Search

PET imaging with targeted novel tracers has been commonly used in the clinical management of prostate cancer. The use of artificial intelligence (AI) in PET imaging is a relatively new approach and in this review article, we will review the current trends and categorize the currently available research into the quantification of tumor burden within the organ, evaluation of metastatic disease, and translational/supplemental research which aims to improve other AI research efforts.

Radiomics has undergone considerable development in recent years. In PET imaging, very promising results concerning the ability of handcrafted features to predict the biological characteristics of lesions and to assess patient prognosis or response to treatment have been reported in the literature. This article presents a checklist for designing a reliable radiomic study, gives an overview of the steps of the pipeline, and outlines approaches for data harmonization. Tips are provided for critical reading of the content of articles. The advantages and limitations of handcrafted radiomics compared with deep-learning approaches for the characterization of PET images are also discussed.

Application of kinetic modeling (KM) on a voxel level in dynamic PET images frequently suffers from high levels of noise, drastically reducing the precision of parametric image analysis. In this paper, we investigate the use of machine learning and artificial neural networks to denoise dynamic PET images. We train a deep denoising autoencoder (DAE) using noisy and noise-free spatiotemporal image patches, extracted from the simulated images of [11C]raclopride, a dopamine D2 receptor agonist. The DAE-processed dynamic and corresponding parametric images (simulated and acquired) are compared with those obtained with conventional denoising techniques, including temporal and spatial Gaussian smoothing, iterative spatiotemporal smoothing/deconvolution, and the highly constrained backprojection processing (HYPR). The simulated (acquired) parametric image non-uniformity was 7.75% (19.49%) with temporal and 5.90% (14.50%) with spatial smoothing, 5.82% (16.21%) with smoothing/deconvolution, 5.49% (13.38%) with HYPR, and 3.52% (11.41%) with DAE. The DAE also produced the best results in terms of the coefficient of variation of voxel values and structural similarity index. Denoising-induced bias in the regional mean binding potential was 7.8% with temporal and 26.31% with spatial smoothing, 28.61% with smoothing/deconvolution, 27.63% with HYPR, and 14.8% with DAE. When the test data did not match the training data, erroneous outcomes were obtained. Our results demonstrate that a deep DAE can provide a substantial reduction in the voxel-level noise compared with the conventional spatiotemporal denoising methods while introducing a similar or lower amount of bias. The better DAE performance comes at the cost of lower generality and requiring appropriate training data.

Purpose It is vital to appropriately power clinical trials towards discovery of novel disease-modifying therapies for Parkinson’s disease (PD). Thus, it is critical to improve prediction of outcome in PD patients. Methods We systematically probed a range of robust predictor algorithms, aiming to find best combinations of features for significantly improved prediction of motor outcome (MDS-UPDRS-III) in PD. We analyzed 204 PD patients with 18 features (clinical measures; dopamine-transporter (DAT) SPECT imaging measures), performing different randomized arrangements and utilizing data from 64%/6%/30% of patients in each arrangement for training/training validation/final testing. We pursued 3 approaches: i) 10 predictor algorithms (accompanied with automated machine learning hyperparameter tuning) were first applied on 32 experimentally created combinations of 18 features, ii) we utilized Feature Subset Selector Algorithms (FSSAs) for more systematic initial feature selection, and iii) considered all possible combinations between 18 features (262,143 states) to assess contributions of individual features. Results A specific set (set 18) applied to the LOLIMOT (Local Linear Model Trees) predictor machine resulted in the lowest absolute error 4.32 ± 0.19, when we firstly experimentally created 32 combinations of 18 features. Subsequently, 2 FSSAs (Genetic Algorithm (GA) and Ant Colony Optimization (ACO)) selecting 5 features, combined with LOLIMOT, reached an error of 4.15 ± 0.46. Our final analysis indicated that longitudinal motor measures (MDS-UPDRS-III years 0 and 1) were highly significant predictors of motor outcome. Conclusions We demonstrate excellent prediction of motor outcome in PD patients by employing automated hyperparameter tuning and optimal utilization of FSSAs and predictor algorithms.

AI for good (AI4G) projects involve developing and applying artificial intelligence (AI) based solutions to further goals in areas such as sustainability, health, humanitarian aid, and social justice. Developing and deploying such solutions must be done in collaboration with partners who are experts in the domain in question and who already have experience in making progress towards such goals. Based on our experiences, we detail the different aspects of this type of collaboration broken down into four high-level categories: communication, data, modeling, and impact, and distill eleven takeaways to guide such projects in the future. We briefly describe two case studies to illustrate how some of these takeaways were applied in practice during our past collaborations., Accepted to AIES-2021

Recent advances in the understanding of brain function are opening new frontiers in the investigation of movement disorders and neurodegeneration. The importance of the brain network-like characteristics is rapidly emerging together with increasing evidence that brain diseases imprint specific alterations on such networks. There is a strong need to determine molecular correlates associated with the network-type alterations to enable understanding of disease origin and mapping between clinical disease manifestations, genetic predispositions, and disease-triggering mechanisms. These considerations justify and highlight the importance of recent technological developments in positron emission tomography (PET) and integration of PET and magnetic resonance imaging (MRI), where the high neurochemical sensitivity of PET is complemented by MRI-derived measures of structural and functional connectivity. Ongoing developments of PET tracers suitable to image novel molecular targets and improvements in image reconstruction and analysis methods are further enhancing the relevance of imaging in addressing the complexity of brain function and disease-induced multidimensional alterations. This paper describes a conceptual justifications for the synergy between PET and MRI as related to neurodegeneration and movement disorders, discusses some predominantly PET-related developments relevant to and catalyzed by such synergy, and describes some novel multimodal metrics relevant to fundamental aspects of brain function altered early by disease.

Quantitative analysis of dopamine transporter (DAT) single-photon emission computed tomography (SPECT) images can enhance diagnostic confidence and improve their potential as a biomarker to monitor the progression of Parkinson’s disease (PD). In the present work, we aim to predict motor outcome from baseline DAT SPECT imaging radiomic features and clinical measures using machine learning techniques. We designed and trained artificial neural networks (ANNs) to analyze the data from 69 patients within the Parkinson’s Progressive Marker Initiative (PPMI) database. The task was to predict the unified PD rating scale (UPDRS) part III motor score in year 4 from 92 imaging features extracted on 12 different regions as well as 6 non-imaging measures at baseline (year 0). We first performed univariate screening (including the adjustment for false discovery) to select 4 regions each having 10 features with significant performance in classifying year 4 motor outcome into two classes of patients (divided by the UPDRS III threshold of 30). The leave-one-out strategy was then applied to train and test the ANNs for individual and combinations of features. The prediction statistics were calculated from 100 rounds of experiments, and the accuracy in appropriate prediction (classification of year 4 outcome) was quantified. Out of the baseline non-imaging features, only the UPDRS III (at year 0) was predictive, while multiple imaging features depicted significance. The different selected features reached a predictive accuracy of 70 % if used individually. Combining the top imaging features from the selected regions significantly improved the prediction accuracy to 75 % (p

Radiomic positron emission tomography (PET) image features are increasingly used in conjunction with machine learning to predict clinical disease measures. However, a thorough understanding of these image features remains challenging due to their relatively high complexity, hampering a-priori selection of optimal features and model parameters for a predictive task. In this paper, we explore the use of a generative disease model (GDM) for feature analysis. The GDM generates a series of synthetic PET images that simulate progressive disease-induced changes in radiotracer binding. These images can be used to obtain the expected values of image features, estimate the effect of various parameters on the feature correlation with clinical measures, and to select optimal features prior to testing them on real data. As an illustrative example, we apply the GDM-based approach to brain PET imaging of Parkinson’s disease subjects. Following initial validation, we use the GDM to understand the trends of change in the measured feature values with disease progression. Interestingly, the GDM revealed many features to change nonmonotonically, even with monotonic changes in radiotracer binding. An important implication of this finding is that different features may be optimal as biomarkers at different disease stages.

Purpose: Reconstructed PET images are typically noisy, especially in dynamic imaging where the acquired data are divided into several short temporal frames. High noise in the reconstructed images translates to poor precision/reproducibility of image features. One important role of "denoising" is therefore to improve the precision of image features. However, typical denoising methods achieve noise reduction at the expense of accuracy. In this work, we present a novel four-dimensional (4D) denoised image reconstruction framework, which we validate using 4D simulations, experimental phantom, and clinical patient data, to achieve 4D noise reduction while preserving spatiotemporal patterns/minimizing error introduced by denoising.Methods: Our proposed 4D denoising operator/kernel is based on HighlY constrained backPRojection (HYPR), which is applied either after each update of OSEM reconstruction of dynamic 4D PET data or within the recently proposed kernelized reconstruction framework inspired by kernel methods in machine learning. Our HYPR4D kernel makes use of the spatiotemporal high frequency features extracted from a 4D composite, generated within the reconstruction, to preserve the spatiotemporal patterns and constrain the 4D noise increment of the image estimate.Results: Results from simulations, experimental phantom, and patient data showed that the HYPR4D kernel with our proposed 4D composite outperformed other denoising methods, such as the standard OSEM with spatial filter, OSEM with 4D filter, and HYPR kernel method with the conventional 3D composite in conjunction with recently proposed High Temporal Resolution kernel (HYPRC3D-HTR), in terms of 4D noise reduction while preserving the spatiotemporal patterns or 4D resolution within the 4D image estimate. Consequently, the error in outcome measures obtained from the HYPR4D method was less dependent on the region size, contrast, and uniformity/functional patterns within the target structures compared to the other methods. For outcome measures that depend on spatiotemporal tracer uptake patterns such as the nondisplaceable Binding Potential (BPND), the root mean squared error in regional mean of voxel BPND values was reduced from similar to 8% (OSEM with spatial or 4D filter) to similar to 3% using HYPRC3D-HTR and was further reduced to similar to 2% using our proposed HYPR4D method for relatively small target structures (similar to 10 mm in diameter). At the voxel level, HYPR4D produced two to four times lower mean absolute error in BPND relative to HYPRC3D-HTR.Conclusion: As compared to conventional methods, our proposed HYPR4D method can produce more robust and accurate image features without requiring any prior information.

Purpose Respiratory motion during positron emission tomography (PET) scans can be a major detriment to image quality in oncological imaging, leading to loss of quantification accuracy and false negative findings. The impact of motion on lesion quantification and detectability can be assessed using anthropomorphic phantoms with realistic anatomy representation and motion modelling. In this work we design and build such a phantom, with careful consideration of system requirements and detailed force analysis.Methods: We start from a previously-developed anatomically-accurate shell of a human torso and add elastic lungs with a highly controllable actuation mechanism which replicates the physics of breathing. The space outside the lungs is filled with a radioactive water solution. To maintain anatomical accuracy in the torso and realistic gamma ray attenuation, all motion mechanisms and actuators are positioned outside of the phantom compartment. The actuation mechanism can produce a plethora of custom respiratory waveforms with breathing rates up to 25 breaths per minute and tidal volumes up to 1200mL.Results: Several tests were performed to validate the performance of the phantom assembly, in which the phantom was filled with water and given respiratory waveforms to execute. All parts demonstrated nominal performance. Force requirements were not exceeded and no leaks were detected, although continued use of the phantom is required to evaluate wear. The respiratory motion was determined to be within a reasonable realistic range. Conclusions: The full mechanical design is described in this paper, as well as a software application with graphical user interface which was developed to plan and visualize respiratory patterns. Both are available open source and linked in this paper. The developed phantom will facilitate future work in evaluating the impact of respiratory motion on lesion quantification and detectability.

Background and ObjectiveRadiomics and deep learning have emerged as two distinct approaches to medical image analysis. However, their relative expressive power remains largely unknown. Theoretically, hand-crafted radiomic features represent a mere subset of features that neural networks can approximate, thus making deep learning a more powerful approach. On the other hand, automated learning of hand-crafted features may require a prohibitively large number of training samples. Here we directly test the ability of convolutional neural networks (CNNs) to learn and predict the intensity, shape, and texture properties of tumors as defined by standardized radiomic features.MethodsConventional 2D and 3D CNN architectures with an increasing number of convolutional layers were trained to predict the values of 16 standardized radiomic features from real and synthetic PET images of tumors, and tested. In addition, several ImageNet-pretrained advanced networks were tested. A total of 4000 images were used for training, 500 for validation, and 500 for testing.ResultsFeatures quantifying size and intensity were predicted with high accuracy, while shape irregularity and heterogeneity features had very high prediction errors and generalized poorly. For example, mean normalized prediction error of tumor diameter with a 5-layer CNN was 4.23 ± 0.25, while the error for tumor sphericity was 15.64 ± 0.93. We additionally found that learning shape features required an order of magnitude more samples compared to intensity and size features.ConclusionsOur findings imply that CNNs trained to perform various image-based clinical tasks may generally under-utilize the shape and texture information that is more easily captured by radiomics. We speculate that to improve the CNN performance, shape and texture features can be computed explicitly and added as auxiliary variables to the networks, or supplied as synthetic inputs.

Quantitative positron emission tomography imaging often requires correcting the image data for deformable motion. With cyclic motion, this is traditionally achieved by separating the coincidence data into a relatively small number of gates, and incorporating the inter-gate image transformation matrices into the reconstruction algorithm. In the presence of non-cyclic deformable motion, this approach may be impractical due to a large number of required gates. In this paper, we propose an alternative approach to iterative image reconstruction with correction for deformable motion, wherein unorganized point clouds are used to model the imaged objects in the image space, and motion is corrected for explicitly by introducing a time-dependence into the point coordinates. The image function is represented using constant basis functions with finite support determined by the boundaries of the Voronoi cells in the point cloud. We validate the quantitative accuracy and stability of the proposed approach by reconstructing noise-free and noisy projection data from digital and physical phantoms. The point-cloud-based maximum likelihood expectation maximization (MLEM) and one-pass list-mode ordered-subset expectation maximization (OSEM) algorithms are validated. The results demonstrate that images reconstructed using the proposed method are quantitatively stable, with noise and convergence properties comparable to image reconstruction based on the use of rectangular and radially-symmetric basis functions.

Measurement of stimulus-induced dopamine release and other types of transient neurotransmitter response (TNR) from dynamic positron emission tomography (PET) images typically suffers from limited detection sensitivity and high false positive (FP) rates. Measurement of TNR of a voxel-level can be particularly problematic due to high image noise. In this work, we perform voxel-level TNR detection using artificial neural networks (ANN) and compare their performance to previously used standard statistical tests. Different ANN architectures were trained and tested using simulated and real human PET imaging data, obtained with the tracer [11C]raclopride (a D2 receptor antagonist). A distinguishing feature of our approach is the use of ‘personalized’ ANNs that are designed to operate on the image from a specific subject and scan. Training of personalized ANNs was performed using simulated images that have been matched with the acquired image in terms of the signal, resolution, and noise. In our tests of TNR detection performance, the F-test of the linear parametric neurotransmitter PET model fit residuals was used as the reference method. For a moderate TNR magnitude, the areas under the receiver operating characteristic curves in simulated tests were 0.64 for the F-test and 0.77–0.79 for the best ANNs. At a fixed FP rate of 0.01, the true positive rates were 0.6 for the F-test and 0.8–0.9 for the ANNs. The F-test detected on average 28% of a 8.4 mm cluster with a strong TNR, while the best ANN detected 47%. When applied to a real image, no significant abnormalities in the ANN outputs were observed. These results demonstrate that personalized ANNs may offer a greater detection sensitivity of dopamine release and other types of TNR compared to previously used method based on the F-test.

There are currently no established disease modifying therapies for PD, and prediction of outcome in PD to power clinical studies is a very important area of research. Assessment of PD is informed by imaging the dopamine system with dopamine transporter (DAT) single-photon emission computed tomography (SPECT) imaging and by the presence of key symptoms. Recently, deep-learning based methods have shown promise for medical image analysis tasks and disease detection. The purpose of this study was to develop a deep-learning based approach to predict outcome of patients with PD using longitudinal clinical data containing imaging and non-imaging information. Features were first extracted from the clinical data by the proposed deep-learning based approach and then combined to predict motor performance (MDS-UPDRS-III) in year 4. The performance of the proposed approach was evaluated via a 10-fold cross-validation. We evaluated the performance of the network on the basis of mean absolute error (MAE) between the predicted and true MDS-UPDRS part III scores in year 4. The proposed approach yielded a MAE of 4.33±3.36 when given only imaging features, 3.71±2.91 when given only non-imaging features, and 3.22±2.71 when given all input data. While the approach given only non-imaging input data outperformed the approach given only imaging data, we found that the performance of the proposed approach substantially improved when given both imaging and non-imaging information. Our results indicate that the addition of imaging data to non-imaging clinical data is helpful for the prediction of outcome in patients with PD. The proposed approach that incorporated both imaging and non-imaging clinical data shows significant promise for prediction of outcome in patients with PD.

In the present work, we systematically probe a range of predictor machines (11 machines), and aim to find the best combinations of features to result in improvements in prediction of outcome in PD. First, we created 32 combinations of 18 conventional features experimentally and selected 4 arrangements of 204 PD subjects. The combinations were applied to the various predictor machines), thereby absolute error of the best combination reached 4.3 (in prediction of MDS-UPDRS-III motor performance in year 4). This is in comparison to previous works that attained errors of around 9. In second part, subset selector machines were used for selecting the best combinations between all features, and GA and ACO selector machines selected the best combinations, further lowering error when combined with LOLIMOT for prediction. Selected features by GA and ACO (UPDRS I-Year 1, UPDRS III-Year 1, left putamen Uptake-Year 1, Age, Gender) had positive effect on prediction of outcome and mean absolute error reached 4.15. Moreover, other subset selector machines also reached acceptable results; mean absolute errors in some predictor machines were below 4.7. Overall, LOLIMOT was seen as the best predictor machine, and GA and ACO as the best feature subset selector machines. Furthermore, MDS-UPDRS III-years 0 and 1, MDS-UPDRS II- year 1, MDS-UPDRS I- years 0 and 1, age and DAT SPECT putamen as well as caudate uptake -year 1 were seen as most important predictors of outcome.

Positron emission tomography (PET) data related to neurodegeneration are most often quantified using methods based on tracer kinetic modeling. In contrast, here we investigate the ability of geometry and texture-based metrics that are independent of kinetic modeling to convey useful information on disease state. The study was performed using data from Parkinson’s disease subjects imaged with 11C-dihydrotetrabenazine and 11C-raclopride. The pattern of the radiotracer distribution in the striatum was quantified using image-based metrics evaluated over multiple regions of interest that were defined on co-registered PET and MRI images. Regression analysis showed a significant degree of correlation between several investigated metrics and clinical evaluations of the disease ( p 2 = 0.94). These results demonstrate that there is clinically relevant quantitative information in the tracer distribution pattern that can be captured using geometric and texture descriptors. Such metrics may provide an alternate and complementary data analysis approach to traditional kinetic modeling.

Background Given the increasing recognition of the significance of non-motor symptoms in Parkinson's disease, we investigate the optimal use of machine learning methods for the prediction of the Montreal Cognitive Assessment (MoCA) score at year 4 from longitudinal data obtained at years 0 and 1. Methods We selected n = 184 PD subjects from the Parkinson's Progressive Marker Initiative (PPMI) database (93 features). A range of robust predictor algorithms (accompanied with automated machine learning hyperparameter tuning) and feature subset selector algorithms (FSSAs) were selected. We utilized 65%, 5% and 30% of patients in each arrangement for training, training validation and final testing respectively (10 randomized arrangements). For further testing, we enrolled 308 additional patients. Results First, we employed 10 predictor algorithms, provided with all 93 features; an error of 1.83 ± 0.13 was obtained by LASSOLAR (Least Absolute Shrinkage and Selection Operator - Least Angle Regression). Subsequently, we used feature subset selection followed by predictor algorithms. GA (Genetic Algorithm) selected 18 features; subsequently LOLIMOT (Local Linear Model Trees) reached an error of 1.70 ± 0.10. DE (Differential evolution) also selected 18 features and coupled with Thiel-Sen regression arrived at a similar performance. NSGAII (Non-dominated sorting genetic algorithm) yielded the best performance: it selected six vital features, which combined with LOLIMOT reached an error of 1.68 ± 0.12. Finally, using this last approach on independent test data, we reached an error of 1.65. Conclusion By employing appropriate optimization tools (including automated hyperparameter tuning), it is possible to improve prediction of cognitive outcome. Overall, we conclude that optimal utilization of FSSAs and predictor algorithms can produce very good prediction of cognitive outcome in PD patients.

Spatial patterns of radiotracer binding in positron emission tomography (PET) images may convey information related to the disease topology. However, this information is not captured by the standard PET image analysis that quantifies the mean radiotracer uptake within a region of interest (ROI). On the other hand, spatial analyses that use more advanced radiomic features may be difficult to interpret. Here we propose an alternative data-driven, voxel-based approach to spatial pattern analysis in brain PET, which can be easily interpreted. We apply principal component analysis (PCA) to identify voxel covariance patterns, and optimally combine several patterns using the least absolute shrinkage and selection operator (LASSO). The resulting models predict clinical disease metrics from raw voxel values, allowing for inclusion of clinical covariates. The analysis is performed on high-resolution PET images from healthy controls and subjects affected by Parkinson's disease (PD), acquired with a pre-synaptic and a post-synaptic dopaminergic PET tracer. We demonstrate that PCA identifies robust and tracer-specific binding patterns in sub-cortical brain structures; the patterns evolve as a function of disease progression. Principal component LASSO (PC-LASSO) models of clinical disease metrics achieve higher predictive accuracy compared to the mean tracer binding ratio (BR) alone: the cross-validated test mean squared error of adjusted disease duration (motor impairment score) was 16.3 ± 0.17 years2 (9.7 ± 0.15) with mean BR, versus 14.4 ± 0.18 years2 (8.9 ± 0.16) with PC-LASSO. We interpret the best-performing PC-LASSO models in the spatial sense and discuss them with reference to the PD pathology and somatotopic organization of the striatum. PC-LASSO is thus shown to be a useful method to analyze clinically-relevant tracer binding patterns, and to construct interpretable, imaging-based predictive models of clinical metrics.

We describe a method to generate a digital phantom of an unrestrained rodent that incorporates non-periodic motion with deformation. The phantom is represented by a deformable point cloud with time-dependent point coordinates, activity and attenuation. Motion is simulated by applying time-varying mesh deformation operators to the point cloud. To compare the simulated motion to the motion of a live rodent, the behavior of an unrestrained mouse was recorded using a depth camera, and the kinematic parameters of motion were measured. To generate simulated coincidence data, the phantom is voxelized and used in the Monte-Carlo emission simulation. The combined emission and motion data can be used for the development and validation of the image reconstruction algorithms with deformable motion correction.

We investigate the correlation between the clinical severity of neurodegenerative disease and texture metrics (such as Haralick features) computed using PET images of the brain. Specifically, we explore how the parameters of feature computation — such as the region of interest definition method, and the direction and distance used for texture quantification — affect the correlation between texture-based image metrics and clinical disease severity. The analysis was based on an ongoing Parkinson's disease imaging study, with co-registered PET and MRI images, and tracer predominantly concentrated in the striatum. Disease duration was used as the primary clinical metric. It was found that the region of interest placement method substantially affected the correlation values. Significant correlation (p

Texture and shape analysis applied to positron emission tomography (PET) or single photon emission computed tomography (SPECT) imaging is a technique based on the characterization of the spatial distribution of a radio tracer using texture and shape descriptors. It has been shown in recent studies to provide functional disease related information. Applied to high resolution PET images of patients suffering from Parkinson's disease (PD), a good correlation has been found between 3D moment invariants (3DMI) which are shape metrics and Parkinson's disease severity. Given the wide availability of SPECT cameras in clinical environments, could texture and shape analysis provide comparable results on lower resolution images to those obtained with state-of-the-art PET cameras? The aim of the present study is to investigate the applicability and robustness of the texture and shape analysis in the specific context of images displaying localized spatial distribution of the tracer with disease induced spatial abnormalities. Applicability and robustness of the method was tested against: i) the choice of the texture and shape descriptors, ii) the image spatial resolution, iii) the image noise level and iv) the definition of the region of interest (ROI). Methods: a magnetic resonance imaging (MRI) scan to provide anatomical information for ROIs placement and a high resolution PET scan providing a dynamic sequence of [11C]dihydrotetrabenazine (DTBZ) images were acquired for 13 PD patients and 6 healthy controls. To simulate the lower spatial resolutions, the reconstructed PET images were smoothed with a 3D Gaussian filter with a full width at half maximum (FWHM) ranging from 2 mm to 20 mm. As an evaluation criterion, Spearman's correlation was calculated between texture and shape metrics and disease severity assessed either by the unified Parkinson's disease rating scale (UPDRS) scores or by the disease duration. Results and conclusion: this study has shown that texture and shape analysis can provide relevant disease related information when applied to images of tracers displaying a localized spatial distribution with disease dependent heterogeneity and/or shape characteristics. In the specific case of Parkinson's disease imaged with the PET tracer [11C]DTBZ, we have shown that the MEAN intensity metric and the 3D moment invariant metrics are strongly correlated with disease severity and that the strength of the correlation persisted on images ranging from the highest spatial resolutions achievable by state-of-the-art PET cameras to the lowest resolutions achievable by any modern clinical SPECT camera.

Image reconstruction techniques that use rectangular basis functions (pixels and voxels) may not be optimal when non-periodic, deformable motion correction is required. Here we propose a new approach to PET image reconstruction and non-rigid motion correction that is based on representing the imaged objects with regularized, spatially bounded sets of disconnected points. Motion correction is performed by explicitly incorporating the object motion into the reconstruction algorithm, though the dynamically adjusted coordinates of the points. Within the proposed approach, the images are reconstructed iteratively in list-mode, and the system matrix calculation is based on the localized estimation of the probabilistic weights for every point in the generated point set, using an optimized point search algorithm. To validate the motion correction, a digital phantom of a freely moving mouse was generated using mesh deformation operators such as armatures and curve modifiers. From the simulated PET list-mode data and a priori known motion trajectory, we reconstructed 3D images corrected for deformable, non-periodic motion without using the traditional gate-based methods. In addition, the stability of the reconstructed images with respect to the point set parameters and deformations was investigated.

The objective of this work was to investigate if geometry and texture-based metrics contain brain disease-related information similar to measures typically derived from kinetic modeling- based approaches. Using co-registered PET and MRI images from an ongoing Parkinson's disease imaging study, the shape, size and texture of the striatal regions containing high tracer concentration were estimated, and regressed against the subject's disease duration. A novel inter-modality region fusion method was used for a systematic metric characterization and evaluation. It was established that several regional measures such as the region volume, surface area, moment invariants, and others were very good predictors of the clinical disease duration. Interestingly, some metrics revealed good correlation with the disease duration only when evaluated on the fused inter-modality PET-MRI regions. These results demonstrate that geometric features that do not rely on kinetic modeling may be used for disease characterization.

Different imaging modalities sample different properties of the tissue, and thus the tissue may appear different depending on the imaging technique. As a consequence, the shapes of organs and homogenous regions in tissues often have different shapes depending on the type of imaging. This presents a problem for ROI-based multi-modality quantitative imaging studies, since it is not clear what modality should be used for data segmentation. An example of such study is the quantitative PET imaging of Parkinson's disease subjects, which often present functional atrophy without an anatomical atrophy. A choice must be made between anatomical (MRI) and radioactivity-based (PET) ROIs. In addition manual ROI placement can be very time consuming and may lack consistency. In this work, we propose a new approach to multi-modality image segmentation. The proposed method generates so-called mixed ROIs that can be computed in a fully automated mode from single modality-based pure ROIs. The computation of the mixed ROIs is based on the fusion of probability images. The use of the fusion principles made it possible to transition between the pure ROI shapes in a smooth fashion. The mixed ROIs were found to be better aligned with the high activity regions than the pure MR ROIs, and had higher anatomical fidelity compared to the pure PET ROIs. Using the method, it is possible to generate a multitude of ROI sets for a particular study starting from one or more previously defined regions.

PET and SPECT images are traditionally reconstructed using voxel grids. In situations where non-rigid motion correction is required, meshes are believed to be better suited than voxels for image reconstruction. With meshes, deformations can be modeled explicitly by changing the coordinates of the mesh nodes. Previously proposed approaches to mesh-based image reconstruction relied either on computing the projection of mesh polyhedrons onto the detector planes, or computing the length-of-intersection between the polyhedrons and the lines-of-response.

Purpose: To evaluate PETimage reconstruction algorithm capable of handling moving, deformable objects, particularly unrestrained conscious rodents. Methods: The proposed point‐based image reconstruction algorithm is based on 2D filtered back‐projection. Reconstruction took place in 2D virtual space filled with dimensionless sampling points. During the backprojection, points acted as spatial count bins, accumulating values of the backprojected LORs. To correct for motion, sampling points in the image space were moved in accord with the scanned object. Thus, motion correction and deformation correction were performed via image bin repositioning. After the backprojection step, images of activity distributions were obtained by fitting a smooth surface over the sampling points. Results: To evaluate the performance of the image reconstruction algorithm, stationary and moving NEMA image quality phantoms were scanned on Siemens Focus 120 microPET scanner, and data was reconstructed using conventional 2D filtered back‐projection and point‐based algorithms. Images obtained using two reconstruction algorithms were in remarkable quantitative agreement, with deviations mostly in the background and on the order of few percent. The location of points during reconstruction had little influence as long as Nyquist sampling criteria was satisfied. For the point‐ based reconstruction, motion‐corrected images preserved quantification and demonstrated good performance in removing the motion artifact. The SNR of motion‐corrected images also improved compared with shorter but motion‐free frames. Conclusions: It had been demonstrated that point reconstruction can provide accurate, quantitative PETimages. As opposed to voxels, points can be moved freely relative to one another. By moving points during the backprojection process, almost any kind of deformation can be taken into account. Plans for the future include further method development of the method to enable full 3D list‐mode reconstruction, which could ultimately lead to unrestrained conscious rodent imaging.

Dopamine transporter (DAT) SPECT imaging is increasingly utilized for diagnostic purposes in suspected Parkinsonian syndromes. We performed a cross-sectional study to investigate whether assessment of texture in DAT SPECT radiotracer uptake enables enhanced correlations with severity of motor and cognitive symptoms in Parkinson's disease (PD), with the long-term goal of enabling clinical utility of DAT SPECT imaging, beyond standard diagnostic tasks, to tracking of progression in PD. Quantitative analysis in routine DAT SPECT imaging, if performed at all, has been restricted to assessment of mean regional uptake. We applied a framework wherein textural features were extracted from the images. Notably, the framework did not require registration to a common template, and worked in the subject-native space. Image analysis included registration of SPECT images onto corresponding MRI images, automatic region-of-interest (ROI) extraction on the MRI images, followed by computation of Haralick texture features. We analyzed 141 subjects from the Parkinson's Progressive Marker Initiative (PPMI) database, including 85 PD and 56 healthy controls (HC) (baseline scans with accompanying 3 T MRI images). We performed univariate and multivariate regression analyses between the quantitative metrics and different clinical measures, namely (i) the UPDRS (part III - motor) score, disease duration as measured from (ii) time of diagnosis (DD-diag.) and (iii) time of appearance of symptoms (DD-sympt.), as well as (iv) the Montreal Cognitive Assessment (MoCA) score. For conventional mean uptake analysis in the putamen, we showed significant correlations with clinical measures only when both HC and PD were included (Pearson correlation r = − 0.74, p-value, Graphical abstract Improved correlation of heterogeneity metrics (vs. conventional mean-uptake) against clinical assessments., Highlights • Aim to enable image-based tracking of progression in Parkinson's disease • Texture analysis of clinical dopamine transporter (DAT) SPECT images (DaTscans) • Significant correlations with clinical, motor and cognitive outcomes