT in the lung

“…Several gates can be combined with registration, either during or after image reconstruction, to generate one single motion corrected PET image Erdi 2008, Rahmim et al 2013). To avoid the increased dose to the patient associated with gated CT, several authors deform a single CT to match the PET gates, for instance based on registering non-attenuation corrected (NAC) PET images (Wells et al 2010, Kalantari andWang 2017) or during reconstruction (Bousse et al 2016, Rezaei et al 2016. However, conventional registration methods follow anatomical tissue deformation and account for spatial mismatch, but do not correct for density and concentration changes associated with expansion.…”