Impact of Tissue Classification in MRI-Guided Attenuation Correction on Whole-Body Patlak PET/MRI

Abstract

Purpose: The aim of this work is to investigate the impact of tissue classification in magnetic resonance imaging (MRI)-guided positron emission tomography (PET) attenuation correction (AC) for whole-body (WB) Patlak net uptake rate constant (K i ) imaging in PET/MRI studies. Procedures: WB dynamic PET/CT data were acquired for 14 patients. The CT images were utilized to generate attenuation maps (μ-map CTAC ) of continuous attenuation coefficient values (A coeff ). The μ-map CTAC were then segmented into four tissue classes (μ-map 4-classes ), namely background (air), lung, fat, and soft tissue, where a predefined A coeff was assigned to each class. To assess the impact of bone for AC, the bones in the μ-map CTAC were then assigned a predefined soft tissue A coeff (0.1 cm −1 ) to produce an AC μ-map without bones (μ-map no-bones ). Thereafter, both WB static SUV and dynamic PET images were reconstructed using μ-map CTAC , μ-map 4-classes , and μ-map no-bones (PET CTAC, PET 4-classes , and PET no-bones ), respectively. WB indirect and direct parametric K i images were generated using Patlak graphical analysis. Malignant lesions were delineated on PET images with an automatic segmentation method that uses an active contour model (MASAC). Then, the quantitative metrics of the metabolically active tumor volume (MATV), target-to-background (TBR), contrast-to-noise ratio (CNR), peak region-of-interest (ROI peak ), maximum region-of-interest (ROI max ), mean region-of-interest (ROI mean ), and metabolic volume product (MVP) were analyzed. The Wilcoxon test was conducted to assess the difference between PET 4-classes and PET no-bones against PET CTAC for all images. The same test was also adopted to compare the differences between SUV, indirect K i , and direct K i images for each evaluated AC method.Results: No significant differences in MATV, TBR, and CNR were observed between PET 4classes and PET CTAC for either SUV or K i images. PET 4-classes significantly overestimated ROI peak , ROI max , ROI mean , as well as MVP scores compared with PET CTAC in both SUV and K i images. SUV images exhibited the highest median relative errors for PET 4-classes with respect to PET CTAC (RE 4-classes ): 6.91 %, 6.55 %, 5.90 %, and 6.56 % for ROI peak , ROI max , ROI mean , and MVP, respectively. On the contrary, K i images showed slightly reduced RE 4-classes (indirect 5.52 %, 5.95 %, 4.43 %, and 5.70 %, direct 6.61 %, 6.33 %, 5.53 %, and 4.96 %) for ROI peak , ROI max , ROI mean , and MVP, respectively. A higher TBR was observed on indirect and direct K i images relative to SUV, while direct K i images demonstrated the highest CNR. Conclusions: Four-tissue class AC may impact SUV and K i parameter estimation but only to a limited extent, thereby suggesting that WB Patlak K i imaging for dynamic WB PET/MRI studies is feasible. Patlak K i imaging can enhance TBR, thereby facilitating lesion segmentation and quantification. However, patient-specific A coeff for each tissue class should be used when possible to address the high inter...

Diagnostic Value of Dynamic 18F-Fluorodeoxyglucose Positron Emission Tomography-Computed Tomography (18F-FDG PET-CT) in Cervical Lymph Node Metastasis of Nasopharyngeal Cancer

“…All data were corrected according to isotope decay, scattering events, and random coincidence, and semi-maximum smoothing was performed using a standard Gaussian filter with a total width of 3 mm. Our research used internally developed computer code for indirect parameter reconstruction processing [16,17]. Assuming irreversible uptake of 18 F-FDG, the physiological parameters of voxel levels were estimated using the ordinary least squares (OLS) Patlak reconstruction regression method for 48 dynamic PET series based on a dual tissue compartment dynamics model [18].…”

Feasibility of Deep Learning-Guided Attenuation and Scatter Correction of Whole-Body 68Ga-PSMA PET Studies in the Image Domain

“…19,20 However, these MRI-guided AC approaches are prone to remarkable errors due to the presence of anatomic abnormalities, intra-atlas misregistration, tissue misclassification, body truncation, image artifacts, noise, and inhomogeneity in MRI scans. 16,21,22 The incorporation of time-of-flight information into the maximum likelihood reconstruction of attenuation and activity map method (MLAA) enabled the prediction of patient-specific AC maps from the PET emission data. This method often suffers from high noise levels, activity-attenuation cross-talk, and quantitative uncertainty of the estimated attenuation coefficients.…”

“…Atlas-based methods rely on the prior knowledge extracted from coregistered MRI and CT atlas image pairs to predict a synthetic CT of the target MRI scan 19,20 . However, these MRI-guided AC approaches are prone to remarkable errors due to the presence of anatomic abnormalities, intra-atlas misregistration, tissue misclassification, body truncation, image artifacts, noise, and inhomogeneity in MRI scans 16,21,22 …”