Brain PET Imaging in Small Animals: Tracer Formulation, Data Acquisition, Image Reconstruction, and Data Analysis

Positron emission tomography (PET) is a noninvasive functional imaging modality that involves in vivo detection of spatiotemporal changes in the binding of radioactive pharmaceuticals (a.k.a. PET tracers) to their target sites in different organs. The development of new PET tracers commonly involves their preclinical evaluation in small rodents. Moreover, laboratory animal PET research is now being used with progressively greater frequency to complement human PET studies, to investigate in greater depth the underlying pathophysiology of human diseases, and to monitor the efficiency of novel therapeutic interventions. Here we describe the steps toward a successful small animal PET study, from tracer formulation and image acquisition to data reconstruction and analysis of the acquired scans, with a particular focus on its utility for the brain.

Enhanced PET imaging using progressive conditional deep image prior

Objective.Unsupervised learning-based methods have been proven to be an effective way to improve the image quality of positron emission tomography (PET) images when a large dataset is not available. However, when the gap between the input image and the target PET image is large, direct unsupervised learning can be challenging and easily lead to reduced lesion detectability. We aim to develop a new unsupervised learning method to improve lesion detectability in patient studies.Approach.We applied the deep progressive learning strategy to bridge the gap between the input image and the target image. The one-step unsupervised learning is decomposed into two unsupervised learning steps. The input image of the first network is an anatomical image and the input image of the second network is a PET image with a low noise level. The output of the first network is also used as the prior image to generate the target image of the second network by iterative reconstruction method.Results.The performance of the proposed method was evaluated through the phantom and patient studies and compared with non-deep learning, supervised learning and unsupervised learning methods. The results showed that the proposed method was superior to non-deep learning and unsupervised methods, and was comparable to the supervised method.Significance.A progressive unsupervised learning method was proposed, which can improve image noise performance and lesion detectability.

MRI-aided kernel PET image reconstruction method based on texture features

We investigate the reconstruction of low-count positron emission tomography (PET) projection, which is an important, but challenging, task. Using the texture feature extraction method of radiomics, i.e. the gray-level co-occurrence matrix (GLCM), texture features can be extracted from magnetic resonance imaging images with high-spatial resolution. In this work, we propose a kernel reconstruction method combining autocorrelation texture features derived from the GLCM. The new kernel function includes the correlations of both the intensity and texture features from the prior image. By regarding the GLCM as a discrete approximation of a probability density function, the asymptotically gray-level-invariant autocorrelation texture feature is generated, which can maintain the accuracy of texture features extracted from small image regions by reducing the number of quantized image gray levels. A computer simulation shows that the proposed method can effectively reduce the noise in the reconstructed image compared to the maximum likelihood expectation maximum method and improve the image quality and tumor region accuracy compared to the original kernel method for low-count PET reconstruction. A simulation study on clinical patient images also shows that the proposed method can improve the whole image quality and that the reconstruction of a high-uptake lesion is more accurate than that achieved by the original kernel method.

Reconstruction, analysis and interpretation of posterior probability distributions of PET images, using the posterior bootstrap

The uncertainty of reconstructed PET images remains difficult to assess and to interpret for the use in diagnostic and quantification tasks. Here we provide (1) an easy-to-use methodology for uncertainty assessment for almost any Bayesian model in PET reconstruction from single datasets and (2) a detailed analysis and interpretation of produced posterior image distributions. We apply a recent posterior bootstrap framework to the PET image reconstruction inverse problem and obtain simple parallelizable algorithms based on random weights and on existing maximuma posteriori(MAP) (posterior maximum) optimization-based algorithms. Posterior distributions are produced, analyzed and interpreted for several common Bayesian models. Their relationship with the distribution of the MAP image estimate over multiple dataset realizations is exposed. The coverage properties of posterior distributions are validated. More insight is obtained for the interpretation of posterior distributions in order to open the way for including uncertainty information into diagnostic and quantification tasks.

Image reconstruction methods affect software-aided assessment of pathologies of [18F]flutemetamol and [18F]FDG brain-PET examinations in patients with neurodegenerative diseases

•

[¹⁸F]Flutemetamol and [¹⁸F]FDG image reconstruction.

•

Software-aided assessment of neurodegenerative disease patients.

•

New developments in brain PET image reconstruction affect quantitative measures.

•

Evaluation of SUVR and z-score measures.

•

Normalizing to pons and whole brain induced greater absolute differences between reconstructions.

Purpose

To assess how some of the new developments in brain positron emission tomography (PET) image reconstruction affect quantitative measures and software-aided assessment of pathology in patients with neurodegenerative diseases.

Methods

PET data were grouped into four cohorts: prodromal Alzheimer’s disease patients and controls receiving [¹⁸F]flutemetamol, and neurodegenerative disease patients and controls receiving [¹⁸F]FDG PET scans. Reconstructed images were obtained by ordered-subsets expectation maximization (OSEM; 3 iterations (i), 16/34 subsets (s), 3/5-mm filter, ±time-of-flight (TOF), ±point-spread function (PSF)) and block-sequential regularized expectation maximization (BSREM; TOF, PSF, β-value 75–300). Standardized uptake value ratios (SUVR) and z-scores were calculated (CortexID Suite, GE Healthcare) using cerebellar gray matter, pons, whole cerebellum and whole brain as reference regions.

Results

In controls, comparable results to the normal database were obtained with OSEM 3i/16 s 5-mm reconstruction. TOF, PSF and BSREM either increased or decreased the relative uptake difference to the normal subjects’ database within the software, depending on the tracer and chosen reference area, i.e. resulting in increased absolute z-scores. Normalizing to pons and whole brain for [¹⁸F]flutemetamol and [¹⁸F]FDG, respectively, increased absolute differences between reconstructions methods compared to normalizing to cerebellar gray matter and whole cerebellum when applying TOF, PSF and BSREM.

Conclusions

Software-aided assessment of patient pathologies should be used with caution when employing other image reconstruction methods than those used for acquisition of the normal database.

Implementation of cylindrical PET scanners with block detector geometry in STIR

BACKGROUND

Software for Tomographic Image Reconstruction (STIR) is an open-source library for PET and SPECT image reconstruction, implementing iterative reconstruction as well as 2D- and 3D-filtered back projection. Quantitative reconstruction of PET data requires the knowledge of the scanner geometry. Typical scanners, clinical as well as pre-clinical ones, use a block-type geometry. Several rectangular blocks of crystals are arranged into regular polygons. Multiple of such polygons are arranged along the scanner axis. However, the geometrical representation of a scanner provided by STIR is a cylinder made of rings of individual crystals equally distributed in axial and transaxial directions. The data of realistic scanners are projected onto such virtual scanners prior to image reconstruction. This results in reduced quality of the reconstructed image. In this study, we implemented the above-described block geometry into the STIR library, permitting the image reconstruction without the interpolation step. In order to evaluate the difference in image quality, we performed Monte Carlo simulation studies of three different scanner designs: two scanners with multiple crystals per block and one with a single crystal per block. Simulated data were reconstructed using the standard STIR method and the newly implemented block geometry.

RESULTS

Visual comparison between the images reconstructed by the two models for the block-type scanners shows that the new implementation enhances the image quality to the extent that the results before normalization correction are comparable with those after normalization correction. The simulation result of a uniform cylinder shows that the coefficient of variation decreases from 25.8% to 20.9% by using the new implementation in STIR. Spatial resolution is enhanced resulting in a lower partial loss of intensity in sources of small size, e.g., the spill-over ratio for spherical sources of 1.8 mm diameter is 0.19 in the block and 0.34 in the cylindrical model.

CONCLUSIONS

Results indicate a significant improvement for the new model in comparison with the old one which mapped the polygonal geometry into a cylinder. The new implementation was tested and is available for use via the library of Swiss Federal Institute of Technology in Zurich (ETH).

A Monte Carlo detector response model for the IRIS PET preclinical scanner

PET preclinical studies require high spatial resolution due to the limited size of the animal under investigation. To achieve this target, iterative image reconstruction algorithms are commonly preferred over the analytical methods because they offer the possibility of accurately modeling the whole imaging process. In this work, we propose an accurate factorized system matrix for the INVISCAN IRIS preclinical PET scanner to be used with an iterative algorithm. The model includes two components: the geometric component and the detector response of the system. The main innovative aspect of the work is the creation of the detector matrix using a Monte Carlo simulation, with a particular focus on the optimization of the simulation process to reduce the calculation time. The new system model is compared with the current IRIS model to evaluate the image quality, following the NEMA Standards NU 4-2008. The comparison showed an enhancement of the image quality, in terms of uniformity and recovery coefficients. This work confirms that the inclusion of the detector response into the system model leads to improved reconstruction results.

Multi-Tracer Guided PET Image Reconstruction

Multi-tracer positron emission tomography (PET) has the potential to enhance PET imaging by providing complementary information from different physiological processes. However, one or more of the images may present high levels of noise. Guided image reconstruction methods transfer information from a guide image into the PET image reconstruction to encourage edge-preserving noise reduction. In this work we aim to reduce noise in poorer quality PET datasets via guidance from higher quality ones by using a weighted quadratic penalty approach. In particular, we applied this methodology to [18F]fluorodeoxyglucose (FDG) and [11C]methionine imaging of gliomas. 3D simulation studies showed that guiding the reconstruction of methionine datasets using pre-existing FDG images reduced reconstruction errors across the whole-brain (-8%) and within a tumour (-36%) compared to maximum likelihood expectation-maximisation (MLEM). Furthermore, guided reconstruction outperformed a comparable non-local means filter, indicating that regularising during reconstruction is preferable to post-reconstruction approaches. Hyperparameters selected from the 3D simulation study were applied to real data, where it was observed that the proposed FDG-guided methionine reconstruction allows for better edge preservation and noise reduction than standard MLEM. Overall, the results in this work demonstrate that transferring information between datasets in multi-tracer PET studies improves image quality and quantification performance.

Improving PET Quantification of Small Animal [68Ga]DOTA-Labeled PET/CT Studies by Using a CT-Based Positron Range Correction

PURPOSE

Image quality of positron emission tomography (PET) tracers that emits high-energy positrons, such as Ga-68, Rb-82, or I-124, is significantly affected by positron range (PR) effects. PR effects are especially important in small animal PET studies, since they can limit spatial resolution and quantitative accuracy of the images. Since generators accessibility has made Ga-68 tracers wide available, the aim of this study is to show how the quantitative results of [⁶⁸Ga]DOTA-labeled PET/X-ray computed tomography (CT) imaging of neuroendocrine tumors in mice can be improved using positron range correction (PRC).

PROCEDURES

Eighteen scans in 12 mice were evaluated, with three different models of tumors: PC12, AR42J, and meningiomas. In addition, three different [⁶⁸Ga]DOTA-labeled radiotracers were used to evaluate the PRC with different tracer distributions: [⁶⁸Ga]DOTANOC, [⁶⁸Ga]DOTATOC, and [⁶⁸Ga]DOTATATE. Two PRC methods were evaluated: a tissue-dependent (TD-PRC) and a tissue-dependent spatially-variant correction (TDSV-PRC). Taking a region in the liver as reference, the tissue-to-liver ratio values for tumor tissue (TLR_tumor), lung (TLR_lung), and necrotic areas within the tumors (TLR_necrotic) and their respective relative variations (ΔTLR) were evaluated.

RESULTS

All TLR values in the PRC images were significantly different (p < 0.05) than the ones from non-PRC images. The relative differences of the tumor TLR values, respect to the case with no PRC, were ΔTLR_tumor 87 ± 41 % (TD-PRC) and 85 ± 46 % (TDSV-PRC). TLR_lung decreased when applying PRC, being this effect more remarkable for the TDSV-PRC method, with relative differences respect to no PRC: ΔTLR_lung = - 45 ± 24 (TD-PRC), - 55 ± 18 (TDSV-PRC). TLR_necrotic values also decreased when using PRC, with more noticeable differences for TD-PRC: ΔTLR_necrotic = - 52 ± 6 (TD-PRC), - 48 ± 8 (TDSV-PRC).

CONCLUSION

The PRC methods proposed provide a significant quantitative improvement in [⁶⁸Ga]DOTA-labeled PET/CT imaging of mice with neuroendocrine tumors, hence demonstrating that these techniques could also ameliorate the deleterious effect of the positron range in clinical PET imaging.

Effect of Post-Reconstruction Gaussian Filtering on Image Quality and Myocardial Blood Flow Measurement with N-13 Ammonia PET

OBJECTIVES

In order to evaluate the effect of post-reconstruction Gaussian filtering on image quality and myocardial blood flow (MBF) measurement by dynamic N-13 ammonia positron emission tomography (PET), we compared various reconstruction and filtering methods with image characteristics.

METHODS

Dynamic PET images of three patients with coronary artery disease (male-female ratio of 2:1; age: 57, 53, and 76 years) were reconstructed, using filtered back projection (FBP) and ordered subset expectation maximization (OSEM) methods. OSEM reconstruction consisted of OSEM_2I, OSEM_4I, and OSEM_6I with 2, 4, and 6 iterations, respectively. The images, reconstructed and filtered by Gaussian filters of 5, 10, and 15 mm, were obtained, as well as non-filtered images. Visual analysis of image quality (IQ) was performed using a 3-grade scoring system by 2 independent readers, blinded to the reconstruction and filtering methods of stress images. Then, signal-to-noise ratio (SNR) was calculated by noise and contrast recovery (CR). Stress and rest MBF and coronary flow reserve (CFR) were obtained for each method. IQ scores, stress and rest MBF, and CFR were compared between the methods, using Chi-square and Kruskal-Wallis tests.

RESULTS

In the visual analysis, IQ was significantly higher by 10 mm Gaussian filtering, compared to other sizes of filter (P<0.001 for both readers). However, no significant difference of IQ was found between FBP and various numbers of iteration in OSEM (P=0.923 and 0.855 for readers 1 and 2, respectively). SNR was significantly higher in 10 mm Gaussian filter. There was a significant difference in stress and rest MBF between several vascular territories. However CFR was not significantly different according to various filtering methods.

CONCLUSION

Post-reconstruction Gaussian filtering with a filter size of 10 mm significantly enhances the IQ of N-13 ammonia PET-CT, without changing the results of CFR calculation.