Feasibility of shortening the ¹⁸F-FDG Patlak scan time in a high-sensitivity short-axial field-of-view positron emission tomography-computed tomography system for oncological studies using deep learning denoising algorithms

Introduction

Whole-body (WB) ¹⁸F fluorodeoxyglucose (¹⁸F-FDG) positron emission tomography-computed tomography (PET/CT) imaging is commonly employed for the clinical assessment of malignant, infectious, and inflammatory diseases. Standardized uptake value (SUV) is the most commonly used semiquantitative metric for ¹⁸F-FDG uptake assessment and is acquired with a single-pass WB scan. The net influx rate (Ki), a quantitative measure of ¹⁸F-FDG metabolism, reduces the incidence of false-positive findings of nonspecific ¹⁸F-FDG uptake as compared to the SUV, thereby improving lesion detectability and disease characterization (1-4). Combining K_i, with SUV allows physicians to accurately assess the metabolic activity of tissues and adopt the optimal treatment strategy. K_i images are calculated from multipass WB scans via Patlak graphic analysis (5). However, calculating the Ki metric poses certain challenges, primarily the long scan duration required, which can be uncomfortable for patients and difficult to integrate into busy clinical workflows. Scanning time optimization in Patlak analysis has become an area of particular research interest in recent years, with the objective being to improve accuracy while minimizing the patient discomfort associated with prolonged scanning segments through the use of various abbreviated protocols appropriate for Patlak analysis (6-9). The challenge of reducing scanning time while maintaining the Patlak K_i accuracy for ¹⁸F-FDG WB-PET depends on the following key factors: (I) arterial input function (AIF) estimation; (II) the number of scan passes required for the Patlak analysis; and (III) the image quality for each pass.

Several strategies can be used to optimize scanning time for AIF generation while maintaining the accuracy of estimated AIF. Although arterial blood sampling is considered the gold standard, it is rarely used in clinical settings because it is invasive and can be uncomfortable for patients. Image-derived AIF in the aorta acquired for more than 60 min in a modern high-sensitivity PET/CT system, particularly a long-axial field-of-view (LAFOV) PET/CT system, is also acceptable as the image-derived input function (IDIF) for ¹⁸F-FDG Patlak analysis (10). Recent Patlak studies in LAFOV PET/CT systems have demonstrated that it is feasible to reduce the scanning time by dividing the whole scanning process into segments, with one segment being dedicated to the acquisition of fast uptake AIF in the aorta, and the others being used for the multipass WB scanning in Patlak fitting (6-8,11). The transition period between segments allows patients to leave their beds, making for a more comfortable experience.

Other options for shortening the WB Patlak scanning time have been explored, including population-based input function (PBIF). In recent years, modern high-sensitivity short-axial field of view (FOV) or LAFOV PET/CT systems have facilitated the development of more accurate PBIFs compared to older analog PET/CT systems. For instance, it was reported that the use of abbreviated scan protocols with PBIFs for accurate kinetic modeling of ¹⁸F-FDG datasets from a LAFOV PET scanner could improve the estimation of kinetic parameters (12). This increase in sensitivity allows for parametric imaging with high temporal resolution, as well as clinically acceptable parametric imaging via a PET scanner with a short-axial FOV (13,14).

Other factors, such as the image quality of each dynamic pass and the number of passes used for Patlak fitting, significantly affect the accuracy of Patlak K_i values. The image quality of each dynamic pass can be enhanced by either increasing the acquisition time or employing advanced denoising algorithms, such as those based on deep learning denoising algorithms. The use of more dynamic passes for Patlak fitting adds to the reliability and accuracy of K_i values (15-17). However, this requires a longer acquisition time, which may not be feasible in routine clinical settings. Therefore, for any solution that shortens the Patlak acquisition time, there should be a balance between the image quality of each pass, total acquisition time, and the number of dynamic passes for Patlak fitting in order to meet routine clinical needs.

Studies have shown the ability of deep learning algorithms to reduce the acquisition time while preserving the accuracy and reliability of Patlak K_i values even when short-time frames in the LAFOV PET/CT system are used (15,18). Therefore, we conducted this study to assess the performance of various abbreviated ¹⁸F-FDG Patlak protocols and identify the one most suited to routine clinical procedures.

Methods

Participants and the dynamic WB ¹⁸F-FDG PET scan protocol

This study was approved by the Institutional Review Board of the China-Japan Friendship Hospital (approval No. 2023-KY-086), and was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. A total of 14 patients were enrolled in this study after providing written informed consent. Fourteen participants completed the standard protocol (i.e., segments I to III, with a range of 0–75 min) used for conventional Patlak analysis. Twelve participants successfully completed the full dual-injection protocol (segments I to IV: 0–86 min). Two participants did not complete the full protocol due to physical discomfort from prolonged lying. The participants received ¹⁸F-FDG bolus injection intravenously, with 323±71 MBq being delivered manually by trained technologists with an injection volume of approximately 1 mL completed within 30 seconds. In total, 12 FDG-avid lesions were manually segmented. The demographics of participants and the injection doses are presented in Table 1. All participants underwent scanning by a high-sensitivity PET/CT system with a 26-cm axial field of view (Biograph Vision 600, Siemens Healthineers, Erlangen, Germany). This dual-time injection strategy was specifically designed for this study to enable the simulation and evaluation of multiple Patlak protocols within a single scan session and not for routine clinical use.

To accommodate all existing AIF acquisition methods or Patlak protocols, the dynamic WB ¹⁸F-FDG PET scan included a dual-time tracer injection protocol. The first injection of 80% ¹⁸F-FDG tracer was followed by AIF acquisition in the chest region (0–6 min p.i.) and a multipass WB scan from the skull base to the thigh (7–75 min p.i.). The remaining 20% ¹⁸F-FDG tracer was then injected for the second time, followed by a 6-minute single-bed AIF acquisition in the chest region that began after about 5 min of rest. This dual-time injection protocol was separated into four segments: (I) a first single-bed AIF scan in the chest region (0–6 min p.i.; reconstructed into 26 frames as follows: 12×5 s + 6×10 s + 8×30 s); (II) an intermediate WB scan from skull base to thigh (7–40 min p.i.; reconstructed as follows: 7×2 min + 4×5 min); (III) a dynamic WB scan for Patlak from the skull base to the thigh (41–75 min p.i.; reconstructed with 5 min/pass); and (IV) second single-bed AIF scan in the chest region (80–86 min p.i.; reconstructed into 26 frames as follows: 12×5 s + 6×10 s + 8×30 s). The ¹⁸F-FDG dynamic frames were reconstructed with parameters listed in Table 2, and the Patlak protocols with different configurations of AIF and number of WB passes for Patlak fitting were defined as detailed in Figure 1. The factor that differed between the protocols was the IDIF derivation method. The Patlak fitting was completed during the same multipass WB scan, which lasted from 41 to 75 min after the first ¹⁸F-FDG tracer injection (80% of the normal dose). Protocol 1 was treated as the standard protocol and consisted of segments I, II, and III, representing the standard Patlak with a full-sampling image-derived AIF; protocol 2 consisted of segments I and III, representing the interleave-sampling image-derived AIF; protocol 3 consisted of segments III and IV, representing the continuously but partially resampled image-derived AIF; and protocol 4 consisted of only segment III with the scaled PBIF as its IDIF. In summary, participants were required to complete only one single comprehensive dual-injection scan in 1 day. All protocols were derived from this single dual-injection scan through the segmentation and combination of the reconstructed dynamic frames.

Figure 1 Four dynamic whole-body Patlak imaging protocols were derived from a dual-injection FDG-PET scanning method. The initial injection occurred at time zero, with the second injection being administered approximately 80 min later. The 41 to 75 min interval was used for Patlak analysis, while the remaining periods (0–6, 7–40, and 80–86 min) were designated for generating various types of AIF. AIF, arterial input function; FDG, fluorodeoxyglucose; PET, positron emission tomography.

Patlak processing workflow

The whole Patlak processing workflow consisted of four steps (Figure S1): (I) deep learning-based FDG-PET denoising; (II) automatic deformative motion correction of dynamic passes; (III) generation of IDIF; and (IV) Patlak fitting. All the dynamic passes were converted into the SUV before processing.

Deep learning-based ¹⁸F-FDG PET denoising and motion correction

A deep learning-based WB ¹⁸F-FDG denoising algorithm was applied to dynamic passes acquired during segment III for Patlak fitting. This denoising network employed a unified noise-aware architecture to adapt to the different noise levels that occur frequently during dynamic acquisition. Both external and internal training datasets were included. The external datasets consisted of 510 participants with a 50% and 100% full-dose WB FDG-PET scan collected from the Ultra-low Dose PET Challenge (https://ultra-low-dose-pet.grand-challenge.org), which includes different levels of low-count FDG-PET and full-dose FDG-PET scans acquired from the Biograph Vision Quadra (Siemens Healthineers) and the uEXPLORER (United Imaging Healthcare, Shanghai, China) scanners. The internal datasets consisted of 100 participants with paired 5 min (representing low counts) and 10 min (representing full dose) per WB FDG-PET pass acquired in the study hospital from the Biograph Vision-600 PET/CT scanner (Siemens Healthineers). This denoising network (see Figure S2 for details) was slightly modified from the published Unified Noise-aware Network (19) and handled four noise levels instead of the original six levels to meet our limited computing resources for deep learning model training. The loss definition, training strategy, and optimization used were from a previous work (19). After the denoising procedure, a fully automated and fast deformative motion correction algorithm (20) was applied to the dynamic WB FDG-PET passes for Patlak fitting (segment III). The last pass was used as the reference.

AIF generation and Patlak fitting

The AIF for various protocols was measured directly from dynamic frames or interpolated by exponentially interpolating population-based AIF into dynamic frame measurements. Two experience clinicians calculated the AIF for protocol 1 by averaging the voxels within a manually drawn region of interest (ROI) with a diameter of 1 cm over the aorta. The population-based AIF was calculated by averaging all participants’ protocol 1 IDIFs and was normalized via the area under the curve (AUC) values of counted dynamic frames (41–75 min) from the same ROI. The IDIF of protocol 2 was interpolated from the signals measured at segments I and III via the nonlinear curve fitting function (nonlinear fitting, SciPy package, version 1.11.2, https://scipy.org/) with parameters fitted from PBIF. For protocol 3, the IDIF was determined through use of the data obtained from segment III and the residual corrected data from segment IV, with the residual signal of the first injection being subtracted from the second injection through extrapolation of the first injected AIF signal in the aorta region (9). Subsequently, the IDIF of protocol 3 was interpolated from the corrected signals in segment IV and the signal in segment III, which facilitated the examination of the same method as that used in protocol 2.

The voxel-wise Patlak fitting was performed from 41 to 75 min p.i., with the configuration of 3, 4, 5, 6, and 7 WB passes (each 5 min in duration) representing 15, 20, 25, 30, and 35 min of WB scanning, respectively, which facilitated evaluation of the optimized scanning time for a clinically applicable Patlak protocol. These configurations were defined by varying t* while maintaining a fixed endpoint at 75 min p.i. The entirety of the AIF interpolation and Patlak fitting processes were performed with an open-source toolbox (NiftyPAD) (21).

Quantitative comparison and statistical analysis

Lesion volumes of interest (VOIs) were manually delineated by two experienced nuclear medicine physicians in consensus using fused PET/CT images. In patients without visible FDG-avid lesions (n=2), the Ki values were not included in the lesion-based statistics but were instead used for PBIF validation and image quality assessment. The quantitative values generated from abbreviated protocols (protocols 2, 3, and 4) were compared to the standard Patlak protocol (protocol 1) using the correlation analysis (Pearson correlation analysis) and the Bland-Altman analysis for difference. The compared quantitative values are listed in Table 3. The dynamic passes from 41 to 75 min p.i. were used for this type of comparison. Moreover, the effect of further reduction in the acquisition time (decrease in the number of WB passes to 3 with an averaged WB acquisition time of 5 min) on the accuracy of K_i values in the abbreviation protocols was also evaluated through a comparison with the K_i values obtained from the standard Patlak imaging protocol (protocol 1 with dynamic passes from 41 to 75 min for Patlak fitting). All statistical analyses were performed in GraphPad Prism 8 (Dotmatics, Boston, MA, USA).

Results

Effect of AIF acquisition time reduction on AIF accuracy

Fourteen participants completed the standard scanning protocol (segments I to III), and 12 completed the full dual-injection protocol (segments I to IV). Therefore, AIF analysis and Ki estimation involving protocol 3 were based on data from these 12 participants who completed all required scan segments. The mean AIF of all participants scaled by the tail portion of IDIF (41–75 min) was defined as the PBIF of this study, which was further normalized by the peak PBIF as shown in the left axis of Figure 2A. The difference between abbreviation protocols and the standard protocol (protocol 1) can be also found in the right axis of Figure 2A. The abbreviated protocols had a relatively lower AUC as compared to the acquired IDIF (Figure 2B), which could lead to an overestimation of Patlak K_i values. In order to quantitatively evaluate these differences and test whether they were statistically significant, the differences between the standard protocol (protocol 1) and the abbreviated protocols (protocols 2, 3, and 4) were evaluated via Bland-Altman, with the results being the following: protocol 2 bias, 3.24%±6.24% (limits of agreement −9.17% to 15.64%); protocol 3 bias, 0.243%±6.302% (limits of agreement −12.11% to 12.60%); and protocol 4 bias, 0.816%±7.40% (−12.68% to 15.31%).

Figure 2 Comparison of AIFs between different protocols. (A) Plot of normalized PBIF generated from protocol 1 (left axes) and the difference in AIFs of protocols 2, 3, and 4 as compared to protocol 1 (right axes); the inset plot shows the 0- to 5-min mean AIFs for protocols 1–4. (B) The Bland-Altman plot of the percentage of difference (%) between the individual AIF AUCs (0–75 min) of protocols 2, 3, and 4 and that of protocol 1. AIF, arterial input function; AUC, area under curve; PBIF, population-based input function.

Effects of AIF and acquisition time on Patlak K_i accuracy

The accuracy of Patlak K_i was quantitatively evaluated at different AIF generation configurations and effective acquisition times for Patlak fitting. In this study, the WB scanning duration was around 5 min and the acquisition configuration for effective Patlak fitting was 15–35 min, corresponding to 3–7 passes of the WB scanning. The Patlak K_i measured from the manually contoured VOIs of FDG-avid lesions was grouped for agreement and difference analysis. The agreement between different configurations was compared to the standard protocol defined in this study, in which the AIF generation was directly measured from dynamic frames (protocol 1) and the 41–75 min p.i. (7 passes with 5 min/pass) was used for Patlak fitting. The correlation and difference between different configurations were compared to the standard protocol defined in this study. Figure 3 presents the result of Pearson correlation analysis, with Figure 3A,3B displaying the correction factors and the correlation plot, respectively. Figure 4 also shows the result of the Bland-Altman analysis. Both correlation and Bland-Altman analyses showed that using only 3 passes resulted in significantly lower correlation coefficients and higher bias as compared to using 4 or more passes.

Figure 3 Pearson correlation analysis of K_i values obtained from protocol 1 and protocols 2, 3, and 4 with all possible whole-body pass configurations for all ¹⁸F-FDG-avid lesions, with protocol 1 with 7 whole-body passes serving as the reference. (A) Matrix (protocol vs. number of WB passes for Patlak fitting) plot of the Pearson correlation coefficient r. (B) Linear fitting of K_i values obtained from protocols 3 and 4 with 7, 6, 5, 4, and 3 WB passes, compared to protocol 1 with 7 WB passes. FDG, fluorodeoxyglucose; WB, whole-body.

Figure 4 Bland-Altman analysis compared K_i values from all FDG‑avid lesions across different protocols against the reference standard (protocol 1 with 7 whole‑body passes). (A) The percentage of bias. (B) The SD of bias (%). (C) Bland-Altman plot of the K_i generated from all WB passes from protocol 4 as compared to the K_i generated from the reference standard (protocol 1 with 7 WB passes). FDG, fluorodeoxyglucose; SD, standard deviation; WB, whole-body.

Figure 5 shows the WB Patlak K_i maximum intensity projection (MIP) images with various passes under protocol 1 and protocol 4 and the SUV MIP images of the seventh pass (71–75 min after ¹⁸F-FDG administration) for a participant with left upper-lobe lung cancer with left hilar lymphatic node metastasis. Visually, the lesion-to-background contrast of the K_i images with three passes was lower than that of the other four K_i images derived from both protocols. The contrast in the K_i images with 4–7 passes was comparable when both input functions were used.

Figure 5 A 65-year-old woman patient with left upper-lobe non-small cell lung cancer with left hilar lymph node metastasis. Example of a visual comparison of Patlak K_i MIP images under (A) protocol 1 and (B) protocol 4 and (C) the SUV MIP. MIP, maximum intensity projection; SUV, standardized uptake value.

Discussion

This study evaluated the feasibility and accuracy of Patlak K_i in a clinical routine setting, as clinical routine requires a relatively short scanning time. The generation of Patlak K_i images for oncological purposes requires AIF and several WB passes, which necessitates longer scanning times. Furthermore, the accuracy of Patlak K_i is influenced by several factors: (I) the fidelity of the AIF; (II) the number of passes used for Patlak fitting, which establishes the temporal resolution and the extent of the linear Patlak plot; (III) the noise level of each pass, which affects the quality and stability of the Patlak fit; and (IV) the motion artifacts among different passes, which can distort the spatial alignment and registration of the PET images.

These factors must be meticulously considered and controlled in the application of Patlak K_i to dynamic PET data analysis. Among these factors, both AIF and the number of frames are directly related to the potential for reducing scanning time to achieve the clinically routine application of Patlak imaging in short-axial FOV PET/CT systems. This study also examined the effect of AIF and scanning passes on the accuracy of K_i by controlling the noise level and motion during dynamic scanning.

Although we used a high-sensitivity PET/CT system, the further reduction of noise level for each pass exerted an impact on the final noise level of K_i images. Moreover, we applied deep learning-based denoising algorithms to further reduce the noise of fitted K_i images. This denoising led to a significant increase in the signal-to-noise ratio (SNR) of dynamic images, as demonstrated in Figure S3. The measurements were based on liver SNR, obtained via a manually drawn spherical VOI with a diameter of 3.0 cm. A comparison between applying and not applying denoising was beyond the scope of this study and is not discussed here. The details of this deep learning-based denoising model are outlined in Figure S2. This model was trained on datasets obtained from high-sensitivity LAFOV PET/CT systems, as detailed in the study by Xie et al. (19). The WB automatic motion correction method developed by Sundar et al. was also employed to correct the interframe motion (20).

As the AIF generation method affects the accuracy of K_i, the AIF estimation from abbreviated protocols (protocol 2, 3, and 4) was compared to the standard protocol (protocol 1). As shown in Figure 2A,2B, there was a strong correlation between protocol 1 and protocols 2, 3, and 4, and the protocols demonstrated no significant differences in terms of AUC (0–75 min). These results indicate that compared to the standard protocol, the protocols 2–4 can provide reliable AIFs to meet the dynamic analysis prerequisites of the Patlak model.

Figure 3 illustrates the correlation of K_i generated from protocols 1, 2, 3, and 4 for all FDG-avid lesions. The data indicates that the coefficient r increases with the number of passes (duration) for each protocol, suggesting that a longer scanning time results in a more accurate K_i of the lesion. When the protocols were compared, vertically, the K_i of the same passes, ranging from 3 to 7 passes, showed good agreement and minimal difference (less than 0.01), indicating that each protocol can reliably estimate K_i images.

The study also demonstrated that the number of passes significantly affects the accuracy of K_i, especially for the PBIF method. Insufficient passes in a fixed but short period may result in inadequate data points or poor linearity, while too many of them in a fixed but short period can introduce noise for each pass even in when denoising algorithms are applied. As shown in Figure 4, when the number of passes used for Patlak fitting increased for protocols 1, 2, and 3, both the bias and precision [standard deviation (SD) of bias] of K_i decreased. For instance, as presented in protocol 3, the bias and precision of K_i for all FDG-avid lesions decreased from 3.183% and 17.93% at 3 passes to 0.2759% and 0.561% at 7 passes, respectively. However, the PBIF approach (protocol 4) showed noticeable deviation from other protocols in terms of bias and precision. The lower Ki estimation in protocol 4 compared to other protocols might be caused by inaccuracies in the AIF estimation. These inaccuracies stemmed from the population-based method, which relied on a fitting process using only a partially measured AIF. Although no statistically significant differences were found in the AUC of the AIFs across protocols, a systematic bias remained. This bias likely resulted from estimating the individual AIF by scaling a population-based model with a limited number of sampling time points. Despite the underestimation for K_i by protocol 4, the acceptable bias and precision of 4 and 5 passes—with a low bias (<3%) and high precision (<15%) could still be acquired, which is consistent with previous reports (12,13) and supports both the reliability of our abbreviated protocol results and their potential application in scientific and clinical settings. In this study, by comparing experimental protocols with a standard protocol in a small patient group, we successfully generated a PBIF and validated the PBIF based on the Patlak protocol. However, the clinical benefit of this PBIF protocol remains to be evaluated in a multicenter study.

When the images from protocols 1 and 4 were compared, the Patlak K_i images with various passes generated with standard IDIF and PBIF and static SUV MIP images from the seventh pass showed good similarity at each scanning time (Figure 5). Visually, the K_i images from the three passes exhibited relatively poor quality, with more noise and less contrast between the lesion and the background. On the other hand, the parametric images from 4 to 7 passes showed good quality and could ensure lesion measurement and aid in the more accurate definition of smaller lesions. The most significant limitation of this study is the small sample size, which restricts the statistical power of comparisons and limits the generalizability of the derived PBIF. Although the initial results are promising, future studies with larger, more diverse patient populations are needed to validate the robustness and clinical utility of the abbreviated protocols. Secondly, the effect of denoising on both IDIF and PBIF K_i images and K_i values was not systematically evaluated. Furthermore, we did not account for the movement within the frame, which was often caused by physiological motion such as breathing. This aspect should be addressed in future studies.

Conclusions

The study findings suggest that the abbreviated protocols can be used for accurate Patlak K_i estimation from high-sensitivity short-axial FOV PET/CT systems. The results showed that the use of a PBIF protocol with 4 passes, enhanced by deep learning-based denoising algorithm, can provide a low bias (<3%) and high precision (<15%) in the estimation of K_i in patients with oncological diseases. This can be achieved in routine clinical procedures in a widely used high-sensitivity short-axial FOV PET/CT system.

Acknowledgments

None.

Footnote

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1757/dss

Funding: This study was financially sponsored by National Natural Science Foundation of China (No. 82071963), Beijing Municipal Natural Science Foundation (No. 7242128), and National High Level Hospital Clinical Research Funding (No. 2025-NHLHCRF-YXHZ-ZD-01).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1757/coif). L.F. reports funding from National Natural Science Foundation of China (No. 82071963), Beijing Municipal Natural Science Foundation (No. 7242128), and National High Level Hospital Clinical Research Funding (No. 2025-NHLHCRF-YXHZ-ZD-01). The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Review Board of the China-Japan Friendship Hospital (approval No. 2023-KY-086) and informed consent was taken from all individual participants.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Sundaraiya S, T R, Nangia S, Sirohi B, Patil S. Role of dynamic and parametric whole-body FDG PET/CT imaging in molecular characterization of primary breast cancer: a single institution experience. Nucl Med Commun 2022;43:1015-25. [Crossref] [PubMed]
2. Zaker N, Kotasidis F, Garibotto V, Zaidi H. Assessment of Lesion Detectability in Dynamic Whole-Body PET Imaging Using Compartmental and Patlak Parametric Mapping. Clin Nucl Med 2020;45:e221-31. [Crossref] [PubMed]
3. Yang M, Lin Z, Xu Z, Li D, Lv W, Yang S, Liu Y, Cao Y, Cao Q, Jin H. Influx rate constant of (18)F-FDG increases in metastatic lymph nodes of non-small cell lung cancer patients. Eur J Nucl Med Mol Imaging 2020;47:1198-208. [Crossref] [PubMed]
4. Kaneko K, Nagao M, Yamamoto A, Yano K, Honda G, Tokushige K, Sakai S. Patlak Reconstruction Using Dynamic 18 F-FDG PET Imaging for Evaluation of Malignant Liver Tumors : A Comparison of HCC, ICC, and Metastatic Liver Tumors. Clin Nucl Med 2024;49:116-23. [Crossref] [PubMed]
5. Dias AH, Pedersen MF, Danielsen H, Munk OL, Gormsen LC. Clinical feasibility and impact of fully automated multiparametric PET imaging using direct Patlak reconstruction: evaluation of 103 dynamic whole-body (18)F-FDG PET/CT scans. Eur J Nucl Med Mol Imaging 2021;48:837-50. [Crossref] [PubMed]
6. van Sluis J, van Snick JH, Brouwers AH, Noordzij W, Dierckx RAJO, Borra RJH, Lammertsma AA, Glaudemans AWJM, Slart RHJA, Yaqub M, Tsoumpas C, Boellaard R. Shortened duration whole body (18)F-FDG PET Patlak imaging on the Biograph Vision Quadra PET/CT using a population-averaged input function. EJNMMI Phys 2022;9:74. [Crossref] [PubMed]
7. Chen Z, Cheng Z, Duan Y, Zhang Q, Zhang N, Gu F, Wang Y, Zhou Y, Wang H, Liang D, Zheng H, Hu Z. Accurate total-body K(i) parametric imaging with shortened dynamic (18) F-FDG PET scan durations via effective data processing. Med Phys 2023;50:2121-34. [Crossref] [PubMed]
8. Wu Y, Feng T, Zhao Y, Xu T, Fu F, Huang Z, Meng N, Li H, Shao F, Wang M. Whole-Body Parametric Imaging of (18)F-FDG PET Using uEXPLORER with Reduced Scanning Time. J Nucl Med 2022;63:622-8. [Crossref] [PubMed]
9. van Sluis J, van Snick JH, Glaudemans AWJM, Slart RHJA, Noordzij W, Brouwers AH, Dierckx RAJO, Lammertsma AA, Tsoumpas C, Boellaard R. Ultrashort Oncologic Whole-Body (18)FFDG Patlak Imaging Using LAFOV PET. J Nucl Med 2024;65:1652-7. [Crossref] [PubMed]
10. Volpi T, Maccioni L, Colpo M, Debiasi G, Capotosti A, Ciceri T, Carson RE, DeLorenzo C, Hahn A, Knudsen GM, Lammertsma AA, Price JC, Sossi V, Wang G, Zanotti-Fregonara P, Bertoldo A, Veronese M. An update on the use of image-derived input functions for human PET studies: new hopes or old illusions? EJNMMI Res 2023;13:97. [Crossref] [PubMed]
11. Wu Y, Feng T, Shen Y, Fu F, Meng N, Li X, Xu T, Sun T, Gu F, Wu Q, Zhou Y, Han H, Bai Y, Wang M. Total-body parametric imaging using the Patlak model: Feasibility of reduced scan time. Med Phys 2022;49:4529-39. [Crossref] [PubMed]
12. Sari H, Eriksson L, Mingels C, Alberts I, Casey ME, Afshar-Oromieh A, Conti M, Cumming P, Shi K, Rominger A. Feasibility of using abbreviated scan protocols with population-based input functions for accurate kinetic modeling of (18)F-FDG datasets from a long axial FOV PET scanner. Eur J Nucl Med Mol Imaging 2023;50:257-65. [Crossref] [PubMed]
13. Dias AH, Smith AM, Shah V, Pigg D, Gormsen LC, Munk OL. Clinical validation of a population-based input function for 20-min dynamic whole-body (18)F-FDG multiparametric PET imaging. EJNMMI Phys 2022;9:60. [Crossref] [PubMed]
14. Dias AH, Hansen AK, Munk OL, Gormsen LC. Normal values for (18)F-FDG uptake in organs and tissues measured by dynamic whole body multiparametric FDG PET in 126 patients. EJNMMI Res 2022;12:15. [Crossref] [PubMed]
15. Li Y, Hu J, Sari H, Xue S, Ma R, Kandarpa S, Visvikis D, Rominger A, Liu H, Shi K. A deep neural network for parametric image reconstruction on a large axial field-of-view PET. Eur J Nucl Med Mol Imaging 2023;50:701-14. [Crossref] [PubMed]
16. Zaker N, Haddad K, Faghihi R, Arabi H, Zaidi H. Direct inference of Patlak parametric images in whole-body PET/CT imaging using convolutional neural networks. Eur J Nucl Med Mol Imaging 2022;49:4048-63. [Crossref] [PubMed]
17. Huang Z, Wu Y, Fu F, Meng N, Gu F, Wu Q, Zhou Y, Yang Y, Liu X, Zheng H, Liang D, Wang M, Hu Z. Parametric image generation with the uEXPLORER total-body PET/CT system through deep learning. Eur J Nucl Med Mol Imaging 2022;49:2482-92. [Crossref] [PubMed]
18. Li S, Abdelhafez YG, Nardo L, Cherry SR, Badawi RD, Wang G. Total-Body Parametric Imaging Using Relative Patlak Plot. J Nucl Med 2025;66:654-61. [Crossref] [PubMed]
19. Xie H, Liu Q, Zhou B, Chen X, Guo X, Wang H, Li B, Rominger A, Shi K, Liu C. Unified Noise-aware Network for Low-count PET Denoising with Varying Count Levels. IEEE Trans Radiat Plasma Med Sci 2024;8:366-78. [Crossref] [PubMed]
20. Shiyam Sundar LK, Lassen ML, Gutschmayer S, Ferrara D, Calabrò A, Yu J, Kluge K, Wang Y, Nardo L, Hasbak P, Kjaer A, Abdelhafez YG, Wang G, Cherry SR, Spencer BA, Badawi RD, Beyer T, Muzik O. Fully Automated, Fast Motion Correction of Dynamic Whole-Body and Total-Body PET/CT Imaging Studies. J Nucl Med 2023;64:1145-53. [Crossref] [PubMed]
21. Jiao J, Heeman F, Dixon R, Wimberley C, Lopes Alves I, Gispert JD, Lammertsma AA, van Berckel BNM, da Costa-Luis C, Markiewicz P, Cash DM, Cardoso MJ, Ourselin S, Yaqub M, Barkhof F. NiftyPAD - Novel Python Package for Quantitative Analysis of Dynamic PET Data. Neuroinformatics 2023;21:457-68. [Crossref] [PubMed]