Lack of repeatability of radiomic features derived from PET scans: Results from a 18 F-DCFPyL test-retest cohort.

OBJECTIVES: PET-based radiomic metrics are increasingly utilized as predictive image biomarkers. However, the repeatability of radiomic features on PET has not been assessed in a test-retest setting. The prostate-specific membrane antigen-targeted compound 18 F-DCFPyL is a high-affinity, high-contrast PET agent that we utilized in a test-retest cohort of men with metastatic prostate cancer (PC). METHODS: Data of 21 patients enrolled in a prospective clinical trial with histologically proven PC underwent two 18 F-DCFPyL PET scans within 7 days, using identical acquisition and reconstruction parameters. Sites of disease were segmented and a set of 29 different radiomic parameters were assessed on both scans. We determined repeatability of quantification by using Pearson's correlations, within-subject coefficient of variation (wCOV), and Bland-Altman analysis. RESULTS: In total, 230 lesions (177 bone, 38 lymph nodes, 15 others) were assessed on both scans. For all investigated radiomic features, a broad range of inter-scan correlation was found (r, 0.07-0.95), with acceptable reproducibility for entropy and homogeneity (wCOV, 16.0% and 12.7%, respectively). On Bland-Altman analysis, no systematic increase or decrease between the scans was observed for either parameter (±1.96 SD: 1.07/-1.30, 0.23/-0.18, respectively). The remaining 27 tested radiomic metrics, however, achieved unacceptable high wCOV (≥21.7%). CONCLUSION: Many common radiomic features derived from a test-retest PET study had poor repeatability. Only Entropy and homogeneity achieved good repeatability, supporting the notion that those image biomarkers may be incorporated in future clinical trials. Those radiomic features based on high frequency aspects of images appear to lack the repeatability on PET to justify further study.

Fisher information analysis of list-mode SPECT emission data for joint estimation of activity and attenuation distribution.

The potential to perform attenuation and scatter compensation (ASC) in single-photon emission computed tomography (SPECT) imaging without a separate transmission scan is highly significant. In this context, attenuation in SPECT is primarily due to Compton scattering, where the probability of Compton scatter is proportional to the attenuation coefficient of the tissue and the energy of the scattered photon and the scattering angle are related. Based on this premise, we investigated whether the SPECT scattered-photon data acquired in list-mode (LM) format and including the energy information can be used to estimate the attenuation map. For this purpose, we propose a Fisher-information-based method that yields the Cramer-Rao bound (CRB) for the task of jointly estimating the activity and attenuation distribution using only the SPECT emission data. In the process, a path-based formalism to process the LM SPECT emission data, including the scattered-photon data, is proposed. The Fisher information method was implemented on NVIDIA graphics processing units (GPU) for acceleration. The method was applied to analyze the information content of SPECT LM emission data, which contains up to first-order scattered events, in a simulated SPECT system with parameters modeling a clinical system using realistic computational studies with 2-D digital synthetic and anthropomorphic phantoms. The method was also applied to LM data containing up to second-order scatter for a synthetic phantom. Experiments with anthropomorphic phantoms simulated myocardial perfusion and dopamine transporter (DaT)-Scan SPECT studies. The results show that the CRB obtained for the attenuation and activity coefficients was typically much lower than the true value of these coefficients. An increase in the number of detected photons yielded lower CRB for both the attenuation and activity coefficients. Further, we observed that systems with better energy resolution yielded a lower CRB for the attenuation coefficient. Overall, the results provide evidence that LM SPECT emission data, including the scattered photons, contains information to jointly estimate the activity and attenuation coefficients.

Intraprocedural C-arm dual-phase cone-beam enhancement patterns correlate with tumor absorbed dose after radioembolization.

BACKGROUND: Recent studies have shown a clear relationship between absorbed dose and tumor response to treatment after hepatic radioembolization. These findings help to create more personalized treatment planning and dosimetry. However, crucial to this goal is the ability to predict the dose distribution prior to treatment. The microsphere distribution is ultimately determined by (i) the hepatic vasculature and the resulting blood flow dynamics and (ii) the catheter position. PURPOSE: To show that pretreatment, intra-procedural imaging of blood flow patterns, as quantified by catheter-directed intra-arterial contrast enhancement, correlate with posttreatment microsphere accumulation and, consequently, absorbed dose. MATERIALS AND METHODS: Patients who participated in a clinical trial (NCT01177007) and for whom both a pretreatment dual-phase contrast-enhanced cone-beam CT (CBCT) and a posttreatment 90Y PET/CT scan were available were included in this retrospective study. Tumors and perfused volumes were manually delineated on the CBCT by an experienced radiologist. The mean, sum, and standard deviation of the voxels in each volume were recorded. The delineations were transferred to the PET-based absorbed dose maps by coregistration of the corresponding CTs. Linear multiple regression was used to correlate pretreatment CBCT enhancement to posttreatment 90Y PET/CT-based absorbed dose in each region. Leave-one-out cross-validation and Bland-Altman analyses were performed on the predicted versus measured absorbed doses. RESULTS: Nine patients, with a total of 23 tumors were included. All presented with hepatocellular carcinoma (HCC). Visually, all patients had a clear correspondence between CBCT enhancement and absorbed dose. The correlation between CBCT enhancement and posttherapy absorbed tumor dose based was strong (R2 = 0.91), and moderate for the non-tumor liver tissue (R2 = 0.61). Limits of agreement were approximately ±55 Gray for tumor tissue. CONCLUSION: There is a linear relationship between pretreatment blood dynamics in HCC tumors and posttreatment absorbed dose, which, if shown to be generalizable, allows for pretreatment tumor absorbed dose prediction.

Girth-based administered activity for pediatric 99m Tc-DMSA SPECT.

BACKGROUND: Pediatric molecular imaging requires a balance between administering an activity that will yield sufficient diagnostic image quality while maintaining patient radiation exposure at acceptable levels. In current clinical practice, this balance is arrived at by the current North American Consensus Guidelines in which patient weight is used to recommend the administered activity (AA). PURPOSE: We have previously demonstrated that girth (waist circumference at the level of the kidneys) is better at equalizing image quality than patient weight for pediatric Tc-99m DMSA renal function imaging. However, the correlation between image quality (IQ), AA, and patient girth has not been rigorously and systematically developed. In this work, we generate a series of curves showing the tradeoff between AA and IQ as a function of patient girth, providing the data for standards bodies to develop the next generation of dosing guideline for pediatric DMSA SPECT. METHODS: An anthropomorphic phantom series that included variations in age (5, 10, and 15 years), gender (M, F), local body morphometry (5, 10, 50, 90, and 95th girth percentiles), and kidney size (±15% standard size), was used to generate realistic SPECT projections. A fixed and clinically challenging defect-to-organ volume percentage (0.49% of renal cortex value) was used to model a focal defect with zero uptake (i.e., full local loss of renal function). Task-based IQ assessment methods were used to rigorously measure IQ in terms of renal perfusion defect detectability. This assessment was performed at multiple count levels (corresponding to various AAs) for groups of patients that had similar girths and defect sizes. Receiver-operating characteristics (ROC) analysis was applied; the area under the ROC curve (AUC) was used as a figure-of-merit for task performance. Curves showing the tradeoff between AUC and AA were generated for these groups of phantoms. RESULTS: Overall, the girth-based dosing method suggested different amounts of AA compared to weight-based dosing for the phantoms that had a relatively large body weight but a small girth or phantoms with relatively small bodyweight but large girth. Reductions of AA to 62.9% compared to weight-based dosing guidelines can potentially be realized while maintaining a baseline (AUC = 0.80) IQ for certain 15-year-olds who have a relatively small girth and large defect size. Note that the task-based IQ results are heavily dependent on the simulated defect size for the defect detection task and the appropriate AUC value must be decided by the physicians for this diagnostic task. These results are based purely on simulation and are subject to future clinical validation. CONCLUSIONS: The study provides simulation-based IQ-AA data for a girth-based dosing method for pediatric renal SPECT, suggesting that patient waist circumference at the level of kidneys should be considered in selecting the AA needed to achieve an acceptable IQ. This data may be useful for standards bodies to develop girth-based dosing guidelines.

Compensation for spill-in and spill-out partial volume effects in cardiac PET imaging.

BACKGROUND: Partial volume effects (PVEs) in PET imaging result in incorrect regional activity estimates due to both spill-out and spill-in from activity in neighboring regions. It is important to compensate for both effects to achieve accurate quantification. In this study, an image-based partial volume compensation (PVC) method was developed and validated for cardiac PET. METHODS AND RESULTS: The method uses volume-of-interest (VOI) maps segmented from contrast-enhanced CTA images to compensate for both spill-in and spill-out in each VOI. The PVC method was validated with simulation studies and also applied to images of dog cardiac perfusion PET data. The PV effects resulting from cardiac motion and myocardial uptake defects were investigated and the efficacy of the proposed PVC method in compensating for these effects was evaluated. RESULTS: Results indicate that the magnitude and the direction of PVEs in cardiac imaging change over time. This affects the accuracy of activity distributions estimates obtained during dynamic studies. The defect regions have different PVEs as compared to the normal myocardium. Cardiac motion contributes around 10% to the PVEs. PVC effectively removed both spill-in and spill-out in cardiac imaging. CONCLUSIONS: PVC improved left ventricular wall uniformity and quantitative accuracy. The best strategy for PVC was to compensate for the PVEs in each cardiac phase independently and treat severe uptake defects as independent regions from the normal myocardium.

Three-dimensional radiobiological dosimetry of kidneys for treatment planning in peptide receptor radionuclide therapy.

PURPOSE: Peptide receptor radionuclide therapy (PRRT) delivers high absorbed doses to kidneys and may lead to permanent nephropathy. Reliable dosimetry of kidneys is thus critical for safe and effective PRRT. The aim of this work was to assess the feasibility of planning PRRT based on 3D radiobiological dosimetry (3D-RD) in order to optimize both the amount of activity to administer and the fractionation scheme, while limiting the absorbed dose and the biological effective dose (BED) to the renal cortex. METHODS: Planar and SPECT data were available for a patient examined with (111)In-DTPA-octreotide at 0.5 (planar only), 4, 24, and 48 h post-injection. Absorbed dose and BED distributions were calculated for common therapeutic radionuclides, i.e., (111)In, (90)Y and (177)Lu, using the 3D-RD methodology. Dose-volume histograms were computed and mean absorbed doses to kidneys, renal cortices, and medullae were compared with results obtained using the MIRD schema (S-values) with the multiregion kidney dosimetry model. Two different treatment planning approaches based on (1) the fixed absorbed dose to the cortex and (2) the fixed BED to the cortex were then considered to optimize the activity to administer by varying the number of fractions. RESULTS: Mean absorbed doses calculated with 3D-RD were in good agreement with those obtained with S-value-based SPECT dosimetry for (90)Y and (177)Lu. Nevertheless, for (111)In, differences of 14% and 22% were found for the whole kidneys and the cortex, respectively. Moreover, the authors found that planar-based dosimetry systematically underestimates the absorbed dose in comparison with SPECT-based methods, up to 32%. Regarding the 3D-RD-based treatment planning using a fixed BED constraint to the renal cortex, the optimal number of fractions was found to be 3 or 4, depending on the radionuclide administered and the value of the fixed BED. Cumulative activities obtained using the proposed simulated treatment planning are compatible with real activities administered to patients in PRRT. CONCLUSIONS: The 3D-RD treatment planning approach based on the fixed BED was found to be the method of choice for clinical implementation in PRRT by providing realistic activity to administer and number of cycles. While dividing the activity in several cycles is important to reduce renal toxicity, the clinical outcome of fractionated PRRT should be investigated in the future.

Development and evaluation of an improved quantitative (90)Y bremsstrahlung SPECT method.

PURPOSE: Yttrium-90 ((90)Y) is one of the most commonly used radionuclides in targeted radionuclide therapy (TRT). Since it decays with essentially no gamma photon emissions, surrogate radionuclides (e.g., (111)In) or imaging agents (e.g., (99m)Tc MAA) are typically used for treatment planning. It would, however, be useful to image (90)Y directly in order to confirm that the distributions measured with these other radionuclides or agents are the same as for the (90)Y labeled agents. As a result, there has been a great deal of interest in quantitative imaging of (90)Y bremsstrahlung photons using single photon emission computed tomography (SPECT) imaging. The continuous and broad energy distribution of bremsstrahlung photons, however, imposes substantial challenges on accurate quantification of the activity distribution. The aim of this work was to develop and evaluate an improved quantitative (90)Y bremsstrahlung SPECT reconstruction method appropriate for these imaging applications. METHODS: Accurate modeling of image degrading factors such as object attenuation and scatter and the collimator-detector response is essential to obtain quantitatively accurate images. All of the image degrading factors are energy dependent. Thus, the authors separated the modeling of the bremsstrahlung photons into multiple categories and energy ranges. To improve the accuracy, the authors used a bremsstrahlung energy spectrum previously estimated from experimental measurements and incorporated a model of the distance between (90)Y decay location and bremsstrahlung emission location into the SIMIND code used to generate the response functions and kernels used in the model. This improved Monte Carlo bremsstrahlung simulation was validated by comparison to experimentally measured projection data of a (90)Y line source. The authors validated the accuracy of the forward projection model for photons in the various categories and energy ranges using the validated Monte Carlo (MC) simulation method. The forward projection model was incorporated into an iterative ordered subsets-expectation maximization (OS-EM) reconstruction code to allow for quantitative SPECT reconstruction. The resulting code was validated using both a physical phantom experiment with spherical objects in a warm background and a realistic anatomical phantom simulation. In the physical phantom study, the authors evaluated the method in terms of quantitative accuracy of activity estimates in the spheres; in the simulation study, the authors evaluated the accuracy and precision of activity estimates from various organs and compared them to results from a previously proposed method. RESULTS: The authors demonstrated excellent agreement between the experimental measurement and Monte Carlo simulation. In the XCAT phantom simulation, the proposed method achieved much better accuracy in the modeling (error in photon counts was -1.1 %) compared to a previously proposed method (errors were more than 20 %); the quantitative accuracy of activity estimates was excellent for all organs (errors were from -1.6 % to 11.9 %) and comparable to previously published results for (131)I using the same collimator. CONCLUSIONS: The proposed (90)Y bremsstrahlung SPECT reconstruction method provided very accurate estimates of organ activities, with accuracies approaching those previously observed for (131)I. The method may be useful in verifying organ doses for targeted radionuclide therapy using (90)Y.

TransMorph: Transformer for unsupervised medical image registration.

In the last decade, convolutional neural networks (ConvNets) have been a major focus of research in medical image analysis. However, the performances of ConvNets may be limited by a lack of explicit consideration of the long-range spatial relationships in an image. Recently, Vision Transformer architectures have been proposed to address the shortcomings of ConvNets and have produced state-of-the-art performances in many medical imaging applications. Transformers may be a strong candidate for image registration because their substantially larger receptive field enables a more precise comprehension of the spatial correspondence between moving and fixed images. Here, we present TransMorph, a hybrid Transformer-ConvNet model for volumetric medical image registration. This paper also presents diffeomorphic and Bayesian variants of TransMorph: the diffeomorphic variants ensure the topology-preserving deformations, and the Bayesian variant produces a well-calibrated registration uncertainty estimate. We extensively validated the proposed models using 3D medical images from three applications: inter-patient and atlas-to-patient brain MRI registration and phantom-to-CT registration. The proposed models are evaluated in comparison to a variety of existing registration methods and Transformer architectures. Qualitative and quantitative results demonstrate that the proposed Transformer-based model leads to a substantial performance improvement over the baseline methods, confirming the effectiveness of Transformers for medical image registration.

Development and evaluation of a model-based downscatter compensation method for quantitative I-131 SPECT.

PURPOSE: The radionuclide 131I has found widespread use in targeted radionuclide therapy (TRT), partly due to the fact that it emits photons that can be imaged to perform treatment planning or posttherapy dose verification as well as beta rays that are suitable for therapy. In both the treatment planning and dose verification applications, it is necessary to estimate the activity distribution in organs or tumors at several time points. In vivo estimates of the 131I activity distribution at each time point can be obtained from quantitative single-photon emission computed tomography (QSPECT) images and organ activity estimates can be obtained either from QSPECT images or quantification of planar projection data. However, in addition to the photon used for imaging, 131I decay results in emission of a number of other higher-energy photons with significant abundances. These higher-energy photons can scatter in the body, collimator, or detector and be counted in the 364 keV photopeak energy window, resulting in reduced image contrast and degraded quantitative accuracy; these photons are referred to as downscatter. The goal of this study was to develop and evaluate a model-based downscatter compensation method specifically designed for the compensation of high-energy photons emitted by 131I and detected in the imaging energy window. METHODS: In the evaluation study, we used a Monte Carlo simulation (MCS) code that had previously been validated for other radionuclides. Thus, in preparation for the evaluation study, we first validated the code for 131I imaging simulation by comparison with experimental data. Next, we assessed the accuracy of the downscatter model by comparing downscatter estimates with MCS results. Finally, we combined the downscatter model with iterative reconstruction-based compensation for attenuation (A) and scatter (S) and the full (D) collimator-detector response of the 364 keV photons to form a comprehensive compensation method. We evaluated this combined method in terms of quantitative accuracy using the realistic 3D NCAT phantom and an activity distribution obtained from patient studies. We compared the accuracy of organ activity estimates in images reconstructed with and without addition of downscatter compensation from projections with and without downscatter contamination. RESULTS: We observed that the proposed method provided substantial improvements in accuracy compared to no downscatter compensation and had accuracies comparable to reconstructions from projections without downscatter contamination. CONCLUSIONS: The results demonstrate that the proposed model-based downscatter compensation method is effective and may have a role in quantitative 131I imaging.

Material separation in x-ray CT with energy resolved photon-counting detectors.

PURPOSE: The objective of the study was to demonstrate that, in x-ray computed tomography (CT), more than two types of materials can be effectively separated with the use of an energy resolved photon-counting detector and classification methodology. Specifically, this applies to the case when contrast agents that contain K-absorption edges in the energy range of interest are present in the object. This separation is enabled via the use of recently developed energy resolved photon-counting detectors with multiple thresholds, which allow simultaneous measurements of the x-ray attenuation at multiple energies. METHODS: To demonstrate this capability, we performed simulations and physical experiments using a six-threshold energy resolved photon-counting detector. We imaged mouse-sized cylindrical phantoms filled with several soft-tissue-like and bone-like materials and with iodine-based and gadolinium-based contrast agents. The linear attenuation coefficients were reconstructed for each material in each energy window and were visualized as scatter plots between pairs of energy windows. For comparison, a dual-kVp CT was also simulated using the same phantom materials. In this case, the linear attenuation coefficients at the lower kVp were plotted against those at the higher kVp. RESULTS: In both the simulations and the physical experiments, the contrast agents were easily separable from other soft-tissue-like and bone-like materials, thanks to the availability of the attenuation coefficient measurements at more than two energies provided by the energy resolved photon-counting detector. In the simulations, the amount of separation was observed to be proportional to the concentration of the contrast agents; however, this was not observed in the physical experiments due to limitations of the real detector system. We used the angle between pairs of attenuation coefficient vectors in either the 5-D space (for non-contrast-agent materials using energy resolved photon-counting acquisition) or a 2-D space (for contrast agents using energy resolved photon-counting acquisition and all materials using dual-kVp acquisition) as a measure of the degree of separation. Compared to dual-kVp techniques, an energy resolved detector provided a larger separation and the ability to separate different target materials using measurements acquired in different energy window pairs with a single x-ray exposure. CONCLUSIONS: We concluded that x-ray CT with an energy resolved photon-counting detector with more than two energy windows allows the separation of more than two types of materials, e.g., soft-tissue-like, bone-like, and one or more materials with K-edges in the energy range of interest. Separating material types using energy resolved photon-counting detectors has a number of advantages over dual-kVp CT in terms of the degree of separation and the number of materials that can be separated simultaneously.

Modeling the performance of a photon counting x-ray detector for CT: energy response and pulse pileup effects.

PURPOSE: Recently, photon counting x-ray detectors (PCXDs) with energy discrimination capabilities have been developed for potential use in clinical computed tomography (CT) scanners. These PCXDs have great potential to improve the quality of CT images due to the absence of electronic noise and weights applied to the counts and the additional spectral information. With high count rates encountered in clinical CT, however, coincident photons are recorded as one event with a higher or lower energy due to the finite speed of the PCXD. This phenomenon is called a "pulse pileup event" and results in both a loss of counts (called "deadtime losses") and distortion of the recorded energy spectrum. Even though the performance of PCXDs is being improved, it is essential to develop algorithmic methods based on accurate models of the properties of detectors to compensate for these effects. To date, only one PCXD (model DXMCT-1, DxRay, Inc., Northridge, CA) has been used for clinical CT studies. The aim of that study was to evaluate the agreement between data measured by DXMCT-1 and those predicted by analytical models for the energy response, the deadtime losses, and the distorted recorded spectrum caused by pulse pileup effects. METHODS: An energy calibration was performed using 99mTc (140 keV), 57Co (122 keV), and an x-ray beam obtained with four x-ray tube voltages (35, 50, 65, and 80 kVp). The DXMCT-1 was placed 150 mm from the x-ray focal spot; the count rates and the spectra were recorded at various tube current values from 10 to 500 microA for a tube voltage of 80 kVp. Using these measurements, for each pulse height comparator we estimated three parameters describing the photon energy-pulse height curve, the detector deadtime tau, a coefficient k that relates the x-ray tube current I to an incident count rate a by a = k x I, and the incident spectrum. The mean pulse shape of all comparators was acquired in a separate study and was used in the model to estimate the distorted recorded spectrum. The agreement between data measured by the DXMCT-1 and those predicted by the models was quantified by the coefficient of variation (COV), i.e., the root mean square difference divided by the mean of the measurement. RESULTS: Photon energy versus pulse height curves calculated with an analytical model and those measured using the DXMCT-1 were in agreement within 0.2% in terms of the COV. The COV between the output count rates measured and those predicted by analytical models was 2.5% for deadtime losses of up to 60%. The COVs between spectra measured and those predicted by the detector model were within 3.7%-7.2% with deadtime losses of 19%-46%. CONCLUSIONS: It has been demonstrated that the performance of the DXMCT-1 agreed exceptionally well with the analytical models regarding the energy response, the count rate, and the recorded spectrum with pulse pileup effects. These models will be useful in developing methods to compensate for these effects in PCXD-based clinical CT systems.

Automated segmentation of lung, liver, and liver tumors from Tc-99m MAA SPECT/CT images for Y-90 radioembolization using convolutional neural networks.

PURPOSE: 90 Y selective internal radiation therapy (SIRT) has become a safe and effective treatment option for liver cancer. However, segmentation of target and organ-at-risks is labor-intensive and time-consuming in 90 Y SIRT planning. In this study, we developed a convolutional neural network (CNN)-based method for automated lungs, liver, and tumor segmentation on 99m Tc-MAA SPECT/CT images for 90 Y SIRT planning. METHODS: 99m Tc-MAA SPECT/CT images and corresponding clinical segmentations were retrospectively collected from 56 patients who underwent 90 Y SIRT. The collected data were used to train three CNN-based segmentation algorithms for lungs, liver, and tumor segmentation. Segmentation performance was evaluated using the Dice similarity coefficient (DSC), surface DSC, and average symmetric surface distance (ASSD). Dosimetric parameters (volume, counts, and lung shunt fraction) were measured from the segmentation results and were compared with clinical reference segmentations. RESULTS: The evaluation results show that the method can accurately segment lungs, liver, and tumor with median [interquartile range] DSCs of 0.98 [0.97-0.98], 0.91 [0.83-0.93], and 0.85 [0.71-0.88]; surface DSCs of 0.99 [0.97-0.99], 0.86 [0.77-0.93], and 0.85 [0.62-0.93], and ASSDs of 0.91 [0.69-1.5], 4.8 [2.6-8.4], and 4.7 [3.5-9.2] mm, respectively. Dosimetric parameters from the three segmentation networks show relationship with those from the reference segmentations. The overall segmentation took about 1 min per patient on an NVIDIA RTX-2080Ti GPU. CONCLUSION: This work presents CNN-based algorithms to segment lungs, liver, and tumor from 99m Tc-MAA SPECT/CT images. The results demonstrated the potential of the proposed CNN-based segmentation method for assisting 90 Y SIRT planning while drastically reducing operator time.

Dosimetric impact of Ac-227 in accelerator-produced Ac-225 for alpha-emitter radiopharmaceutical therapy of patients with hematological malignancies: a pharmacokinetic modeling analysis.

BACKGROUND: Actinium-225 is an alpha-particle emitter under investigation for use in radiopharmaceutical therapy. To address limited supply, accelerator-produced 225Ac has been recently made available. Accelerator-produced 225Ac via 232Th irradiation (denoted 225/7Ac) contains a low percentage (0.1-0.3%) of 227Ac (21.77-year half-life) activity at end of bombardment. Using pharmacokinetic modeling, we have examined the dosimetric impact of 227Ac on the use of accelerator-produced 225Ac for radiopharmaceutical therapy. We examine the contribution of 227Ac and its daughters to tissue absorbed doses. The dosimetric analysis was performed for antibody-conjugated 225/7Ac administered intravenously to treat patients with hematological cancers. Published pharmacokinetic models are used to obtain the distribution of 225/7Ac-labeled antibody and also the distribution of either free or antibody-conjugated 227Th. RESULTS: Based on our modeling, the tissue specific absorbed dose from 227Ac would be negligible in the context of therapy, less than 0.02 mGy/MBq for the top 6 highest absorbed tissues and less than 0.007 mGy/MBq for all other tissues. Compared to that from 225Ac, the absorbed dose from 227Ac makes up a very small component (less than 0.04%) of the total absorbed dose delivered to the 6 highest dose tissues: red marrow, spleen, endosteal cells, liver, lungs and kidneys when accelerator produced 225/7Ac-conjugated anti-CD33 antibody is used to treat leukemia patients. For all tissues, the dominant contributor to the absorbed dose arising from the 227Ac is 227Th, the first daughter of 227Ac which has the potential to deliver absorbed dose both while it is antibody-bound and while it is free. CONCLUSIONS: These results suggest that the absorbed dose arising from 227Ac to normal organs would be negligible for an 225/7Ac-labeled antibody that targets hematological cancer.

Learning fuzzy clustering for SPECT/CT segmentation via convolutional neural networks.

PURPOSE: Quantitative bone single-photon emission computed tomography (QBSPECT) has the potential to provide a better quantitative assessment of bone metastasis than planar bone scintigraphy due to its ability to better quantify activity in overlapping structures. An important element of assessing the response of bone metastasis is accurate image segmentation. However, limited by the properties of QBSPECT images, the segmentation of anatomical regions-of-interests (ROIs) still relies heavily on the manual delineation by experts. This work proposes a fast and robust automated segmentation method for partitioning a QBSPECT image into lesion, bone, and background. METHODS: We present a new unsupervised segmentation loss function and its semi- and supervised variants for training a convolutional neural network (ConvNet). The loss functions were developed based on the objective function of the classical Fuzzy C-means (FCM) algorithm. The first proposed loss function can be computed within the input image itself without any ground truth labels, and is thus unsupervised; the proposed supervised loss function follows the traditional paradigm of the deep learning-based segmentation methods and leverages ground truth labels during training. The last loss function is a combination of the first and the second and includes a weighting parameter, which enables semi-supervised segmentation using deep learning neural network. EXPERIMENTS AND RESULTS: We conducted a comprehensive study to compare our proposed methods with ConvNets trained using supervised, cross-entropy and Dice loss functions, and conventional clustering methods. The Dice similarity coefficient (DSC) and several other metrics were used as figures of merit as applied to the task of delineating lesion and bone in both simulated and clinical SPECT/CT images. We experimentally demonstrated that the proposed methods yielded good segmentation results on a clinical dataset even though the training was done using realistic simulated images. On simulated SPECT/CT, the proposed unsupervised model's accuracy was greater than the conventional clustering methods while reducing computation time by 200-fold. For the clinical QBSPECT/CT, the proposed semi-supervised ConvNet model, trained using simulated images, produced DSCs of 0.75 and 0.74 for lesion and bone segmentation in SPECT, and a DSC of 0.79 bone segmentation of CT images. These DSCs were larger than that for standard segmentation loss functions by > 0.4 for SPECT segmentation, and > 0.07 for CT segmentation with P-values < 0.001 from a paired t-test. CONCLUSIONS: A ConvNet-based image segmentation method that uses novel loss functions was developed and evaluated. The method can operate in unsupervised, semi-supervised, or fully-supervised modes depending on the availability of annotated training data. The results demonstrated that the proposed method provides fast and robust lesion and bone segmentation for QBSPECT/CT. The method can potentially be applied to other medical image segmentation applications.

DeepAMO: a multi-slice, multi-view anthropomorphic model observer for visual detection tasks performed on volume images.

Purpose: We propose a deep learning-based anthropomorphic model observer (DeepAMO) for image quality evaluation of multi-orientation, multi-slice image sets with respect to a clinically realistic 3D defect detection task. Approach: The DeepAMO is developed based on a hypothetical model of the decision process of a human reader performing a detection task using a 3D volume. The DeepAMO is comprised of three sequential stages: defect segmentation, defect confirmation (DC), and rating value inference. The input to the DeepAMO is a composite image, typical of that used to view 3D volumes in clinical practice. The output is a rating value designed to reproduce a human observer's defect detection performance. In stages 2 and 3, we propose: (1) a projection-based DC block that confirms defect presence in two 2D orthogonal orientations and (2) a calibration method that "learns" the mapping from the features of stage 2 to the distribution of observer ratings from the human observer rating data (thus modeling inter- or intraobserver variability) using a mixture density network. We implemented and evaluated the DeepAMO in the context of Tc 99 m -DMSA SPECT imaging. A human observer study was conducted, with two medical imaging physics graduate students serving as observers. A 5 × 2 -fold cross-validation experiment was conducted to test the statistical equivalence in defect detection performance between the DeepAMO and the human observer. We also compared the performance of the DeepAMO to an unoptimized implementation of a scanning linear discriminant observer (SLDO). Results: The results show that the DeepAMO's and human observer's performances on unseen images were statistically equivalent with a margin of difference ( Δ AUC ) of 0.0426 at p < 0.05 , using 288 training images. A limited implementation of an SLDO had a substantially higher AUC (0.99) compared to the DeepAMO and human observer. Conclusion: The results show that the DeepAMO has the potential to reproduce the absolute performance, and not just the relative ranking of human observers on a clinically realistic defect detection task, and that building conceptual components of the human reading process into deep learning-based models can allow training of these models in settings where limited training images are available.

Prospective SPECT-CT Organ Dosimetry-Driven Radiation-Absorbed Dose Escalation Using the In-111 (111In)/Yttrium 90 (90Y) Ibritumomab Tiuxetan (Zevalin®) Theranostic Pair in Patients with Lymphoma at Myeloablative Dose Levels.

PURPOSE: We prospectively evaluated the feasibility of SPECT-CT/planar organ dosimetry-based radiation dose escalation radioimmunotherapy in patients with recurrent non-Hodgkin's lymphoma using the theranostic pair of 111In and 90Y anti-CD20 ibritumomab tiuxetan (Zevalin®) at myeloablative radiation-absorbed doses with autologous stem cell support. We also assessed acute non-hematopoietic toxicity and early tumor response in this two-center outpatient study. METHODS: 24 patients with CD20-positive relapsed or refractory rituximab-sensitive, low-grade, mantle cell, or diffuse large-cell NHL, with normal organ function, platelet counts > 75,000/mm3, and <35% tumor involvement in the marrow were treated with Rituximab (375 mg/m2) weekly for 4 consecutive weeks, then one dose of cyclophosphamide 2.5 g/m2 with filgrastim 10 mcg/kg/day until stem cell collection. Of these, 18 patients with successful stem cell collection (at least 2 × 106 CD34 cells/kg) proceeded to RIT. A dosimetric administration of 111In ibritumomab tiuxetan (185 MBq) followed by five sequential quantitative planar and one SPECT/CT scan was used to determine predicted organ radiation-absorbed dose. Two weeks later, 90Y ibritumomab tiuxetan was administered in an outpatient setting at a cohort- and patient-specific predicted organ radiation-absorbed dose guided by a Continuous Response Assessment (CRM) methodology with the following cohorts for dose escalation: 14.8 MBq/kg, and targeted 18, 24, 28, and 30.5 Gy to the liver. Autologous stem cell infusion occurred when the estimated marrow radiation-absorbed dose rate was predicted to be <1 cGy/h. Feasibility, short-term toxicities, and tumor response were assessed. RESULTS: Patient-specific hybrid SPECT/CT + planar organ dosimetry was feasible in all 18 cases and used to determine the patient-specific therapeutic dose and guide dose escalation (26.8 ± 7.3 MBq/kg (mean), 26.3 MBq/kg (median) of 90Y (range: 12.1-41.4 MBq/kg)) of ibritumomab tiuxetan that was required to deliver 10 Gy to the liver. Infused stem cells engrafted rapidly. The most common treatment-related toxicities were hematological and were reversible following stem cell infusion. No significant hepatotoxicity was seen. One patient died from probable treatment-related causes-pneumonia at day 27 post-transplant. One patient at dose level 18 Gy developed myelodysplastic syndrome (MDS), 4 patients required admission post-90Y RIT for febrile neutropenia, 16/18 patients receiving 90Y ibritumomab tiuxetan (89%) responded to the therapy, with 13 CR (72%) and 3/18 PR (17%), at 60 days post-treatment. Two patients had progressive disease at sixty days. One patient was lost to follow-up. Median time to progression was estimated to be at least 13 months. MTD to the liver is greater than 28 Gy, but the MTD was not reached as the study was terminated due to unexpected discontinuation of availability of the therapeutic agent. CONCLUSIONS: Patient-specific outpatient 90Y ibritumomab tiuxetan RIT with myeloablative doses of RIT up to a targeted 30.5 Gy to the liver is feasible, guided by prospective SPECT/CT + planar imaging with the theranostic pair of 111In and 90Y anti-CD20, with outpatient autologous stem cell transplant support. Administered activity over 5 times the standard FDA-approved activity was well-tolerated. The non-hematopoietic MTD in this study exceeds 28 Gy to the liver. Initial tumor responses were common at all dose levels. This study supports the feasibility of organ dosimetry-driven patient-specific dose escalation in the treatment of NHL with stem cell transplant and provides additional information on the radiation tolerance of the normal liver to radiopharmaceutical therapy.

Renal 99mTc-DMSA pharmacokinetics in pediatric patients.

99mTc-DMSA is one of the most commonly used pediatric nuclear medicine imaging agents. Nevertheless, there are no pharmacokinetic (PK) models for 99mTc-DMSA in children, and currently available pediatric dose estimates for 99mTc-DMSA use pediatric S values with PK data derived from adults. Furthermore, the adult PK data were collected in the mid-70's using quantification techniques and instrumentation available at the time. Using pediatric imaging data for DMSA, we have obtained kinetic parameters for DMSA that differ from those applicable to adults. METHODS: We obtained patient data from a retrospective re-evaluation of clinically collected pediatric SPECT images of 99mTc-DMSA in 54 pediatric patients from Boston's Children Hospital (BCH), ranging in age from 1 to 16 years old. These were supplemented by prospective data from twenty-three pediatric patients (age range: 4 months to 6 years old). RESULTS: In pediatric patients, the plateau phase in fractional kidney uptake occurs at a fractional uptake value closer to 0.3 than the value of 0.5 reported by the International Commission on Radiological Protection (ICRP) for adult patients. This leads to a 27% lower time-integrated activity coefficient in pediatric patients than in adults. Over the age range examined, no age dependency in uptake fraction at the clinical imaging time was observed. Female pediatric patients had a 17% higher fractional kidney uptake at the clinical imaging time than males (P < 0.001). CONCLUSIONS: Pediatric 99mTc-DMSA kinetics differ from those reported for adults and should be considered in pediatric patient dosimetry. Alternatively, the differences obtained in this study could reflect improved quantification methods and the need to re-examine DMSA kinetics in adults.

Overcoming Barriers to Radiopharmaceutical Therapy (RPT): An Overview From the NRG-NCI Working Group on Dosimetry of Radiopharmaceutical Therapy.

Radiopharmaceutical therapy (RPT) continues to demonstrate tremendous potential in improving the therapeutic gains in radiation therapy by specifically delivering radiation to tumors that can be well assessed in terms of dosimetry and imaging. Dosimetry in external beam radiation therapy is standard practice. This is not the case, however, in RPT. This NRG (acronym formed from the first letter of the 3 original groups: National Surgical Adjuvant Breast and Bowel Project, the Radiation Therapy Oncology Group, and the Gynecologic Oncology Group)-National Cancer Institute Working Group review describes some of the challenges to improving RPT. The main priorities for advancing the field include (1) developing and adopting best practice guidelines for incorporating patient-specific dosimetry for RPT that can be used at both large clinics with substantial resources and more modest clinics that have limited resources, (2) establishing and improving strategies for introducing new radiopharmaceuticals for clinical investigation, (3) developing approaches to address the radiophobia that is associated with the administration of radioactivity for cancer therapy, and (4) solving the financial and logistical issues of expertise and training in the developing field of RPT.

Effects of shortened acquisition time on accuracy and precision of quantitative estimates of organ activity.

PURPOSE: Quantitative estimation of in vivo organ uptake is an essential part of treatment planning for targeted radionuclide therapy. This usually involves the use of planar or SPECT scans with acquisition times chosen based more on image quality considerations rather than the minimum needed for precise quantification. In previous simulation studies at clinical count levels (185 MBq 111In), the authors observed larger variations in accuracy of organ activity estimates resulting from anatomical and uptake differences than statistical noise. This suggests that it is possible to reduce the acquisition time without substantially increasing the variation in accuracy. METHODS: To test this hypothesis, the authors compared the accuracy and variation in accuracy of organ activity estimates obtained from planar and SPECT scans at various count levels. A simulated phantom population with realistic variations in anatomy and biodistribution was used to model variability in a patient population. Planar and SPECT projections were simulated using previously validated Monte Carlo simulation tools. The authors simulated the projections at count levels approximately corresponding to 1.5-30 min of total acquisition time. The projections were processed using previously described quantitative SPECT (QSPECT) and planar (QPlanar) methods. The QSPECT method was based on the OS-EM algorithm with compensations for attenuation, scatter, and collimator-detector response. The QPlanar method is based on the ML-EM algorithm using the same model-based compensation for all the image degrading effects as the QSPECT method. The volumes of interests (VOIs) were defined based on the true organ configuration in the phantoms. The errors in organ activity estimates from different count levels and processing methods were compared in terms of mean and standard deviation over the simulated phantom population. RESULTS: There was little degradation in quantitative reliability when the acquisition time was reduced by half for the QSPECT method (the mean error changed by < 1%, e.g., 0.9%-0.3% = 0.6% for the spleen). The magnitude of the errors and variations in errors for large organ with high uptake were still acceptable for 1.5 min scans, even though the errors were slightly larger than those for the 30 min scans (i.e., < 2% for liver, < 3% for heart). The errors over the ranges of scan times studied for the QPlanar method were all within 0.3% for all organs. CONCLUSIONS: These data indicate that, for the purposes of organ activity estimation, acquisition times could be reduced at least by a factor of 2 for the QSPECT and QPlanar methods with little effect on the errors in organ activity estimates. The acquisition time can be further reduced for the QPlanar method, assuming well-registered VOIs are available and the activity distribution in organs can be treated as uniform. Although the differences in accuracy and precision were statistically significant for all the acquisition times shorter than 30 min, the magnitude of the changes in accuracy and precision were small and likely not clinically important. The reduction in SPECT acquisition time justified by this study makes the use of SPECT a more clinically practical alternative to conventional planar scanning for targeted radiotherapy treatment planning.

An analytical model of the effects of pulse pileup on the energy spectrum recorded by energy resolved photon counting x-ray detectors.

PURPOSE: Recently, novel CdTe photon counting x-ray detectors (PCXDs) with energy discrimination capabilities have been developed. When such detectors are operated under a high x-ray flux, however, coincident pulses distort the recorded energy spectrum. These distortions are called pulse pileup effects. It is essential to compensate for these effects on the recorded energy spectrum in order to take full advantage of spectral information PCXDs provide. Such compensation can be achieved by incorporating a pileup model into the image reconstruction process for computed tomography, that is, as a part of the forward imaging process, and iteratively estimating either the imaged object or the line integrals using, e.g., a maximum likelihood approach. The aim of this study was to develop a new analytical pulse pileup model for both peak and tail pileup effects for nonparalyzable detectors. METHODS: The model takes into account the following factors: The bipolar shape of the pulse, the distribution function of time intervals between random events, and the input probability density function of photon energies. The authors used Monte Carlo simulations to evaluate the model. RESULTS: The recorded spectra estimated by the model were in an excellent agreement with those obtained by Monte Carlo simulations for various levels of pulse pileup effects. The coefficients of variation (i.e., the root mean square difference divided by the mean of measurements) were 5.3%-10.0% for deadtime losses of 1%-50% with a polychromatic incident x-ray spectrum. CONCLUSIONS: The proposed pulse pileup model can predict recorded spectrum with relatively good accuracy.