EJNMMI Physics最新文献

Background: In preclinical PET/MR, attenuation correction (AC) uses a mix of pre-calculated attenuation maps accounting for animal cradles and organ segmentation obtained from whole-body MR using a volume coil. The development of X-nuclei such as ²³Na or ³¹P may benefit from high sensitivity surface coils, which could lead to inaccurate PET quantification. In this study, we evaluated the benefit of a pre-calculated attenuation map including a dedicated cradle enclosing a surface coil for ¹⁸F-FDG PET imaging.

Materials and methods: We developed a 3D-printed cradle (DC) embedding a 20 mm surface coil made of PLA (polylactic acid), coated with a thin cap of epoxy. An attenuation map was generated using computed tomography and integrated into PET reconstruction. To validate AC, we compared PET images to those obtained using a conventional cradle and a volume coil (CC). Various image quality metrics were evaluated in various phantoms including a NEMA NU-4 (recovery coefficients (RC), uniformity (%STD), spillover ratio (SOR)), a homogenous phantom (slice and inter-slice uniformity) and a resolution phantom (spatial resolution). Finally, cardiac ¹⁸F-FDG PET images acquired with the 2 cradles were compared in Sprague-Dawley rats.

Results: RC obtained using DC and CC configurations were not significantly different. However, the %STD was significantly increased with the DC (5.07 ± 0.18) vs. CC (3.29 ± 0.53, p = 0.0002) leading to a decreased CNR with DC. The SOR was similar between the two cradles. In homogenous phantom, there was a non-significant 1.58% underestimation of the PET signal near the surface coil and 5% overestimation on the opposite side when using the DC vs. CC. Uniformity between slice was significantly higher in DC than in CC (3.15 ± 1.2% for DC vs. 2.24 ± 0.7% for CC, p = 0.02) while uniformity inter-slice was similar in DC and CC (2.25 ± 0.83% for DC vs. 1.93 ± 0.44% for CC, p = 0.2). Spatial resolution was similar between DC and CC (axial: 1.80 vs. 1.63 mm, tangential: 1.62 vs. 1.79 mm and radial resolution: 1.53 vs. 1.76 mm for DC and CC respectively). In-vivo, cardiac standardized uptake values were similar between the two cradles (3.55 ± 0.89 and 3.59 ± 1.02 for DC and CC respectively, p = 0.81).

Conclusion: Pre-calculated attenuation map using a standardized positioning of MR surface coil provided similar ¹⁸F-FDG PET images compared to a conventional PET/MR system with a volume coil, allowing its usage for combined PET and X-nuclei MR.

Purpose: Respiratory motion (RM)-related artifacts significantly impact image quality and diagnostic accuracy in PET/CT imaging. This study aimed to prospectively evaluate the clinical utility of the unified data-driven respiratory motion correction (uRMC) algorithm utilizing deep learning neural networks for diagnosing upper abdominal lesions.

Methods: A total of 100 patients with suspected upper abdominal lesions who underwent ¹⁸F-FDG PET/CT were enrolled. Two senior physicians independently conducted subjective visual assessments and semi-quantitative analyses of the PET/CT images before and after applying uRMC. Subjective visual evaluation parameters included overall image quality, PET-CT misalignment, and lesion distortion. Additionally, physicians identified involved upper-abdominal lesions in both images. Semi-quantitative metrics recorded for each lesion included maximum standardized uptake value (SUV_max), metabolic tumor volume (MTV), tumor-to-background ratio (TBR), and horizontal-to-vertical ratio (HV_ratio) before and after correction. Percentage changes in lesion SUV_max and MTV were calculated, and subgroup analyses were performed to assess the impact of lesion uptake, volume, and displacement on the performance of the uRMC algorithm.

Results: Compared to no motion correction (NMC) images, 78% (78/100) of patients demonstrated improved overall image quality after uRMC reconstruction, with 68.9% (155/225) of lesion showing improved PET-CT alignment and 64.0% (144/225) demonstrating reduced lesion distortion (all p < 0.001). The RM-corrected images exhibited a significantly higher SUV_max (9.07 [6.45, 11.79] vs.7.46 [5.69, 10.00], p < 0.001) and TBR (3.65 [2.54, 4.98] vs. 3.17 [2.40, 4.38], p < 0.001). The number of detected lesions increased from 171 (NMC) to 181 (uRMC) in 62 patients, with 10 additional suspicious lesions identified in 14.5% (9/62) of cases. Moreover, 7 lesions in 9.7% (6/62) of patients exhibited improved PET-CT alignment after uRMC correction. The uRMC algorithm also significantly reduced lesion MTV (1359.6 [690.8, 3837.6] mm³ vs.1710.5 [899.1, 4013.0] mm³, p < 0.01) and HV_ratio (0.99 [0.82, 1.09] vs. 1.16 [1.02, 1.44], p < 0.01). Subgroup-based analyses revealed that uRMC outperformed NMC in detecting low-uptake and small-volume lesions, with SUV_max improvements being more pronounced in lesions with larger displacement (17.8% vs. 9.8%, p < 0.001).

Conclusion: Compared with conventional NMC reconstruction, the uRMC algorithm significantly enhances overall image quality, PET-CT alignment, and lesion delineation. Furthermore, it improves the detection of low-uptake and small-volume lesions in the upper abdomen, thereby increasing the accuracy and reliability of clinical diagnoses and supporting more informed therapeutic decision-making.

Background: Improving the image quality of cardiac gating myocardial perfusion single-photon emission computed tomography (CG MP-SPECT) is crucial for accurate diagnosis. Diffusion model (DM) has recently shown promise in MP-SPECT image denoising, but traditional DM typically require extensive computational resources and prolonged processing times. The study aims to develop and evaluate a lightweight generalized DM for efficient denoising in CG MP-SPECT images.

Methods: We propose a novel step mapping generalized diffusion model (SMGDiff) that incorporates cardiac-gated MP-SPECT images as diffusion endpoints instead of traditional Gaussian noise, alongside a novel mean-preserving degradation operator to significantly reduce sampling steps and inference time. Additionally, a stepwise mapping and error optimization module (SMEO) was designed to precisely calibrate stepwise features using contextual information, thereby minimizing cumulative errors during reconstruction. A retrospective dataset of 50 MP-SPECT scans from 36 patients was used, each gated into 8 (CG-8) or 16 (CG-16) cardiac phases, generating 400/800 image pairs for CG-8/CG-16, respectively. The dataset was divided into training (35 scans), validation (5 scans), and testing (10 scans). Peak signal-to-noise ratio (PSNR), structural similarity (SSIM), normalized mean square error (NMSE), joint histogram, linear regression analysis and a paired two-tail t-test were employed for quantitative evaluation. Two board-certified nuclear medicine physicians performed a blinded and randomized reader study on resulted images. Images were rated on 5-point Likert scales for image quality and diagnostic confidence, with significance evaluated by Wilcoxon signed-rank tests.

Results: The SMGDiff model with 5 diffusion steps (SMGDiff-5) achieved the best overall performance across all evaluation metrics for both gating configurations. SMGDiff-5 also demonstrated superior computational efficiency, requiring only 0.024 s per slice compared to 4.982 s for the 1000-step diffusion model. Furthermore, SMGDiff-5 significantly outperformed established deep learning methods including CNN, U-Net, GAN, and the denoising diffusion probabilistic model, as evidenced by higher PSNR and SSIM and lower NMSE (p < 0.05). Joint histogram and linear regression analyses confirmed these quantitative findings. Reader study results aligned with the quantitative trends which SMGDiff-5 received the highest or near-reference ratings for image quality (CG-8 4.725; CG-16 4.550) and diagnostic confidence (CG-8 4.600; CG-16 4.525), clearly above original gated images and comparable to static MP-SPECT.

Conclusions: The proposed SMGDiff-5 model provides robust and efficient denoising of CG MP-SPECT images, offering superior performance compared to traditional deep learning methods with significantly reduced computational demand.

Aim: Predicting the time-activity curve (TAC) of [¹⁷⁷Lu]Lu-DOTA-TATE for organs at risk and neuroendocrine tumours (NETs) is an essential element in the calculation of the absorbed dose (AD) and a critical step for individualising peptide receptor radionuclide therapy (PRRT) treatment planning. This study aims to predict the TAC of [¹⁷⁷Lu]Lu-DOTA-TATE using a single quantitative image of [⁶⁸Ga]Ga-DOTA-TATE and population data with a physiologically based pharmacokinetic (PBPK) model.

Methods: A PBPK model was developed for [⁶⁸Ga]Ga-DOTA-TATE and [¹⁷⁷Lu]Lu-DOTA-TATE, including organs and NETs. To generate reference TACs, general physiological parameters were taken from the literature, while individual model parameters were estimated using pre-therapy (PET/CT) and post-therapy (planar and SPECT/CT) image-based organ activity measurements from patients with NETs. Different error models were evaluated to determine the best one. To predict the TAC of [¹⁷⁷Lu]Lu-DOTA-TATE from a single [⁶⁸Ga]Ga-DOTA-TATE PET/CT, individual model parameters were estimated using only [⁶⁸Ga]Ga-DOTA-TATE organ and tumour activity measurements. Finally, the predicted [¹⁷⁷Lu]Lu-DOTA-TATE TACs for modelled organs and NETs were compared to the reference.

Results: The best error model was the proportional data-based error model, where the proportionality parameter b differs between diagnostic and therapeutic data, and between tumours and organs: b_{T, Organ}, b_{T, Tumour}, and b_{D, Organ}, b_{D, Tumour}. The medians for b_{T, Organ}, b_{T, Tumour} and b_{D, Organ}, b_{D, Tumour} were determined to be 0.16, 0.39, 0.35, and 0.27, respectively. For the prediction, b_{D, Organ} and b_{D, Tumour} were used as patient-specific proportional errors. The relative prediction error (RPE) was calculated for the predicted time-integrated activity (TIA). The mean and standard deviation for the RPEs were found to be (- 5 ± 51)%, (- 4 ± 22)%, (- 13 ± 40)%, and (- 10 ± 21)% for tumours, kidneys, liver, and spleen, respectively. The mean absolute percentage errors (MAPEs) were 43%, 18%, 31% and 17% for tumours, kidney, liver, and spleen, respectively.

Conclusion: The integration of the PBPK model with a data-based proportional error model represents a significant improvement in predicting TACs for estimating tumour and organ ADs following [¹⁷⁷Lu]Lu-DOTA-TATE therapy, using single-time-point PET/CT imaging with [⁶⁸Ga]Ga-DOTA-TATE. These results emphasise the importance of error model analysis in PBPK modelling.

Purpose: This study aimed to assess the imaging performance of a 3D CZT SPECT/CT system for Tb-161 and compare its quantitative accuracy and image quality with those of a conventional [NaI(Tl)]-based SPECT/CT system.

Methods: A NEMA Image Quality (IQ) phantom was scanned with a 3D CZT SPECT/CT (Veriton, Spectrum Dynamics) for 8 min per bed position and with a conventional SPECT/CT (Symbia T16, by Siemens) for 16 min per bed position. The 3D CZT SPECT acquisition time was virtually shortened by reparsing the listmode data. The NEMA IQ phantom was scanned with a cold background and with a sphere-to-background ratio (SBR) of 20:1 and 10:1, with 0.71 MBq/mL in the spheres. Clinical reconstruction protocols were chosen by maximizing the SBR (for the four largest spheres) while minimizing the background noise.

Results: The optimal SBR-to-noise trade-off was achieved at 48 updates for the Symbia and 50 updates for the Veriton. The Veriton exhibited a 20% and 34% higher SBR with a 19% and 10% lower CoV compared to the Symbia for the 10:1 and 20:1 spheres to background ratios, respectively. The calibration factors were determined as 15.17 and 4.8 [Formula: see text] cps/MBq for the Symbia and Veriton, respectively. Reducing the scan time had minimal effect on SBR, but a twofold count reduction increased background noise by 11%.

Conclusion: Quantitative Tb-161 SPECT imaging can be performed using both a conventional and a 3D CZT SPECT/CT. Given its higher sensitivity, the 3D CZT system is preferred, as it achieved better image quality with an 8-min acquisition compared to a conventional system with a 16-min acquisition.

Background: Lead-212 (²¹²Pb) is being investigated for alpha therapies, but it can be challenging to image. To investigate the quantitative accuracy of ²¹²Pb SPECT images for patient geometries and low activity concentrations, we imaged an anthropomorphic phantom with ²¹²Pb, and studied the deviations of SPECT derived activity concentrations.

Methods: Fillable phantom compartment shells of the kidneys, liver and five vertebrae (T11-L3) were 3D-printed based on a patient's CT-images. The same patient's [¹⁸F]F-PSMA-1007 PET image was used as a basis for the relative distribution of ²¹²Pb activity within the phantom compartments. The phantom was imaged with a Siemens Symbia Intevo Bold SPECT/CT, with a total of 3.4-3.8 MBq ²¹²Pb for three acquisitions and 1.0-1.1 MBq ²¹²Pb for three acquisitions, while recording energy windows centred at 79 keV and 239 keV. The SPECT images were reconstructed with Siemens' Flash-3D with a variety of iterations, subsets, and matrix sizes. Activity concentrations for each phantom compartment were measured from the images using a calibration factor measured in a uniform cross calibration phantom and compared to the activity concentrations in 1 ml samples extracted from each compartment, which were analysed using a gamma counter.

Results: Quantification was relatively stable across energy windows and matrix sizes, but best results were achieved using 30-120 reconstruction updates. Low activity concentration volumes representing background and vertebral bodies (0.03-0.05 kBq/ml) were not quantifiable with deviations over 400% for all investigated reconstructions. The activity concentrations in the liver and kidneys were underestimated by 10-50% compared to the gamma counter measurements. Precision between SPECT acquisitions was higher for the larger image matrix, with standard deviations of liver and kidney measurements less than 6% for the higher activity images, and less than 8% for the lower activity images.

Conclusion: We found that larger volumes, such as liver and kidneys with at least 210 Bq/ml, may be quantifiable with an accuracy of approx. 30-40%. While very low activity concentrations below 54 Bq/ml were not quantifiable, this still indicate carefully used imaging results to be of value in dosimetric calculations, also when characterising latter parts of time activity curves.

Purpose: In this study, we constructed an anomaly detection system based on the Wasserstein generative model in brain single-photon emission computed tomography (SPECT) images, and conducted a basic study on the feasibility of the system.

Methods: The proposed method used a Wasserstein generative adversarial network approach based on optimal transport theory. Anomaly detection was performed using only healthy images, and based on the anomaly score calculated from the loss function. Theoretical validation was performed using a numerical phantom which simulated medical image with varying image noise and signal levels. The results were evaluated with receiver operating characteristic curves and area under the curve (AUC). Brain SPECT images from clinical scenarios were used to investigate the feasibility of this system through subtraction images and various quantitative evaluations.

Results: The numerical phantom showed the relationship between the noise and signal values of the system, and the anomaly detection ability based on the anomaly score value indicated an AUC of 0.9994. The correlation between the anomaly score and the anomaly image was confirmed in the brain SPECT image, and the detection region was clearly observed in the difference image.

Conclusion: We proposed and investigated the feasibility of an anomaly detection system based on the Wasserstein generative model. Its effectiveness is expected to reduce the workload of human operators.

Objective: This study explored a hybrid curve-fitting method optimized for radiation dosimetry evaluation in total-body dynamic positron emission tomography/computed tomography (PET/CT) imaging, compared to conventional methods reliant on multi-time-point static acquisition protocols, and evaluated the radiation dosimetry results and organ biodistribution of [⁶⁸Ga]Ga-NOTA-SNA002.

Methods: A total of 16 patients with solid tumors underwent a 60-min dynamic PET acquisition immediately after administration of [⁶⁸Ga]Ga-NOTA-SNA002, followed by a 20-min static PET scan at about 120 min post-injection. Volumes of interest (VOIs) were manually delineated on the CT images for subsequent radiation dosimetry estimation and biodistribution analysis. Two datasets were created to generate organ time-activity curves (TACs): (1) two PET frames reconstructed at 15-35 min and 40-60 min from the dynamic PET dataset, and one frame from static PET acquisition, thereby simulating a conventional multi-time-point static protocol; (2) 55 frames from the dynamic PET dataset and a frame from static PET. To calculate the time-integrated activity coefficients (TIACs) for the tracer in source organs, two methods were used: a routine method (RM) that fitted Dataset 1 by using a conventional bi-exponential curve fitting approach, and a hybrid method (HM) that fitted Dataset 2 by combining rectangular integration for the early phase of TACs with exponential fitting for the later phase. Absorbed and effective radiation doses were subsequently estimated using OLINDA/EXM version 1.1.

Results: The [⁶⁸Ga]Ga-NOTA-SNA002 primarily accumulated in the urinary system, with relatively low overall uptake levels in other source organs. Analysis of tumor-to-organ standardized uptake value (SUV) ratios revealed that all ratios reached their highest values at 120 min post-injection. The TIACs derived from the HM were consistently higher than those obtained using the RM method, while preserving a similar rank order of distribution across organs. In both methods, the kidneys, liver, and lungs exhibited the largest TIACs. The bladder wall, kidneys, and spleen received the highest absorbed doses. The total effective dosimetry calculated with RM and HM were 20.47 ± 3.08 µSv/MBq and 36.33 ± 6.18 µSv/MBq, respectively.

Conclusion: The radiation dosimetry results obtained from both the RM and the HM consistently support the safety and feasibility of total-body PET/CT imaging using [⁶⁸Ga]Ga-NOTA-SNA002. Meanwhile, the HM provides theoretically more accurate results compared to the RM, and is therefore recommended for radiation dosimetry evaluation of other novel tracers.

Purpose: Actinium-225 (²²⁵Ac)-labeled prostate-specific membrane antigen (PSMA) radiopharmaceuticals represent a promising therapeutic approach for metastatic castration-resistant prostate cancer (mCRPC), yet clinical implementation remains limited by the absence of accurate dosimetric assessment methods. The complex decay chain and non-imaging alpha emissions of ²²⁵Ac pose substantial challenges for quantitative imaging. We aimed to evaluate the feasibility of quantitative single photon emission computed tomography (SPECT)-based dosimetry for ²²⁵Ac-PSMA-CY313 therapy by exploiting gamma emissions from daughter radionuclides francium-221 (²²¹Fr) and bismuth-213 (²¹³Bi).

Methods: Four mCRPC patients received 185.8 ± 11.7 µCi ²²⁵Ac-PSMA-CY313 and underwent multi-timepoint SPECT/CT and whole-body planar imaging at 6, 24, 48, and 96 h post-injection. Quantitative SPECT reconstruction used ordered-subsets expectation-maximization with comprehensive corrections for attenuation, scatter, resolution blur, and crosstalk. Volume of interest were defined using co-registered ¹⁸F-PSMA-CY313 positron emission tomography /computed tomography (PET/CT). Time-activity curves were fitted with mono- or bi-exponential models, and absorbed doses were calculated using validated Monte Carlo-based software and International Commission on Radiological Protection reference phantoms.

Results: High-quality quantitative imaging was successfully achieved across all timepoints. Among normal organs, kidneys and liver exhibited the highest absorbed doses (1.55 ± 0.38 Gy and 1.07 ± 0.19 Gy, respectively), corresponding to dose coefficients of 0.23 ± 0.07 Gy/MBq and 0.16 ± 0.03 Gy/MBq. Soft-tissue lesions exhibited higher absorbed doses than bone metastases (5.03 ± 5.51 Gy versus 1.61 ± 2.28 Gy), with corresponding dose coefficients of 0.73 ± 0.80 Gy/MBq and 0.25 ± 0.33 Gy/MBq. Tumor-to-critical organ dose ratios indicated favorable therapeutic windows, with red marrow showing the highest ratio (14.84), followed by adrenal glands (6.35) and salivary glands (4.96), while the dose-limiting kidneys demonstrated a ratio of 1.55.

Conclusion: Quantitative SPECT-based dosimetry for ²²⁵Ac-PSMA-CY313 therapy is clinically feasible using standard imaging systems. This methodology demonstrates preferential tumor targeting with acceptable organ-at-risk dose distributions, supporting the therapeutic potential of ²²⁵Ac-PSMA-CY313 for mCRPC and providing a practical framework for personalized dosimetry in targeted alpha therapy.