Jean-Baptiste Michaud

The Hardware and Signal Processing Architecture of LabPET™, a Small Animal APD-Based Digital PET Scanner

The highly multiplexed analog processing front-end of current Positron Emission Tomography (PET) scanners yields high accuracy for timing but adds significant dead time and offers little flexibility for improvement. A new fully digital APD-based scanner architecture is proposed wherein nuclear pulses are sampled directly at the output of the Charge Sensitive Preamplifier (CSP) with one free-running ADC per channel. This approach offers the opportunity to explore new digital signal processing algorithms borrowed from other fields like command and control theory, as well as advanced heuristics such as neural networks. The analog front-end consists of a dedicated 0.18- mum, 16-channel CMOS charge sensitive preamplifier. Digitization is performed with off-the-shelf dual 8-bit analog-to-digital converters running at 45-MSPS. Digital processing is shared between a FPGA and a Digital Signal Processor (DSP), which can process the data from up to 64 parallel channels without dead time. The FPGA deals with the initial signal analysis for energy measurement and time stamping, while crystal identification is deferred to the DSP running computation-intensive recursive algorithms. The entire system is controlled serially through a Firewire link by a Graphic User Interface. The initial LabPETtrade implementation of the system is a dedicated small animal scanner holding up to 4608 APD channels at an averaged count rate of up to 10 000 events/s each.

System Architecture of the LabPET Small Animal PET Scanner

To address modern molecular imaging requirements, a digital positron emission tomography (PET) scanner for small animals has been developed at Universite de Sherbrooke. Based on individual readout of avalanche photodiodes (APD) coupled to LYSO/LGSO phoswich detectors, the scanner supports up to 4608 channels in a 16.2 cm diameter, 11.25 cm axial field of view with an isotropic ~ 1.2 mm FWHM intrinsic spatial resolution at the center of the field of view. Custom data acquisition boards preprocess and sample APD signals at 45 MHz and compute in real time crystal identification, energy and timing information of detected events at an average sustained rate of up to 1250 raw counts per second per mm2 (10 000 cps/channel). Real time digital signal analysis also filters out events outside the pre-selected energy window with crystal granularity to eliminate Compton events and electronic noise. Retained events are then merged into a single stream through a real-time sorting tree, at which end prompt and delayed coincidences are extracted. A single Firewire link handles both control and data transfers with a host computer. The LabPET features four data recording modes, giving the user the choice to retain data for research or to minimize file size for high coincidence count rate and imaging purposes. The electronic system also supports time synchronized data insertion for flags such as vital signs used in gated image reconstruction. Aside from data acquisition, hardware can generate live energy and discrimination spectra suitable for fast, automatic channel calibration.

Architecture of a dual-modality, high-resolution, fully digital positron emission tomography/computed tomography (PET/CT) scanner for small animal imaging

Contemporary positron emission tomography (PET) scanners are commonly implemented with very large scale integration analog front-end electronics to reduce power consumption, space, noise, and cost. Analog processing yields excellent results in dedicated applications, but offers little flexibility for sophisticated signal processing or for more accurate measurements with newer, fast scintillation crystals. Design goals of the new Sherbrooke PET/computed tomography (CT) scanner are: 1) to achieve 1 mm resolution in both emission (PET) and transmission (CT) imaging using the same detector channels; 2) to be able to count and discriminate individual X-ray photons in CT mode. These requirements can be better met by sampling the analog signal from each individual detector channel as early as possible, using off-the-shelf, 8-b, 100-MHz, high-speed analog-to-digital converters (ADC) and digital processing in field programmable gate arrays (FPGAs). The core of the processing units consists of Xilinx SpartanIIe that can hold up to 16 individual channels. The initial architecture is designed for 1024 channels, but modularity allows extending the system up to 10 K channels or more. This parallel architecture supports count rates in excess of a million hits/s/scintillator in CT mode and up to 100 K events/s/scintillator in PET mode, with a coincidence time window of less than 10 ns full-width at half-maximum.

The architecture of LabPET/spl trade/, a small animal APD-based digital PET scanner

The highly multiplexed, integrated analog processing front-end of current PET scanners yields high accuracy for timing and crystal identification, but also adds significant dead time and offers little flexibility for improvement. A new fully digital APD-based scanner architecture is proposed wherein nuclear pulses are sampled directly at the output of the charge sensitive preamplifier with one free-running ADC per channel. This approach offers the opportunity to explore new digital signal processing algorithms borrowed to command and control theory, as well as advanced heuristics such as neural networks. The analog front-end consists of a dedicated 0.18-mum, 16-channel CMOS charge sensitive preamplifier. Digitization is performed with off-the-shelf dual 8-bit 45-MHz analog-to-digital converters. Digital processing is concentrated in a dual processor FPGA and a digital signal processor (DSP), which can process the data from up to 64 parallel channels with no dead time. The FPGA deals with the initial signal analysis for energy measurement and time stamping, while crystal identification is deferred to the DSP running computation-intensive auto-regressive algorithms. The entire system is controlled serially through a Firewire link by a graphic user interface. The initial LabPETtrade implementation of the system is a dedicated small animal scanner holding up to 3072 APD channels at an averaged count rate of up to 10 000 events/s each

Real time digital signal processing implementation for an APD-based PET scanner with phoswich detectors

Recent progress in advanced digital signal processing provides an opportunity to expand the computation power required for real time extraction of event characteristics in APD-based positron emission tomography (PET) scanners. These developments are made possible by a highly parallel data acquisition (DAQ) system based on an integrated analog front-end and a high-speed fully digital signal processing section that directly samples the output of each preamplifier with a free-running, off-the-shelf, 45-MHz MAX1193 analog-to-digital converter that feeds the sampled data into a field programmable gate array VirtexII PRO from Xilinx. This FPGA features ~31,000 logic cells and 2 PowerPC processors, which allows up to 64 channels to be processed simultaneously. Each channel has its own digital signal processing chain including a trigger, a baseline restorer and a timestamp algorithm. Various timestamp algorithms have been tested so far, achieving a coincidence timing resolution of 3.2 ns FWHM for APD-LSO and 11.4 ns FWHM for APD-BGO detectors, respectively. Channels are then multiplexed into a TMS320C6414 DSP processor from Texas Instruments for crystal identification by an ARMAX recursive algorithm borrowed from identification and vector quantization theory. The system can sustain an event rate of 10 000 events/s/channel without electronic dead time

Experimental results of identification and vector quantization algorithms for DOI measurement in digital PET scanners with phoswich detectors

DOI measurement in phoswich PET scanners still relies mostly on traditional Pulse Shape Discrimination (PSD), transposed from analog electronics. PSD performance is limited in two conditions: measurement noise increases the error rate, as with low-energy Compton photons; and phoswich stacking of the newer, fast crystal materials like LSO, LYSO and LuAP show intrinsic low discrimination success. These impairments somewhat limit the widespread use of such stacking, as well as recuperation and treatment of Compton photons. We propose two new algorithms adapted from other fields of electrical engineering, but unused in radiation detection so far, that mostly circumvent these problems: identification, from command-and-control applications, followed by vector quantization, from speech recognition. These algorithms exhibit operational properties that mitigate the above problems. In our previous work, we explained the steps required to adapt the algorithms to DOI application. This paper presents discrimination results for all photons of energy greater than 100 keV detected in any stacking of BGO, LSO, LYSO, LuAP and/or GSO materials. Errors are un-correlated with crystal statistical noise and/or energy resolution, with electronics white noise and with timestamp uncertainty. For all measurements made (N=40,000), the error rate is null, except for Compton discrimination with the faster crystals, where it does not exceed 0.5%. This far surpasses conventional PSD results.

Crystal Identification Based on Recursive-Least-Squares and Least-Mean-Squares Auto-Regressive Models for Small Animal Pet

Most positron emission tomography (PET) scanners still partly rely on analog processing to sort out events from the PET detector front-end. Recent all-digital architectures enable the use of more complex algorithms to solve common problems in PET scanners, such as crystal identification and parallax error. Auto-regressive exogeneous variable (ARX) algorithms were shown to be among the most powerful methods of crystal identification by pulse shape discrimination (PSD) for parallax mitigation or resolution improvement with phoswich detectors. Although ARX algorithms achieve a nearly 100% discrimination accuracy even in a noisy environment, such methods are computationally expensive and can hardly be implemented in a real time digital PET system. A crystal identification method based on adaptive filter theory using an auto-regressive (AR) model is proposed to enable real time crystal identification in a noisy environment.

Time discrimination techniques using artificial neural networks for positron emission tomography

Relevant information in positron emission tomography is currently being obtained mostly by analog signal-processing methods. New digital PET scanner architectures are now becoming available, which offer greater flexibility and easier reconfiguration capability as compared to previous PET designs. Moreover, new strategies can be devised to extract more information with better accuracy from the digitized detector signals. Trained artificial neural networks (ANN) have been investigated to improve coincidence timing resolution with different types of Avalanche PhotoDiode (APD)-based detectors. The signal at the output of a charge-sensitive preamplifier was digitized with an off-the-shelf, free-running 100-MHz, 8-bit analog-to-digital converter and time discrimination was performed with ANNs implemented in field-programmable gate array (FPGA). Results show that ANNs can be particularly efficient with slow and low light output scintillators, such as BGO, but less so with faster luminous crystals, such as LSO. In reference to a fast PMT-plastic detector, a time resolution of 6.5 ns was achieved with a BGO-APD detector. With LSO, the ANN was found to be competitive with other digital techniques developed in previous works. ANNs implemented in FPGAs provide a fast and flexible circuit that can be easily reconfigured to accommodate various detectors under different signal/noise conditions.

Sensitivity in PET: Neural networks as an alternative to compton photons LOR analysis

In high-resolution small-animal positron emission tomography (PET), sensitivity remains an active issue. Sensitivity can be increased by lowering the energy threshold to include more Compton-scattered events, but then computation of the correct annihilation line-of-response (LOR) proves problematic. The complexity of Compton-kinematics analysis, compounded with finite energy resolution and detection position quantization of finite-size detectors, yields unaffordable methods with rather poor success rates. As an alternative, this paper proposes an artificial neural network (ANN) approach, which forfeits all explicit handling of equations at the expense of a priori statistical training, and which has the potential to better handle the previous measurement impairments. The method first consists in a preprocessing step involving geometrical transformations, which simplifies the actual use of the neural network, in the second step. This paper presents the methods proof-of-concept. It focuses on a simple yet prevalent inter-crystal scatter scenario, where a 511-keV annihilation photon is detected coincidently with two inter-crystal-scattered photons whose energy sum accounts for the whole 511 keV annihilation energy. It shows, in preliminary simulations, a promising correct LOR computation rate in the range from 90 to 94%. Finally, it discusses the steps and requirements for the eventual implementation of the method, including further validation, hardware requirements, system- level issues and possible other applications.

System Integration of the LabPET Small Animal PET Scanner

To address modern molecular imaging requirements, a digital positron emission tomography scanner for small animals has been developed at Universite de Sherbrooke. Based on individual readout of avalanche photodiodes (APD) coupled to a LYSO/LGSO phoswich array, the scanner supports up to 3072 channels in a 16.2 cm diameter, 7.5 cm axial field of view with an isotropic 1.2 mm FWHM intrinsic spatial resolution at the center of the FOV. Custom data acquisition boards sample APD signals at 45 MHz and compute in real time crystal identification, energy and timing information of detected events at rates of up to 1250 raw counts per second per mm2 (10k cps/channel). Real time digital signal analysis also filters out events outside the photopeak with crystal granularity to eliminate Compton events and electronic noise. Retained events are then merged into a single stream through a real-time sorting tree, at which end the prompt and delayed coincidences are extracted. A single Firewire link handles both control and data transfers with a computer. The LabPETtrade features four data recording modes, giving the user the choice to retain data for research or to minimize file size for high coincidence count rate and imaging purposes. The electronic system also supports time synchronized data insertion for flags such as vital signs used in gated image reconstruction. Aside from data acquisition, hardware can generate live energy and discrimination histograms suitable for fast, automatic channel calibration.