The reconstruction of electrons and photons in CMS depends on the topological clustering of the energy deposited by an incident particle in different crystals of the electromagnetic calorimeter (ECAL). The currently used algorithm cannot account for the energy deposits coming from the pileup (secondary collisions) efficiently. The performance of this algorithm is expected to degrade during the LHC Run 3 because of the larger average pileup level and the increasing level of noise due to the aging of the ECAL detector. In this paper, we explore new techniques for energy reconstruction in ECAL using state-of-the-art machine learning algorithms like graph neural networks and self-attention modules.
Radiomics features extract quantitative information from medical images, towards the derivation of biomarkers for clinical tasks, such as diagnosis, prognosis, or treatment response assessment. Different image discretization parameters (e.g. bin number or size), convolutional filters, segmentation perturbation, or multi-modality fusion levels can be used to generate radiomics features and ultimately signatures. Commonly, only one set of parameters is used; resulting in only one value or flavour for a given RF. We propose tensor radiomics (TR) where tensors of features calculated with multiple combinations of parameters (i.e. flavours) are utilized to optimize the construction of radiomics signatures. We present examples of TR as applied to PET/CT, MRI, and CT imaging invoking machine learning or deep learning solutions, and reproducibility analyses: (1) TR via varying bin sizes on CT images of lung cancer and PET-CT images of head neck cancer (HNC) for overall survival prediction. A hybrid deep neural network, referred to as TR-Net, along with two ML-based flavour fusion methods showed improved accuracy compared to regular rediomics features. (2) TR built from different segmentation perturbations and different bin sizes for classification of late-stage lung cancer response to first-line immunotherapy using CT images. TR improved predicted patient responses. (3) TR via multi-flavour generated radiomics features in MR imaging showed improved reproducibility when compared to many single-flavour features. (4) TR via multiple PET/CT fusions in HNC. Flavours were built from different fusions using methods, such as Laplacian pyramids and wavelet transforms. TR improved overall survival prediction. Our results suggest that the proposed TR paradigm has the potential to improve performance capabilities in different medical imaging tasks.
A Phase-II upgrade of the ATLAS detector for the High Luminosity LHC will affect most of the detector’s subsystems, including the Trigger and Data Acquisition (TDAQ) subsystem. The core part of the TDAQ subsystem in Phase-II is the Global Trigger, which is capable of performing algorithms on a full-granularity calorimeter cells as well as applying topological requirements and refining the trigger objects from the upstream Feature Extractors. A Global Common Module (GCM) is the main building block of the Global Trigger. A design of an additional, standalone, Global Trigger Versatile Module (GVM) has been completed in compliance with the hardware specifications of the Global Trigger prototype. While possessing high processing and bandwidth capabilities, the GVM is intended to be used as an auxiliary hardware component for operational, testing and development purposes of the Global Trigger as well as other resource and bandwidth critical projects. A high-density PCB of the GVM is designed for high-speed data transmission and features 25.8 Gb/s capable Finisar BOA optical modules, a modern Xilinx Ultrascale+ VU13P FPGA as well as other hardware components essential for the Global Trigger. GCM development firmware tests, VU13P FPGA’s and Finisar BOA optical modules’ performance evaluation as well as the module’s main hardware functionality verification composes a full testing program for the GVM, which was passed successfully.
Many nuclear safety applications need fast, portable, and accurate imagers to better locate radiation sources. The Rotating Scatter Mask (RSM) system is an emerging device with the potential to meet these needs. The main challenge is the under-determined nature of the data acquisition process: the dimension of the measured signal is far less than the dimension of the image to be reconstructed. To address this challenge, this work aims to fuse model-based sparsity-promoting regularization and a data-driven deep neural network denoising image prior to perform image reconstruction. An efficient algorithm is developed and produces superior reconstructions relative to current approaches.
In astrophysics, several key questions in the hard X-/soft Gamma-ray range (>100 keV) require sensitivity and angular resolution that are hardly achievable with current technologies. Therefore, a new kind of instrument able to focus hard X and gamma-rays is essential. Broad band Laue lenses seem to be the only solution to fulfil these requirements, significantly improving the sensitivity and angular resolution of the X-/gamma-ray telescopes. This type of high-energy optics will require highly performing focal plane detectors in terms of detection efficiency, spatial resolution, and spectroscopy. This paper presents the results obtained in the project "Technological Readiness Increase for Laue Lenses (TRILL)" framework using a Caliste-HD detector module. This detector is a pixel spectrometer developed at CEA (Commissariat à Energie Atomique, Saclay, France). It is used to acquire spectroscopic images of the focal spot produced by Laue Lens bent crystals under a hard X-ray beam at the LARIX facility (University of Ferrara, Italy).
Hyperspectral Computed Tomography (HCT) Data is often visualized using dimension reduction algorithms. However, these methods often fail to adequately differentiate between materials with similar spectral signatures. Previous work showed that a combination of image preprocessing, clustering, and dimension reduction techniques can be used to colorize simulated HCT data and enhance the contrast between similar materials. In this work, we evaluate the efficacy of these existing methods on experimental HCT data and propose new improvements to the robustness of these methods. We introduce an automated channel selection method and compare the Feldkamp, Davis, and Kress filtered back-projection (FBP) algorithm with the maximum-likelihood estimation-maximization (MLEM) algorithm in terms of HCT reconstruction image quality and its effect on different colorization methods. Additionally, we propose adaptations to the colorization process that eliminate the need for a priori knowledge of the number distinct materials for material classification. Our results show that these methods generalize to materials in real-world experimental HCT data for both colorization and classification tasks; both tasks have applications in industry, medicine, and security, wherever rapid visualization and identification is needed.
Inspired by the recent success of transformers for Natural Language Processing and vision transformer for Computer Vision, many researchers in the medical imaging community have flocked to transformer-based networks for various main stream medical tasks such as classification [1] - [3] segmentation [4] - [6] and registration [7], [8]. In this study, we analyze, two recently published transformer-based network architectures for the task of multimodal head-and-tumor segmentation and compare their performance to the de facto standard 3D segmentation network – the nnU-Net. Our results showed that modelling long-range dependencies may be helpful in cases where large structures are present and/or large field of view is needed. However, for small structures such as head-and-neck tumor, the convolution-based U-Net architecture seemed to perform well, especially when training dataset is small and computational resource is limited.
Our group at the University of Pennsylvania has designed and built a dedicated high spatial resolution time-of-flight (TOF)-capable breast PET (BPET) scanner integrated with a digital breast tomosynthesis (DBT) unit in a common gantry to provide co-registered PET-DBT images. The BPET scanner is comprised of two detector heads, with each head composed of a 4x2 array of PET detectors built using 1.5x1.5x15 mm3 LYSO crystals. The PET detector head separation is set to 9 cm, providing a PET FOV of 20x9x10 cm3. In comparison with conventional dual-headed PET scanners, TOF information will help in alleviating limited-angle image artifacts and improve lesion quantification. This dedicated scanner will thus provide the ability to more accurately measure radiotracer uptake in smaller lesions that are prevalent in breast cancer. A custom data acquisition system performs fast signal waveform sampling at 4 Gsps with minimal deadtime. This paper describes the full system design and presents early imaging performance of the BPET scanner. In particular, results from reconstructed spatial resolution and phantom measurements are presented.
We present a new framework for PET-MR reconstruction using deep learning. The proposed network named Dense Syn-Net is a new version of the synergistic network Syn-Net. This method is built on two model-based image reconstruction algorithms, the maximum a posteriori expectation-maximization algorithm for PET and the Landweber algorithm for MR, that we fully connect to each other. To avoid the use of handcrafted regularisations, the gradient of the priors for PET and MR are learned from training data along with regularisation strengths for both modalities. Two major modifications of the original Syn-Net have been introduced: i) iteration-dependent targets are used to ensure that the output of every module matches the corresponding iteration of the reconstruction of high quality data, ii) all the previous PET and MR estimates are used to guide the regularisation in a given module allowing both self and inter-modality guidance. Results on 2D simulated data show that Dense-Syn-Net outperforms conventional independent PET and MR reconstruction algorithms. For MR, our method offers improvements compared to deep learned independent methods and synergistic PET-MR reconstruction with mutually weighted quadratic priors. For PET reconstruction, our network shows greater robustness towards mismatches than MR-guided methods by better preserving modality-unique features. Dense Syn-Net improved global image reconstruction accuracy compared to Syn-Net, however the latter performs better for regions of mismatch. Future work will need to focus on assessing the performance of the network on 3D real data and performing ablation studies to investigate the need of fully connections.
Simultaneous positron emission tomography / magnetic resonance imaging (PET/MRI) acquires fused anatomical and molecular information of diseases. However, this diagnostically attractive idea is highly technically challenging due to the mutual interference between PET and MRI. Currently, clinical integrated PET/MRI systems are available, but only three support simultaneous PET/MRI and only two support time-of-flight (ToF). For small brain structures, a brain-dedicated PET insert provides better spatial resolution and sensitivity performances at a more affordable cost since it can be inserted into any existing MR system. Our first generation RF-penetrable brain PET insert has proven RF-penetrability and MR-compatibility. In the second generation system, SiPM signals are digitized inside detectors which results in better PET performance but presents a greater technical challenge for achieving MR-compatibility. In this paper, we report our initial MR-compatibility results for two fully assembled TOF PET detector modules. We studied the MR noise spectrum and MR image quality change with PET detectors present. With the presence of PET detectors, no noise peaks were induced in the MR acquisition, but the average noise level was increased by 15%, which led to a -3.1 to -4.2 dB degradation in MR image SNR. The uniformity of MR images was influenced by less than 2%. No ghosting artifacts were detected (ghosting intensity level (GIL) < 2%). The changes seen with just two detectors may not exist once we scale up to the full PET ring.
This paper describes the transmission tomography extension to the Open-source MATLAB Emission Tomography Software (OMEGA), now named as Open-source MATLAB Emission and Transmission Tomography Software. The OMEGA toolkit was originally designed for the easy image reconstruction of positron emission tomography (PET) data. Now we extend many of the features available for PET imaging to also for transmission tomography, with examples provided for cone beam computed tomography (CBCT). OMEGA supports both MAT-LAB and the open-source alternative GNU Octave. All built-in algorithms for PET are available for the transmission tomography case. These include 10 (Poisson) maximum likelihood algorithms, 8 maximum a posteriori algorithms and 11 priors. Furthermore, custom algorithms using Gaussian-based reconstruction can be implemented with a MATLAB/Octave class object that can be used to compute either the forward projection or backward projection. OMEGA reconstructions, whether using the built-in functions or the class object, can be performed by using parallel computing utilizing either OpenCL for GPUs or OpenMP for CPUs. Both methods allow for completely matrix-free computations, enabling the use of high resolution scanners with the optimized use of memory. OpenCL and OpenMP also enable the use of any computer hardware in the reconstruction. Additionally, as with PET data, built-in support exists for GATE simulated CBCT data allowing easy reconstructions of Monte Carlo simulated data. This extension to transmission tomography extends the usability of OMEGA software to the efficient reconstruction of transmission tomography data.
The front-end electronics of the ATLAS muon drift-tube chambers will be upgraded in the experiment’s phase-II upgrade to comply with the new trigger and read-out scheme at the HL-LHC. A new amplifier shaper discriminator chip was developed in 130 nm Global Foundries technology for this upgrade. A preproduction of 7500 chips was launched in 2019 and tested in 2020. The presentation will summarize the functionality of the new ASD chip, the test set-up and testing procedure as well as the test results which show a production yield of 93%. Based on the successful test of the preproduction chip the serious production of 80,000 chips was carried out in fall 2020. The tests of a sample of 1000 production chips show the same yield as the preproduction chips.
We discuss the architecture and characterization of an advanced frontside-illuminated (FSI) multi-channel digital silicon photomultiplier (MD-SiPM) fabricated in 0.18 µm/0.18 µm 3D-stacked CMOS technology. The top-tier chip houses FSI photodetectors, the bottom-tier chip photon timestamping, signal processing, and communication logic. The total chip size is 7.5×4.2 mm2, comprising two arrays of 8×8 clusters, each composed of 64 single-photon avalanche diodes (SPADs). The sensor was electro-optically tested using a laser-based setup. Key parameters, such as dark count rate, hot pixel distribution, TSV yield, response linearity and saturation were characterized and complemented by preliminary radiation measurements on LYSO scintillators coupled to a 22Na source.
Dimension reduction techniques have frequently been used to summarize information from high dimensional hyperspectral data, usually done in effort to classify or visualize the materials contained in the hyperspectral image. The main challenge in applying these techniques to Hyperspectral Computed Tomography (HCT) data is that if the materials in the field of view are of similar composition then it can be difficult for a visualization of the hyperspectral image to differentiate between the materials. We propose novel alternative methods of preprocessing and summarizing HCT data in a single colorized image and novel measures to assess desired qualities in the resultant colored image, such as the contrast between different materials and the consistency of color within the same object. Proposed processes in this work include a new majority-voting method for multi-level thresholding, binary erosion, median filters, PAM clustering for grouping pixels into objects (of homogeneous materials) and mean/median assignment along the spectral dimension for representing the underlying signature, UMAP or GLMs to assign colors, and quantitative coloring assessment with developed measures. Strengths and weaknesses of various combinations of methods are discussed. These results have the potential to create more robust material identification methods from HCT data that has wide use in industrial, medical, and security-based applications for detection and quantification, including visualization methods to assist with rapid human interpretability of these complex hyperspectral signatures.
Spectroscopic imagers based on high-Z and wide-bandgap compound semiconductor detectors are widely proposed for the detection of prompt gamma rays in boron neutron capture therapy (BNCT). BNCT is a therapy based on the neutron capture reaction 10B (n,α)7Li. To perform a real-time monitoring of the spatial distribution of 10B during the treatments, the detection of the prompt gamma rays (478 keV), produced by the 7Li recoil nuclei, can be helpful. In this work, we presented the potentialities of new high-resolution CZT drift strip detectors, recently developed by our group, for BNCT measurements. The detectors, exploiting the analysis of the collected-induced charge pulses from anodes, cathodes and drift strips, show excellent energy resolution < 1% at 662 keV at room temperature. The results of preliminary gamma ray measurements under thermal neutrons at the T.R.I.G.A. Mark II research nuclear reactor of Pavia University (Italy) are shown.
The CMS detector at the CERN Large Hadron Collider is undergoing an extensive Phase II upgrade program to prepare for the challenging conditions of the High-Luminosity LHC. A new timing detector in CMS will measure minimum ionizing particles (MIPs) with a time resolution of 30-40 ps for MIP signals at a rate of 2.5 Mhit/s per channel at the beginning of HL-LHC operation. The precision time information from this detector will be used to reduce the effects of the high levels of pileup expected at the HL-LHC, bringing new capabilities to the CMS detector. The central barrel part of the detector, Barrel Timing Layer (BTL), will be based on LYSO:Ce crystals read out with SiPMs with TOFHIR ASICs for the front-end readout. The BTL will use elongated crystal bars, read out by a SiPM on each end of the crystal, in order to maximize detector performance within the constraints of space, cost, and channel count. This geometry enables to cover large surfaces with a minimal active area of the photodetectors, thus reducing noise and power consumption. The conceptual design of BTL have been validated with measurements with test beam of high energy particles. The close to final prototypes of the detector modules and readout electronics have been produced and are being tested.
The electronics of the CMS Drift Tube (DT) chambers will need to be replaced in order to cope with the increased occupancy and rates at the high luminosity phase in the HL-LHC (High-Luminosity Large Hadron Collider). A new electronic system is being designed, that will allow forwarding signals from all DT chambers at the maximum resolution to the backend system, where improved trigger primitives generation will take place. The on-detector boards, which will be attached to the DT chambers inside the CMS wheels are called OBDT (On Board electronics for Drift Tubes) and are built around a Microsemi Polarfire FPGA that performs the time digitization of the chamber signals at ~1 ns resolution, and forwards them to a high-speed link, optimizing bandwidth and latency. Thirteen prototypes of this board have been installed in parallel to the present system in the CMS detector, fully instrumenting one out of sixty sectors, and are taking data routinely integrated in the CMS DAQ chain. The firmware implementation of this FPGA performs the time digitization of up to 240 input channels, the high-speed link implementation and several slow control functionalities, including a very thorough clock and reset signals distribution that allows ensuring that the time measurements are stable across power cycles and reconfigurations of all the CMS clock chain. The description and implementation details, together with the results of the performance of this firmware in the data taking campaigns at CMS are presented in this contribution.
Pixelated CZT detectors have been used in a variety of molecular imaging applications for many years. The interplay of gamma camera and collimator geometric design, gantry motion, and image reconstruction determines the image quality and dose-time-FOV trade-offs. In particular, Molecular Breast Imaging (MBI) has been shown to provide excellent diagnostic results in patients with dense breast tissue, but higher than mammography patient dose and long imaging time impede its wide adoption. We propose a new transformative system concept combining the advantages of CZT detectors (superior energy and position resolution and depth of interaction sensing), multi-pinhole collimation and novel image reconstruction to mitigate those drawbacks without compromising diagnostic content. The closely spaced pinholes allow tomographic image reconstruction, improve sensitivity and angular sampling, but result in significant multiplexing. Novel de-multiplexing algorithms have been developed to mitigate the adverse multiplexing artefacts using the DOI. GATE simulations of the new camera demonstrate a potential to reduce the patient dose by at least a factor of 5 in comparison to planar MBI, thus reducing the dose to the level of an average mammography scan. The first prototype has been built at Kromek with 3D position sensitive CZT detectors and is being evaluated using an "activity-painting" setup with a point 57Co source. Initial results demonstrate the expected performance improvement with the use of sub-pixelisation and DOI. The next steps of the development will include accurate evaluation of the image quality and the dose reduction followed by building a larger scale clinical prototype using optimised detector design.
