IEEE Nuclear Science Symposium conference record. Nuclear Science Symposium最新文献

Recent advances in deep learning (DL) have greatly improved the performance of positron emission tomography (PET) denoising performance. However, DL model performance can vary a lot across subjects, due to the large variability of the count levels and spatial distributions. A generalizable DL model that mitigates the subject-wise variations is highly expected toward a reliable and trustworthy system for clinical application. In this work, we propose a contrastive adversarial learning framework for subject-wise domain generalization (DG). Specifically, we configure a contrastive discriminator in addition to the UNet-based denoising module to check the subject-related information in the bottleneck feature, while the denoising module is adversarially trained to enforce the extraction of subject-invariant features. The sampled low-count realizations from the list-mode data are used as anchor-positive pairs to be close to each other, while the other subjects are used as negative samples to be distributed far away. We evaluated on 97 ¹⁸F-MK6240 tau PET studies, each having 20 noise realizations with 25% fractions of events. Training, validation, and testing were implemented using 1400, 120, and 420 pairs of 3D image volumes in a subject-independent manner. The proposed contrastive adversarial DG demonstrated superior denoising performance than conventional UNet without subject-wise DG and cross-entropy-based adversarial DG.

Diffusion models (DM) built from a hierarchy of denoising autoencoders have achieved remarkable progress in image generation, and are increasingly influential in the field of image restoration (IR) tasks. In the meantime, its backbone of autoencoders also evolved from UNet to vision transformer, e.g. Restormer. Therefore, it is important to disentangle the contribution of backbone networks and the additional generative learning scheme. Notably, DM shows varied performance across IR tasks, and the performance of recent advanced transformer-based DM on PET denoising is under-explored. In this study, we further raise an intuitive question, "{if we have a sufficiently powerful backbone, whether DM can be a general add-on generative learning scheme to further boost PET denoising}". Specifically, we investigate one of the best-in-class IR models, i.e., DiffIR, which is a latent DM based on the Restormer backbone. We provide a qualitative and quantitative comparison with UNet, SR3 (UNet+pixel DM), and Restormer, on the 25% low dose ¹⁸F-FDG whole-body PET denoising task, aiming to identify the best practices. We trained and tested on 93 and 12 subjects, and each subject has 644 slices. It appears that Restormer outperforms UNet in terms of PSNR and MSE. However, additional latent DM over Restormer does not contribute to better MSE, SSIM, or PSNR in our task, which is even inferior to the conventional UNet. In addition, SR3 with pixel space DM is not stable to synthesize satisfactory results. The results are consistent with the natural image super-resolution tasks, which also suffer from limited spatial information. A possible reason would be the denoising iteration at latent feature space cannot well support detailed structure and texture restoration. This issue is more crucial in the IR tasks taking inputs with limited details, e.g., SR and PET denoising.

Delineating lesions and anatomical structure is important for image-guided interventions. Point-supervised medical image segmentation (PSS) has great potential to alleviate costly expert delineation labeling. However, due to the lack of precise size and boundary guidance, the effectiveness of PSS often falls short of expectations. Although recent vision foundational models, such as the medical segment anything model (MedSAM), have made significant advancements in bounding-box-prompted segmentation, it is not straightforward to utilize point annotation, and is prone to semantic ambiguity. In this preliminary study, we introduce an iterative framework to facilitate semantic-aware point-supervised MedSAM. Specifically, the semantic box-prompt generator (SBPG) module has the capacity to convert the point input into potential pseudo bounding box suggestions, which are explicitly refined by the prototype-based semantic similarity. This is then succeeded by a prompt-guided spatial refinement (PGSR) module that harnesses the exceptional generalizability of MedSAM to infer the segmentation mask, which also updates the box proposal seed in SBPG. Performance can be progressively improved with adequate iterations. We conducted an evaluation on BraTS2018 for the segmentation of whole brain tumors and demonstrated its superior performance compared to traditional PSS methods and on par with box-supervised methods.

Instrumentation research in small animal Positron Emission Tomography (PET) imaging is driven by improving timing, spatial resolution and sensitivity. Conventional PET scanners are built of multiple detectors placed in a cylindrical geometry with gaps between them in both the transaxial and axial planes. These gaps decrease sensitivity and degrade spatial resolution towards the edges of the system field of view (FOV). To mitigate these problems, we have designed and validated an edgeless pre-clinical PET system based on a single LYSO annulus with an inner diameter of 62 mm and 10 outer facets of 26 × 52 mm² each. The scintillation light is read out using the row and columns of Silicon Photomultipliers (SiPMs) mounted in magnetic-field compatible PCBs. The objective of this work is to provide a calibration method for this system. The particular design of the annulus produces some undesirable effects in the light distributions (LD) at the module joints, which needs to be addressed. Nevertheless, after calibration, the system allows one to properly retrieve both, the energy and 3D photon impact positions.

In this study, we investigated large scale radoimics on 116 breast cancer patients. We are particularly interested in unsupervised learning to bicluster patients and features in order to associate such biclusters with the disease characteristics. The results show that radiomics features with wavelet features have a better biclustering ability. And 172 radiomics features have shown a better classification capability.

X-ray grating-based differential phase-contrast imaging is able to obtain excellent soft-tissue contrast of phase, attenuation, and small angle scatter. In this work we model the performance of an X-ray interferometer wherein the phase gratings are replaced with a single Fresnel micro-bi-prism. Our goal is to develop imaging systems based on bi-prism interferometry with improved polychromatic performance. In our investigation we obtain an analytical expression for the irradiance distribution of the bi-prism. The localized regions of fringe visibility within the irradiance distribution are non-periodic. Following the work of Pfeiffer et al., we then develop a method for reconstructing scattering directions that can be used to obtain a three-dimensional tensor field. This will eventually be used in modified bi-prism-based differential phase-contrast imaging to obtain tissue properties through mathematical reconstruction of tensor tomographic data.

A new low-power application-specific integrated circuit (ASIC) for Cadmium Zinc Telluride (CZT) detectors for single-photon emission computed tomography (SPECT) application is being developed at BNL. As the first step, a 32-channel prototype ASIC was designed and tested recently. Each channel has a preamplifier followed by CR-RC³ shaping circuits and three independent energy bins with comparators and 16-bit counters. The ASIC was fabricated with TSMC 0.35-μm complementary metal-oxide-semiconductor (CMOS) process and tested in laboratories. The power consumption is around 1 mW/ch with a 2.5-V supply. With a gain of 400 mV/fC and the peaking time of 500 ns, the equivalent noise charge (ENC) of 360 e- has been measured in room temperature while the crosstalk rate is less than 0.3%. The 10-bit DACs for global thresholds have an integral nonlinearity (INL) less than 0.56% and differential nonlinearity (DNL) less than 0.33%. In the presentation, we will report the detailed test results with this ASIC.

In this work we have parallelized the Maximum Likelihood Expectation-Maximization (MLEM) and Ordered Subset Expectation Maximization (OSEM) algorithms for improving efficiency of reconstructions of multiple pinholes SPECT, and cone-bean CT data. We implemented the parallelized versions of the algorithms on a General Purpose Graphic Processing Unit (GPGPU): 448 cores of a NVIDIA Tesla M2070 GPU with 6GB RAM per thread of computing. We compared their run times against those from the corresponding CPU implementations running on 8 cores CPU of an AMD Opteron 6128 with 32 GB RAM. We have further shown how an optimization of thread balancing can accelerate the speed of the GPU implementation.

The primary goal of this project is to implement the iterative statistical image reconstruction algorithm, in this case maximum likelihood expectation maximum (MLEM) used for dynamic cardiac single photon emission computed tomography, on Spark/GraphX. This involves porting the algorithm to run on large-scale parallel computing systems. Spark is an easy-to- program software platform that can handle large amounts of data in parallel. GraphX is a graph analytic system running on top of Spark to handle graph and sparse linear algebra operations in parallel. The main advantage of implementing MLEM algorithm in Spark/GraphX is that it allows users to parallelize such computation without any expertise in parallel computing or prior knowledge in computer science. In this paper we demonstrate a successful implementation of MLEM in Spark/GraphX and present the performance gains with the goal to eventually make it useable in clinical setting.

Coronary steal (CS) is a physiological process that induces absolute decrease in blood flow in collateralized myocardium compared to resting flow during coronary vasodilation due to redistribution of blood away from collateral-dependent myocardium. Although, CS has been well known for decades, there are very few noninvasive perfusion studies in humans that quantitatively predict the existence of CS. In this study, we show that the quantitative measurement of absolute value of regional myocardial blood flow (MBF) and coronary flow reserve (CFR) using dynamic single photon emitted computed tomography (SPECT) can help estimate the presence of CS in myocardium with obstructed coronary artery and collateral circulation.