The predictive value of pretherapy [68Ga]Ga-DOTA-TATE PET and biomarkers in [177Lu]Lu-PRRT tumor dosimetry.

PURPOSE: Metastatic neuroendocrine tumors (NETs) overexpressing type 2 somatostatin receptors are the target for peptide receptor radionuclide therapy (PRRT) through the theragnostic pair of 68Ga/177Lu-DOTATATE. The main purpose of this study was to develop machine learning models to predict therapeutic tumor dose using pre therapy 68Ga -PET and clinicopathological biomarkers. METHODS: We retrospectively analyzed 90 segmented metastatic NETs from 25 patients (M14/F11, age 63.7 ± 9.5, range 38-76) treated by 177Lu-DOTATATE at our institute. Patients underwent both pretherapy [68Ga]Ga-DOTA-TATE PET/CT and four timepoints SPECT/CT at ~ 4, 24, 96, and 168 h post-177Lu-DOTATATE infusion. Tumors were segmented by a radiologist on baseline CT or MRI and transferred to co-registered PET/CT and SPECT/CT, and normal organs were segmented by deep learning-based method on CT of the PET and SPECT. The SUV metrics and tumor-to-normal tissue SUV ratios (SUV_TNRs) were calculated from 68Ga -PET at the contour-level. Posttherapy dosimetry was performed based on the co-registration of SPECT/CTs to generate time-integrated-activity, followed by an in-house Monte Carlo-based absorbed dose estimation. The correlation between delivered 177Lu Tumor absorbed dose and PET-derived metrics along with baseline clinicopathological biomarkers (such as Creatinine, Chromogranin A and prior therapies) were evaluated. Multiple interpretable machine-learning algorithms were developed to predict tumor dose using these pretherapy information. Model performance on a nested tenfold cross-validation was evaluated in terms of coefficient of determination (R2), mean-absolute-error (MAE), and mean-relative-absolute-error (MRAE). RESULTS: SUVmean showed a significant correlation (q-value < 0.05) with absorbed dose (Spearman ρ = 0.64), followed by TLSUVmean (SUVmean of total-lesion-burden) and SUVpeak (ρ = 0.45 and 0.41, respectively). The predictive value of PET-SUVmean in estimation of posttherapy absorbed dose was stronger compared to PET-SUVpeak, and SUV_TNRs in terms of univariate analysis (R2 = 0.28 vs. R2 ≤ 0.12). An optimal trivariate random forest model composed of SUVmean, TLSUVmean, and total liver SUVmean (normal and tumoral liver) provided the best performance in tumor dose prediction with R2 = 0.64, MAE = 0.73 Gy/GBq, and MRAE = 0.2. CONCLUSION: Our preliminary results demonstrate the feasibility of using baseline PET images for prediction of absorbed dose prior to 177Lu-PRRT. Machine learning models combining multiple PET-based metrics performed better than using a single SUV value and using other investigated clinicopathological biomarkers. Developing such quantitative models forms the groundwork for the role of 68Ga -PET not only for the implementation of personalized treatment planning but also for patient stratification in the era of precision medicine.

A deep neural network for fast and accurate scatter estimation in quantitative SPECT/CT under challenging scatter conditions.

PURPOSE: A major challenge for accurate quantitative SPECT imaging of some radionuclides is the inadequacy of simple energy window-based scatter estimation methods, widely available on clinic systems. A deep learning approach for SPECT/CT scatter estimation is investigated as an alternative to computationally expensive Monte Carlo (MC) methods for challenging SPECT radionuclides, such as 90Y. METHODS: A deep convolutional neural network (DCNN) was trained to separately estimate each scatter projection from the measured 90Y bremsstrahlung SPECT emission projection and CT attenuation projection that form the network inputs. The 13-layer deep architecture consisted of separate paths for the emission and attenuation projection that are concatenated before the final convolution steps. The training label consisted of MC-generated "true" scatter projections in phantoms (MC is needed only for training) with the mean square difference relative to the model output serving as the loss function. The test data set included a simulated sphere phantom with a lung insert, measurements of a liver phantom, and patients after 90Y radioembolization. OS-EM SPECT reconstruction without scatter correction (NO-SC), with the true scatter (TRUE-SC) (available for simulated data only), with the DCNN estimated scatter (DCNN-SC), and with a previously developed MC scatter model (MC-SC) were compared, including with 90Y PET when available. RESULTS: The contrast recovery (CR) vs. noise and lung insert residual error vs. noise curves for images reconstructed with DCNN-SC and MC-SC estimates were similar. At the same noise level of 10% (across multiple realizations), the average sphere CR was 24%, 52%, 55%, and 67% for NO-SC, MC-SC, DCNN-SC, and TRUE-SC, respectively. For the liver phantom, the average CR for liver inserts were 32%, 73%, and 65% for NO-SC, MC-SC, and DCNN-SC, respectively while the corresponding values for average contrast-to-noise ratio (visibility index) in low-concentration extra-hepatic inserts were 2, 19, and 61, respectively. In patients, there was high concordance between lesion-to-liver uptake ratios for SPECT reconstruction with DCNN-SC (median 4.8, range 0.02-13.8) compared with MC-SC (median 4.0, range 0.13-12.1; CCC = 0.98) and with 90Y PET (median 4.9, range 0.02-11.2; CCC = 0.96) while the concordance with NO-SC was poor (median 2.8, range 0.3-7.2; CCC = 0.59). The trained DCNN took ~ 40 s (using a single i5 processor on a desktop computer) to generate the scatter estimates for all 128 views in a patient scan, compared to ~ 80 min for the MC scatter model using 12 processors. CONCLUSIONS: For diverse 90Y test data that included patient studies, we demonstrated comparable performance between images reconstructed with deep learning and MC-based scatter estimates using metrics relevant for dosimetry and for safety. This approach that can be generalized to other radionuclides by changing the training data is well suited for real-time clinical use because of the high speed, orders of magnitude faster than MC, while maintaining high accuracy.

A review of 177Lu dosimetry workflows: how to reduce the imaging workloads?

177 Lu radiopharmaceutical therapy is a standardized systemic treatment, with a typical dose of 7.4 GBq per injection, but its response varies from patient to patient. Dosimetry provides the opportunity to personalize treatment, but it requires multiple post-injection images to monitor the radiopharmaceutical's biodistribution over time. This imposes an additional imaging burden on centers with limited resources. This review explores methods to lessen this burden by optimizing acquisition types and minimizing the number and duration of imaging sessions. After summarizing the different steps of dosimetry and providing examples of dosimetric workflows for 177 Lu -DOTATATE and 177 Lu -PSMA, we examine dosimetric workflows based on a reduced number of acquisitions, or even just one. We provide a non-exhaustive description of simplified methods and their assumptions, as well as their limitations. Next, we detail the specificities of each normal tissue and tumors, before reviewing dose-response relationships in the literature. In conclusion, we will discuss the current limitations of dosimetric workflows and propose avenues for improvement.

An International Study of Factors Affecting Variability of Dosimetry Calculations, Part 3: Contribution from Calculating Absorbed Dose from Time-Integrated Activity.

Image-based dosimetry-guided radiopharmaceutical therapy has the potential to personalize treatment by limiting toxicity to organs at risk and maximizing the therapeutic effect. The 177Lu dosimetry challenge of the Society of Nuclear Medicine and Molecular Imaging consisted of 5 tasks assessing the variability in the dosimetry workflow. The fifth task investigated the variability associated with the last step, dose conversion, of the dosimetry workflow on which this study is based. Methods: Reference variability was assessed by 2 medical physicists using different software, methods, and all possible combinations of input segmentation formats and time points as provided in the challenge. General descriptive statistics for absorbed dose values from the global submissions from participants were calculated, and variability was measured using the quartile coefficient of dispersion. Results: For the liver, which included lesions with high uptake, variabilities of up to 36% were found. The baseline analysis showed a variability of 29% in absorbed dose results for the liver from datasets where lesions included and excluded were grouped, indicating that variation in how lesions in normal liver were treated was a significant source of variability. For other organs and lesions, variability was within 7%, independently of software used except for the local deposition method. Conclusion: The choice of dosimetry method or software had a small contribution to the overall variability of dose estimates.

Linear Boltzmann equation solver for voxel-level dosimetry in radiopharmaceutical therapy: Comparison with Monte Carlo and kernel convolution.

BACKGROUND: With recent interest in patient-specific dosimetry for radiopharmaceutical therapy (RPT) and selective internal radiation therapy (SIRT), an increasing number of voxel-based algorithms are being evaluated. Monte Carlo (MC) radiation transport, generally considered to be the most accurate among different methods for voxel-level absorbed dose estimation, can be computationally inefficient for routine clinical use. PURPOSE: This work demonstrates a recently implemented grid-based linear Boltzmann transport equation (LBTE) solver for fast and accurate voxel-based dosimetry in RPT and SIRT and benchmarks it against MC. METHODS: A deterministic LBTE solver (Acuros MRT) was implemented within a commercial RPT dosimetry package (Velocity 4.1). The LBTE is directly discretized using an adaptive mesh refined grid and then the coupled photon-electron radiation transport is iteratively solved inside specified volumes to estimate radiation doses from both photons and charged particles in heterogeneous media. To evaluate the performance of the LBTE solver for RPT and SIRT applications, 177 Lu SPECT/CT, 90 Y PET/CT, and 131 I SPECT/CT images of phantoms and patients were used. Multiple lesions (2-1052 mL) and normal organs were delineated for each study. Voxel dosimetry was performed with the LBTE solver, dose voxel kernel (DVK) convolution with density correction, and a validated in-house MC code using the same time-integrated activity and density maps as input to the different dose engines. The resulting dose maps, difference maps, and dose-volume-histogram (DVH) metrics were compared, to assess the voxel-level agreement. Evaluation of mean absorbed dose included comparison with structure-level estimates from OLINDA. RESULTS: In the phantom inserts/compartments, the LBTE solver versus MC and DVK convolution demonstrated good agreement with mean absorbed dose and DVH metrics agreeing to within 5% except for the D90 and D70 metrics of a very low activity concentration insert of 90 Y where the agreement was within 15%. In the patient studies (five patients imaged after 177 Lu DOTATATE RPT, five after 90 Y SIRT, and two after 131 I radioimmunotherapy), in general, there was better agreement between the LBTE solver and MC than between LBTE solver and DVK convolution for mean absorbed dose and voxel-level evaluations. Across all patients for all three radionuclides, for soft tissue structures (kidney, liver, lesions), the mean absorbed dose estimates from the LBTE solver were in good agreement with those from MC (median difference < 1%, maximum 9%) and those from DVK (median difference < 5%, maximum 9%). The LBTE and OLINDA estimates for mean absorbed dose in kidneys and liver agreed to within 10%, but differences for lesions were larger with a maximum 14% for 177 Lu, 23% for 90 Y, and 26% for 131 I. For bone regions, the agreement in mean absorbed doses between LBTE and both MC and DVK were similar (median < 11%, max 11%) while for lung the agreement between LBTE and MC (median < 1%, max 8%) was substantially better than between LBTE and DVK (median < 16%, max 33%). Voxel level estimates for soft tissue structures also showed good agreement between the LBTE solver and both MC and DVK with a median difference < 5% (maximum < 13%) for the DVH metrics with all three radionuclides. The largest difference in DVH metrics was for the D90 and D70 metric in lung and bone where the uptake was low. Here, the difference between LBTE and MC had a median value < 14% (maximum 23%) for bone and < 4% (maximum 37%) for lung, while the corresponding differences between LBTE and DVK were < 23% (maximum 31%) and < 67% (maximum 313%), respectively. For a typical patient with a matrix size of 166 × 166 × 129 (voxel size 3 × 3 × 3 mm3 ), voxel dosimetry using the LBTE solver was as fast as ∼2 min on a desktop computer. CONCLUSION: Having established good agreement between the LBTE solver and MC for RPT and SIRT applications, the LBTE solver is a viable option for voxel dosimetry that can be faster than MC. Further analysis is being performed to encompass the broad range of radionuclides and conditions encountered clinically.

Enhancing precision: A predictive model for 177Lu-DOTATATE treatment response in neuroendocrine tumors using quantitative 68Ga-DOTATATE PET and clinicopathological biomarkers.

Purpose: This study aims to elucidate the role of quantitative SSTR-PET metrics and clinicopathological biomarkers in the progression-free survival (PFS) and overall survival (OS) of neuroendocrine tumors (NETs) treated with peptide receptor radionuclide therapy (PRRT). Methods: A retrospective analysis including 91 NET patients (M47/F44; age 66 years, range 34-90 years) who completed four cycles of standard 177Lu-DOTATATE was conducted. SSTR-avid tumors were segmented from pretherapy SSTR-PET images using a semiautomatic workflow with the tumors labeled based on the anatomical regions. Multiple image-based features including total and organ-specific tumor volume and SSTR density along with clinicopathological biomarkers including Ki-67, chromogranin A (CgA) and alkaline phosphatase (ALP) were analyzed with respect to the PRRT response. Results: The median OS was 39.4 months (95% CI: 33.1-NA months), while the median PFS was 23.9 months (95% CI: 19.3-32.4 months). Total SSTR-avid tumor volume (HR = 3.6; P = 0.07) and bone tumor volume (HR = 1.5; P = 0.003) were associated with shorter OS. Also, total tumor volume (HR = 4.3; P = 0.01), liver tumor volume (HR = 1.8; P = 0.05) and bone tumor volume (HR = 1.4; P = 0.01) were associated with shorter PFS. Furthermore, the presence of large lesion volume with low SSTR uptake was correlated with worse OS (HR = 1.4; P = 0.03) and PFS (HR = 1.5; P = 0.003). Among the biomarkers, elevated baseline CgA and ALP showed a negative association with both OS (CgA: HR = 4.9; P = 0.003, ALP: HR = 52.6; P = 0.004) and PFS (CgA: HR = 4.2; P = 0.002, ALP: HR = 9.4; P = 0.06). Similarly, number of prior systemic treatments was associated with shorter OS (HR = 1.4; P = 0.003) and PFS (HR = 1.2; P = 0.05). Additionally, tumors originating from the midgut primary site demonstrated longer PFS, compared to the pancreas (HR = 1.6; P = 0.16), and those categorized as unknown primary (HR = 3.0; P = 0.002). Conclusion: Image-based features such as SSTR-avid tumor volume, bone tumor involvement, and the presence of large tumors with low SSTR expression demonstrated significant predictive value for PFS, suggesting potential clinical utility in NETs management. Moreover, elevated CgA and ALP, along with an increased number of prior systemic treatments, emerged as significant factors associated with worse PRRT outcomes.

SPECT/CT Image-Derived Absorbed Dose to Red Marrow Correlates with Hematologic Toxicity in Patients Treated with [177Lu]Lu-DOTATATE.

Hematologic toxicity, although often transient, is the most common limiting adverse effect during somatostatin peptide receptor radionuclide therapy. This study investigated the association between Monte Carlo-derived absorbed dose to the red marrow (RM) and hematologic toxicity in patients being treated for their neuroendocrine tumors. Methods: Twenty patients each receiving 4 treatment cycles of [177Lu]Lu-DOTATATE were included. Multiple-time-point 177Lu SPECT/CT imaging-based RM dosimetry was performed using an artificial intelligence-driven workflow to segment vertebral spongiosa within the field of view (FOV). This workflow was coupled with an in-house macroscale/microscale Monte Carlo code that incorporates a spongiosa microstructure model. Absorbed dose estimates to RM in lumbar and thoracic vertebrae within the FOV, considered as representations of the whole-body RM absorbed dose, were correlated with hematologic toxicity markers at about 8 wk after each cycle and at 3- and 6-mo follow-up after completion of all cycles. Results: The median of absorbed dose to RM in lumbar and thoracic vertebrae within the FOV (D median,vertebrae) ranged from 0.019 to 0.11 Gy/GBq. The median of cumulative absorbed dose across all 4 cycles was 1.3 Gy (range, 0.6-2.5 Gy). Hematologic toxicity was generally mild, with no grade 2 or higher toxicity for platelets, neutrophils, or hemoglobin. However, there was a decline in blood counts over time, with a fractional value relative to baseline at 6 mo of 74%, 97%, 57%, and 97%, for platelets, neutrophils, lymphocytes, and hemoglobin, respectively. Statistically significant correlations were found between a subset of hematologic toxicity markers and RM absorbed doses, both during treatment and at 3- and 6-mo follow-up. This included a correlation between the platelet count relative to baseline at 6-mo follow up: D median,vertebrae (r = -0.64, P = 0.015), D median,lumbar (r = -0.72, P = 0.0038), D median,thoracic (r = -0.58, P = 0.029), and D average,vertebrae (r = -0.66, P = 0.010), where D median,lumbar and D median,thoracic are median absorbed dose to the RM in the lumbar and thoracic vertebrae, respectively, within the FOV and D average,vertebrae is the mass-weighted average absorbed dose of all vertebrae. Conclusion: This study found a significant correlation between image-derived absorbed dose to the RM and hematologic toxicity, including a relative reduction of platelets at 6-mo follow up. These findings indicate that absorbed dose to the RM can potentially be used to understand and manage hematologic toxicity in peptide receptor radionuclide therapy.

Technical note: Impact of dose voxel kernel (DVK) values on dosimetry estimates in 177 Lu and 90 Y radiopharmaceutical therapy (RPT) applications.

BACKGROUND: Radiopharmaceutical therapy (RPT) is an increasingly adopted modality for treating cancer. There is evidence that the optimization of the treatment based on dosimetry can improve outcomes. However, standardization of the clinical dosimetry workflow still represents a major effort. Among the many sources of variability, the impact of using different Dose Voxel Kernels (DVKs) to generate absorbed dose (AD) maps by convolution with the time-integrated activity (TIA) distribution has not been systematically investigated. PURPOSE: This study aims to compare DVKs and assess the differences in the ADs when convolving the same TIA map with different DVKs. METHODS: DVKs of 3 × 3 × 3 mm3 sampling-nine for 177 Lu, nine for 90 Y-were selected from those most used in commercial/free software or presented in prior publications. For each voxel within a 11 × 11 × 11 matrix, the coefficient of variation (CoV) and the percentage difference between maximum and minimum values (% maximum difference) were calculated. The total absorbed dose per decay (SUM), calculated as the sum of all the voxel values in each kernel, was also compared. Publicly available quantitative SPECT images for two patients treated with 177 Lu-DOTATATE and PET images for two patients treated with 90 Y-microspheres were used, including organs at risk (177 Lu: kidneys; 90 Y: liver and healthy liver) and tumors' segmentations. For each patient, the mean AD to the volumes of interest (VOIs) was calculated using the different DVKs, the same TIA map and the same software tool for dose convolution, thereby focusing on the DVK impact. For each VOI, the % maximum difference of the mean AD between maximum and minimum values was computed. RESULTS: The CoV (% maximum difference) in voxels of normalized coordinates [0,0,0], [0,1,0], and [0,1,1] were 5%(21%), 9%(35%), and 10%(46%) for the 177 Lu DVKs. For the case of 90 Y, these values were 2%(9%), 4%(14%), and 4%(16%). The CoV (% maximum difference) for SUM was 9%(33%) for 177 Lu, and 4%(15%) for 90 Y. The variability of the mean tumor and organ AD was up to 19% and 15% in 177 Lu-DOTATATE and 90 Y-microspheres patients, respectively. CONCLUSIONS: This study showed a considerable AD variability due exclusively to the use of different DVKs. A concerted effort by the scientific community would contribute to decrease these discrepancies, strengthening the consistency of AD calculation in RPT.

PET/CT-Based Absorbed Dose Maps in 90Y Selective Internal Radiation Therapy Correlate with Spatial Changes in Liver Function Derived from Dynamic MRI.

Functional liver parenchyma can be damaged from treatment of liver malignancies with 90Y selective internal radiation therapy (SIRT). Evaluating functional parenchymal changes and developing an absorbed dose (AD)-toxicity model can assist the clinical management of patients receiving SIRT. We aimed to determine whether there is a correlation between 90Y PET AD voxel maps and spatial changes in the nontumoral liver (NTL) function derived from dynamic gadoxetic acid-enhanced MRI before and after SIRT. Methods: Dynamic gadoxetic acid-enhanced MRI scans were acquired before and after treatment for 11 patients undergoing 90Y SIRT. Gadoxetic acid uptake rate (k1) maps that directly quantify spatial liver parenchymal function were generated from MRI data. Voxel-based AD maps, derived from the 90Y PET/CT scans, were binned according to AD. Pre- and post-SIRT k1 maps were coregistered to the AD map. Absolute and percentage k1 loss in each bin was calculated as a measure of loss of liver function, and Spearman correlation coefficients between k1 loss and AD were evaluated for each patient. Average k1 loss over the patients was fit to a 3-parameter logistic function based on AD. Patients were further stratified into subgroups based on lesion type, baseline albumin-bilirubin scores and alanine transaminase levels, dose-volume effect, and number of SIRT treatments. Results: Significant positive correlations (ρ = 0.53-0.99, P < 0.001) between both absolute and percentage k1 loss and AD were observed in most patients (8/11). The average k1 loss over 9 patients also exhibited a significant strong correlation with AD (ρ ≥ 0.92, P < 0.001). The average percentage k1 loss of patients across AD bins was 28%, with a logistic function model demonstrating about a 25% k1 loss at about 100 Gy. Analysis between patient subgroups demonstrated that k1 loss was greater among patients with hepatocellular carcinoma, higher alanine transaminase levels, larger fractional volumes of NTL receiving an AD of 70 Gy or more, and sequential SIRT treatments. Conclusion: Novel application of multimodality imaging demonstrated a correlation between 90Y SIRT AD and spatial functional liver parenchymal degradation, indicating that a higher AD is associated with a larger loss of local hepatocyte function. With the developed response models, PET-derived AD maps can potentially be used prospectively to identify localized damage in liver and to enhance treatment strategies.

Accuracy and uncertainty analysis of reduced time point imaging effect on time-integrated activity for 177Lu-DOTATATE PRRT in patients and clinically realistic simulations.

BACKGROUND: Dosimetry promises many advantages for radiopharmaceutical therapies but repeat post-therapy imaging for dosimetry can burden both patients and clinics. Recent applications of reduced time point imaging for time-integrated activity (TIA) determination for internal dosimetry following 177Lu-DOTATATE peptide receptor radionuclide therapy have shown promising results that allow for the simplification of patient-specific dosimetry. However, factors such as scheduling can lead to sub-optimal imaging time points, but the resulting impact on dosimetry accuracy is still under investigation. We use four-time point 177Lu SPECT/CT data for a cohort of patients treated at our clinic to perform a comprehensive analysis of the error and variability in time-integrated activity when reduced time point methods with various combinations of sampling points are employed. METHODS: The study includes 28 patients with gastroenteropancreatic neuroendocrine tumors who underwent post-therapy SPECT/CT imaging at approximately 4, 24, 96, and 168 h post-therapy (p.t.) following the first cycle of 177Lu-DOTATATE. The healthy liver, left/right kidney, spleen and up to 5 index tumors were delineated for each patient. Time-activity curves were fit with either monoexponential or biexponential functions for each structure, based on the Akaike information criterion. This fitting was performed using all 4 time points as a reference and various combinations of 2 and 3 time points to determine optimal imaging schedules and associated errors. 2 commonly used methods of single time point (STP) TIA estimation are also evaluated. A simulation study was also performed with data generated by sampling curve fit parameters from log-normal distributions derived from the clinical data and adding realistic measurement noise to sampled activities. For both clinical and simulation studies, error and variability in TIA estimates were estimated with various sampling schedules. RESULTS: The optimal post-therapy imaging time period for STP estimates of TIA was found to be 3-5 days (71-126 h) p.t. for tumor and organs, with one exception of 6-8 days (144-194 h) p.t. for spleen with one STP approach. At the optimal time point, STP estimates give mean percent errors (MPE) within ± 5% and SD < 9% across all structures with largest magnitude error for kidney TIA (MPE = - 4.1%) and highest variability also for kidney TIA (SD = 8.4%). The optimal sampling schedule for 2TP estimates of TIA is 1-2 days (21-52 h) p.t. followed by 3-5 days (71-126 h) p.t. for kidney, tumor, and spleen. Using the optimal sampling schedule, the largest magnitude MPE for 2TP estimates is 1.2% for spleen and highest variability is in tumor with SD = 5.8%. The optimal sampling schedule for 3TP estimates of TIA is 1-2 days (21-52 h) p.t. followed by 3-5 days (71-126 h) p.t. and 6-8 days (144-194 h) p.t. for all structures. Using the optimal sampling schedule, the largest magnitude MPE for 3TP estimates is 2.5% for spleen and highest variability is in tumor with SD = 2.1%. Simulated patient results corroborate these findings with similar optimal sampling schedules and errors. Many sub-optimal reduced time point sampling schedules also exhibit low error and variability. CONCLUSIONS: We show that reduced time point methods can be used to achieve acceptable average TIA errors over a wide range of imaging time points and sampling schedules while maintaining low uncertainty. This information can improve the feasibility of dosimetry for 177Lu-DOTATATE and elucidate the uncertainty associated with non-ideal conditions.

Training End-to-End Unrolled Iterative Neural Networks for SPECT Image Reconstruction.

Training end-to-end unrolled iterative neural networks for SPECT image reconstruction requires a memory-efficient forward-backward projector for efficient backpropagation. This paper describes an open-source, high performance Julia implementation of a SPECT forward-backward projector that supports memory-efficient backpropagation with an exact adjoint. Our Julia projector uses only ~5% of the memory of an existing Matlab-based projector. We compare unrolling a CNN-regularized expectation-maximization (EM) algorithm with end-to-end training using our Julia projector with other training methods such as gradient truncation (ignoring gradients involving the projector) and sequential training, using XCAT phantoms and virtual patient (VP) phantoms generated from SIMIND Monte Carlo (MC) simulations. Simulation results with two different radionuclides (90Y and 177Lu) show that: 1) For 177Lu XCAT phantoms and 90Y VP phantoms, training unrolled EM algorithm in end-to-end fashion with our Julia projector yields the best reconstruction quality compared to other training methods and OSEM, both qualitatively and quantitatively. For VP phantoms with 177Lu radionuclide, the reconstructed images using end-to-end training are in higher quality than using sequential training and OSEM, but are comparable with using gradient truncation. We also find there exists a trade-off between computational cost and reconstruction accuracy for different training methods. End-to-end training has the highest accuracy because the correct gradient is used in backpropagation; sequential training yields worse reconstruction accuracy, but is significantly faster and uses much less memory.

Accuracy and uncertainty analysis of reduced time point imaging effect on time-integrated activity for 177Lu-DOTATATE PRRT in clinical patients and realistic simulations.

Background. Dosimetry promises many advantages for radiopharmaceutical therapies but repeat post-therapy imaging for dosimetry can burden both patients and clinics. Recent applications of reduced time point imaging for time-integrated activity (TIA) determination for internal dosimetry following 177 Lu-DOTATATE peptide receptor radionuclide therapy have shown promising results that allow for the simplification of patient-specific dosimetry. However, factors such as scheduling can lead to undesirable imaging time points, but the resulting impact on dosimetry accuracy is unknown. We use four-time point 177 Lu SPECT/CT data for a cohort of patients treated at our clinic to perform a comprehensive analysis of the error and variability in time-integrated activity when reduced time point methods with various combination of sampling points are employed. Methods. The study includes 28 patients with gastroenteropancreatic neuroendocrine tumors who underwent post-therapy SPECT/CT imaging at approximately 4, 24, 96, and 168 hours post-therapy (p.t.) following the first cycle of 177 Lu-DOTATATE. The healthy liver, left/right kidney, spleen and up to 5 index tumors were delineated for each patient. Time-activity curves were fit with either monoexponential or biexponential functions for each structure, based on the Akaike information criterion. This fitting was performed using all 4 time points as a reference and various combinations of 2 and 3 time points to determine optimal imaging schedules and associated errors. 2 commonly used methods of single time point (STP) TIA estimation are also evaluated. A simulation study was also performed with data generated by sampling curve fit parameters from log-normal distributions derived from the clinical data and adding realistic measurement noise to sampled activities. For both clinical and simulation studies, error and variability in TIA estimates were estimated with various sampling schedules. Results . The optimal post-therapy imaging time period for STP estimates of TIA was found to be 3-5 days (71-126 h) p.t. for tumor and organs, with one exception of 6-8 days (144-194 h) p.t. for spleen with one STP approach. At the optimal time point, STP estimates give mean percent errors (MPE) within +/-5% and SD < 9% across all structures with largest magnitude error for kidney TIA (MPE=-4.1%) and highest variability also for kidney TIA (SD=8.4%). The optimal sampling schedule for 2TP estimates of TIA is 1-2 days (21-52 h) p.t. followed by 3-5 days (71-126 h) p.t. for kidney, tumor, and spleen. Using the optimal sampling schedule, the largest magnitude MPE for 2TP estimates is 1.2% for spleen and highest variability is in tumor with SD=5.8%. The optimal sampling schedule for 3TP estimates of TIA is 1-2 days (21-52 h) p.t. followed by 3-5 days (71-126 h) p.t. and 6-8 days (144-194 h) p.t. for all structures. Using the optimal sampling schedule, the largest magnitude MPE for 3TP estimates is 2.5% for spleen and highest variability is in tumor with SD=2.1%. Simulated patient results corroborate these findings with similar optimal sampling schedules and errors. Many sub-optimal reduced time point sampling schedules also exhibit low error and variability. Conclusions. We show that reduced time point methods can be used to achieve acceptable average TIA errors over a wide range of imaging time points and sampling schedules while maintaining low uncertainty. This information can improve the feasibility of dosimetry for 177 Lu-DOTATATE and elucidate the uncertainty associated with non-ideal conditions.

SPECT reconstruction with a trained regularizer using CT-side information: Application to 177Lu SPECT imaging.

Improving low-count SPECT can shorten scans and support pre-therapy theranostic imaging for dosimetry-based treatment planning, especially with radionuclides like 177Lu known for low photon yields. Conventional methods often underperform in low-count settings, highlighting the need for trained regularization in model-based image reconstruction. This paper introduces a trained regularizer for SPECT reconstruction that leverages segmentation based on CT imaging. The regularizer incorporates CT-side information via a segmentation mask from a pre-trained network (nnUNet). In this proof-of-concept study, we used patient studies with 177Lu DOTATATE to train and tested with phantom and patient datasets, simulating pre-therapy imaging conditions. Our results show that the proposed method outperforms both standard unregularized EM algorithms and conventional regularization with CT-side information. Specifically, our method achieved marked improvements in activity quantification, noise reduction, and root mean square error. The enhanced low-count SPECT approach has promising implications for theranostic imaging, post-therapy imaging, whole body SPECT, and reducing SPECT acquisition times.

90Y SPECT scatter estimation and voxel dosimetry in radioembolization using a unified deep learning framework.

PURPOSE: 90Y SPECT-based dosimetry following radioembolization (RE) in liver malignancies is challenging due to the inherent scatter and the poor spatial resolution of bremsstrahlung SPECT. This study explores a deep-learning-based absorbed dose-rate estimation method for 90Y that mitigates the impact of poor SPECT image quality on dosimetry and the accuracy-efficiency trade-off of Monte Carlo (MC)-based scatter estimation and voxel dosimetry methods. METHODS: Our unified framework consists of three stages: convolutional neural network (CNN)-based bremsstrahlung scatter estimation, SPECT reconstruction with scatter correction (SC) and absorbed dose-rate map generation with a residual learning network (DblurDoseNet). The input to the framework is the measured SPECT projections and CT, and the output is the absorbed dose-rate map. For training and testing under realistic conditions, we generated a series of virtual patient phantom activity/density maps from post-therapy images of patients treated with 90Y-RE at our clinic. To train the scatter estimation network, we use the scatter projections for phantoms generated from MC simulation as the ground truth (GT). To train the dosimetry network, we use MC dose-rate maps generated directly from the activity/density maps of phantoms as the GT (Phantom + MC Dose). We compared performance of our framework (SPECT w/CNN SC + DblurDoseNet) and MC dosimetry (SPECT w/CNN SC + MC Dose) using normalized root mean square error (NRMSE) and normalized mean absolute error (NMAE) relative to GT. RESULTS: When testing on virtual patient phantoms, our CNN predicted scatter projections had NRMSE of 4.0% ± 0.7% on average. For the SPECT reconstruction with CNN SC, we observed a significant improvement on NRMSE (9.2% ± 1.7%), compared to reconstructions with no SC (149.5% ± 31.2%). In terms of virtual patient dose-rate estimation, SPECT w/CNN SC + DblurDoseNet had a NMAE of 8.6% ± 5.7% and 5.4% ± 4.8% in lesions and healthy livers, respectively; compared to 24.0% ± 6.1% and 17.7% ± 2.1% for SPECT w/CNN SC + MC Dose. In patient dose-rate maps, though no GT was available, we observed sharper lesion boundaries and increased lesion-to-background ratios with our framework. For a typical patient data set, the trained networks took ~ 1 s to generate the scatter estimate and ~ 20 s to generate the dose-rate map (matrix size: 512 × 512 × 194) on a single GPU (NVIDIA V100). CONCLUSION: Our deep learning framework, trained using true activity/density maps, has the potential to outperform non-learning voxel dosimetry methods such as MC that are dependent on SPECT image quality. Across comprehensive testing and evaluations on multiple targeted lesions and healthy livers in virtual patients, our proposed deep learning framework demonstrated higher (66% on average in terms of NMAE) estimation accuracy than the current "gold-standard" MC method. The enhanced computing speed with our framework without sacrificing accuracy is highly relevant for clinical dosimetry following 90Y-RE.

An International Study of Factors Affecting Variability of Dosimetry Calculations, Part 2: Overall Variabilities in Absorbed Dose.

Dosimetry for personalized radiopharmaceutical therapy has gained considerable attention. Many methods, tools, and workflows have been developed to estimate absorbed dose (AD). However, standardization is still required to reduce variability of AD estimates across centers. One effort for standardization is the Society of Nuclear Medicine and Molecular Imaging 177Lu Dosimetry Challenge, which comprised 5 tasks (T1-T5) designed to assess dose estimate variability associated with the imaging protocol (T1 vs. T2 vs. T3), segmentation (T1 vs. T4), time integration (T4 vs. T5), and dose calculation (T5) steps of the dosimetry workflow. The aim of this work was to assess the overall variability in AD calculations for the different tasks. Methods: Anonymized datasets consisting of serial planar and quantitative SPECT/CT scans, organ and lesion contours, and time-integrated activity maps of 2 patients treated with 177Lu-DOTATATE were made available globally for participants to perform dosimetry calculations and submit their results in standardized submission spreadsheets. The data were carefully curated for formal mistakes and methodologic errors. General descriptive statistics for ADs were calculated, and statistical analysis was performed to compare the results of different tasks. Variability in ADs was measured using the quartile coefficient of dispersion. Results: ADs to organs estimated from planar imaging protocols (T2) were lower by about 60% than those from pure SPECT/CT (T1), and the differences were statistically significant. Importantly, the average differences in dose estimates when at least 1 SPECT/CT acquisition was available (T1, T3, T4, T5) were within ±10%, and the differences with respect to T1 were not statistically significant for most organs and lesions. When serial SPECT/CT images were used, the quartile coefficients of dispersion of ADs for organs and lesions were on average less than 20% and 26%, respectively, for T1; 20% and 18%, respectively, for T4 (segmentations provided); and 10% and 5%, respectively, for T5 (segmentation and time-integrated activity images provided). Conclusion: Variability in ADs was reduced as segmentation and time-integration data were provided to participants. Our results suggest that SPECT/CT-based imaging protocols generate more consistent and less variable results than planar imaging methods. Effort at standardizing segmentation and fitting should be made, as this may substantially reduce variability in ADs.

Physics-Guided Deep Scatter Estimation by Weak Supervision for Quantitative SPECT.

Accurate scatter estimation is important in quantitative SPECT for improving image contrast and accuracy. With a large number of photon histories, Monte-Carlo (MC) simulation can yield accurate scatter estimation, but is computationally expensive. Recent deep learning-based approaches can yield accurate scatter estimates quickly, yet full MC simulation is still required to generate scatter estimates as ground truth labels for all training data. Here we propose a physics-guided weakly supervised training framework for fast and accurate scatter estimation in quantitative SPECT by using a 100× shorter MC simulation as weak labels and enhancing them with deep neural networks. Our weakly supervised approach also allows quick fine-tuning of the trained network to any new test data for further improved performance with an additional short MC simulation (weak label) for patient-specific scatter modelling. Our method was trained with 18 XCAT phantoms with diverse anatomies / activities and then was evaluated on 6 XCAT phantoms, 4 realistic virtual patient phantoms, 1 torso phantom and 3 clinical scans from 2 patients for 177Lu SPECT with single / dual photopeaks (113, 208 keV). Our proposed weakly supervised method yielded comparable performance to the supervised counterpart in phantom experiments, but with significantly reduced computation in labeling. Our proposed method with patient-specific fine-tuning achieved more accurate scatter estimates than the supervised method in clinical scans. Our method with physics-guided weak supervision enables accurate deep scatter estimation in quantitative SPECT, while requiring much lower computation in labeling, enabling patient-specific fine-tuning capability in testing.

A Multicenter Study on Observed Discrepancies Between Vendor-Stated and PET-Measured 90Y Activities for Both Glass and Resin Microsphere Devices.

Dosimetry-guided treatment planning in selective internal radiation therapy relies on accurate and reproducible measurement of administered activity. This 4-center, 5-PET-device study compared the manufacturer-declared 90Y activity in vials with quantitative 90Y PET/CT assessment of the same vials. We compared 90Y PET-measured activity (APET) for 56 90Y-labeled glass and 18 90Y-labeled resin microsphere vials with the calibrated activity specified by the manufacturer (AM). Additionally, the same analysis was performed for 4 90Y-chloride vials. The mean APET/AM ratio was 0.79 ± 0.04 (range, 0.71-0.89) for glass microspheres and 1.15 ± 0.06 (range, 1.05-1.25) for resin microspheres. The mean APET/AM ratio for 90Y-chloride vials was 1.00 ± 0.04 (range, 0.96-1.06). Thus, we found an average difference of 46% between glass and resin microsphere activity calibrations, whereas close agreement was found for chloride solutions. We expect that the reported discrepancies will promote further investigations to establish reliable and accurate patient dosimetry and dose-effect assessments.

Single-Time-Point Imaging for Dosimetry After [177Lu]Lu-DOTATATE: Accuracy of Existing Methods and Novel Data-Driven Models for Reducing Sensitivity to Time-Point Selection.

Estimation of the time-integrated activity (TIA) for dosimetry from imaging at a single time point (STP) facilitates the clinical translation of dosimetry-guided radiopharmaceutical therapy. However, the accuracy of the STP methods for TIA estimation varies on the basis of time-point selection. We constructed patient data-driven regression models to reduce the sensitivity to time-point selection and to compare these new models with commonly used STP methods. Methods: SPECT/CT performed at time period (TP) 1 (3-5 h), TP2 (days 1-2), TP3 (days 3-5), and TP4 (days 6-8) after cycle 1 of [177Lu]Lu-DOTATATE therapy involved 27 patients with 100 segmented tumors and 54 kidneys. Influenced by the previous physics-based STP models of Madsen et al. and Hänscheid et al., we constructed an STP prediction expression, TIA = A(t) × g(t), in a SPECT data-driven way (model 1), in which A(t) is the observed activity at imaging time t, and the curve, g(t), is estimated with a nonparametric generalized additive model by minimizing the normalized mean square error relative to the TIA derived from 4-time-point SPECT (reference TIA). Furthermore, we fit a generalized additive model that incorporates baseline biomarkers as auxiliary data in addition to the single activity measurement (model 2). Leave-one-out cross validation was performed to evaluate STP models using mean absolute error (MAE) and mean square error between the predicted and reference TIA. Results: At days 3-5, all evaluated STP methods performed very well, with an MAE of less than 7% (between-patient SD of <10%) for both kidneys and tumors. At other TPs, the Madsen method and data-driven models 1 and 2 performed reasonably well (MAEs < 17% for kidneys and < 32% for tumors), whereas the error with the Hänscheid method was substantially higher. The proof of concept of adding baseline biomarkers to the prediction model was demonstrated and showed a moderate enhancement at TP1, especially for estimating kidney TIA (MAE ± SD from 15.6% ± 1.3% to 11.8% ± 1.0%). Evaluations on 500 virtual patients using clinically relevant time-activity simulations showed a similar performance. Conclusion: The performance of the Madsen method and proposed data-driven models is less sensitive to TP selection than is the Hänscheid method. At the earliest TP, which is the most practical, the model incorporating baseline biomarkers outperforms other methods that rely only on the single activity measurement.

Experimental validation of Monte Carlo dosimetry for therapeutic beta emitters with radiochromic film in a 3D-printed phantom.

PURPOSE: Validation of dosimetry software, such as Monte Carlo (MC) radiation transport codes used for patient-specific absorbed dose estimation, is critical prior to their use in clinical decision making. However, direct experimental validation in the clinic is generally not performed for low/medium-energy beta emitters used in radiopharmaceutical therapy (RPT) due to the challenges of measuring energy deposited by short-range particles. Our objective was to design a practical phantom geometry for radiochromic film (RF)-based absorbed dose measurements of beta-emitting radionuclides and perform experiments to directly validate our in-house developed Dose Planning Method (DPM) MC code dedicated to internal dosimetry. METHODS: The experimental setup was designed for measuring absorbed dose from beta emitters that have a range sufficiently penetrating to ∼200 µm in water as well as to capture any photon contributions to absorbed dose. Assayed 177 Lu and 90 Y liquid sources, 13-450 MBq estimated to deliver 0.5-10 Gy to the sensitive layer of the RF, were injected into the cavity of two 3D-printed half-cylinders that had been sealed with 12.7 µm or 25.4 µm thick Kapton Tape. A 3.8 × 6 cm strip of GafChromic EBT3 RF was sandwiched between the two taped half-cylinders. After 2-48 h exposures, films were retrieved and wipe tested for contamination. Absorbed dose to the RF was measured using a commercial triple-channel dosimetry optimization method and a calibration generated via 6 MV photon beam. Profiles were analyzed across the central 1 cm2 area of the RF for validation. Eleven experiments were completed with 177 Lu and nine with 90 Y both in saline and a bone equivalent solution. Depth dose curves were generated for 177 Lu and 90 Y stacking multiple RF strips between a single filled half-cylinder and an acrylic backing. All experiments were modeled in DPM to generate voxelized MC absorbed dose estimates. We extended our study to benchmark general purpose MC codes MCNP6 and EGSnrc against the experimental results as well. RESULTS: A total of 20 experiments showed that both the 3D-printed phantoms and the final absorbed dose values were reproducible. The agreement between the absorbed dose estimates from the RF measurements and DPM was on average -4.0% (range -10.9% to 3.2%) for all single film 177 Lu experiments and was on average -1.0% (range -2.7% to 0.7%) for all single film 90 Y experiments. Absorbed depth dose estimates by DPM agreed with RF on average 1.2% (range -8.0% to 15.2%) across all depths for 177 Lu and on average 4.0% (range -5.0% to 9.3%) across all depths for 90 Y. DPM absorbed dose estimates agreed with estimates from EGSnrc and MCNP across the board, within 4.7% and within 3.4% for 177 Lu and 90 Y respectively, for all geometries and across all depths. MC showed that absorbed dose to RF from betas was greater than 92% of the total (betas + other radiations) for 177 Lu, indicating measurement of dominant beta contribution with our design. CONCLUSIONS: The reproducible results with a RF insert in a simple phantom designed for liquid sources demonstrate that this is a reliable setup for experimentally validating dosimetry algorithms used in therapies with beta-emitting unsealed sources. Absorbed doses estimated with the DPM MC code showed close agreement with RF measurement and with results from two general purpose MC codes, thereby validating the use of this algorithms for clinical RPT dosimetry.

177Lu-DOTATATE Theranostics: Predicting Renal Dosimetry From Pretherapy 68Ga-DOTATATE PET and Clinical Biomarkers.

PURPOSE: Pretreatment predictions of absorbed doses can be especially valuable for patient selection and dosimetry-guided individualization of radiopharmaceutical therapy. Our goal was to build regression models using pretherapy 68Ga-DOTATATE PET uptake data and other baseline clinical factors/biomarkers to predict renal absorbed dose delivered by 177Lu-DOTATATE peptide receptor radionuclide therapy (177Lu-PRRT) for neuroendocrine tumors. We explore the combination of biomarkers and 68Ga PET uptake metrics, hypothesizing that they will improve predictive power over univariable regression. PATIENTS AND METHODS: Pretherapy 68Ga-DOTATATE PET/CTs were analyzed for 25 patients (50 kidneys) who also underwent quantitative 177Lu SPECT/CT imaging at approximately 4, 24, 96, and 168 hours after cycle 1 of 177Lu-PRRT. Kidneys were contoured on the CT of the PET/CT and SPECT/CT using validated deep learning-based tools. Dosimetry was performed by coupling the multi-time point SPECT/CT images with an in-house Monte Carlo code. Pretherapy renal PET SUV metrics, activity concentration per injected activity (Bq/mL/MBq), and other baseline clinical factors/biomarkers were investigated as predictors of the 177Lu SPECT/CT-derived mean absorbed dose per injected activity to the kidneys using univariable and bivariable models. Leave-one-out cross-validation (LOOCV) was used to estimate model performance using root mean squared error and absolute percent error in predicted renal absorbed dose including mean absolute percent error (MAPE) and associated standard deviation (SD). RESULTS: The median therapy-delivered renal dose was 0.5 Gy/GBq (range, 0.2-1.0 Gy/GBq). In LOOCV of univariable models, PET uptake (Bq/mL/MBq) performs best with MAPE of 18.0% (SD = 13.3%), and estimated glomerular filtration rate (eGFR) gives an MAPE of 28.5% (SD = 19.2%). Bivariable regression with both PET uptake and eGFR gives LOOCV MAPE of 17.3% (SD = 11.8%), indicating minimal improvement over univariable models. CONCLUSIONS: Pretherapy 68Ga-DOTATATE PET renal uptake can be used to predict post-177Lu-PRRT SPECT-derived mean absorbed dose to the kidneys with accuracy within 18%, on average. Compared with PET uptake alone, including eGFR in the same model to account for patient-specific kinetics did not improve predictive power. Following further validation of these preliminary findings in an independent cohort, predictions using renal PET uptake can be used in the clinic for patient selection and individualization of treatment before initiating the first cycle of PRRT.