Continuously Sampled Digital Pulse Processing for ...

1 downloads 0 Views 1MB Size Report
Abstract–The QuickSilver Event Processing Module (EPM) is a key component of a high performance data acquisition platform from Siemens Molecular Imaging ...
2007 IEEE Nuclear Science Symposium Conference Record

M26-36

Continuously Sampled Digital Pulse Processing for Inveon Small Animal PET Scanner Aaron R. McFarland, Member, IEEE, Stefan Siegel, Member, IEEE, Danny F. Newport, Member, IEEE, Robert Mintzer, Member, IEEE, Blake Atkins, Senior Member, IEEE, and Mark Lenox Abstract–The QuickSilver Event Processing Module (EPM) is a key component of a high performance data acquisition platform from Siemens Molecular Imaging (Knoxville, TN) for use in the Inveon™ line of multimodal PET and SPECT preclinical imaging systems. The card’s main purpose is to condition, digitize and process incoming analog pulses from PMT or APD based PET or SPECT detectors. Analog pulses from a detector are digitized using a 100MHz continuous sampling ADC and read into a Xilinx Virtex II Pro FPGA for processing. The FPGA performs digital integration, baseline offset correction and pileup rejection. Because these functions are done in the digital domain, different algorithms can be quickly re-implemented and tested. The EPM has the ability to capture raw event ADC samples, allowing for the quick development and comparison of new algorithms in software on actual event samples. The Inveon™ small animal PET scanner uses a larger LSO block detector and new analog front end than previous generation scanners which increases the likelihood of pileup events. The digital pulse processing methods presented here have been evaluated to obtain the best energy and positioning performance from the high pixel count Inveon™ detectors while maintaining high stability across countrates.

and D channels simultaneously. Once the 16 samples are acquired for an event, the critical task of calculating the energy of A, B, C, D must be accomplished while accounting for baseline shifts, pileup events and other effects.

Figure 1 Event Processing Module Board

I. INTRODUCTION

Inveon PET Analog Front End Block Diagram

Inveon™ small animal PET scanner [1][2] detector Tandhereadout electronics consist of a 20x20mm pixilated LSO block, tapered light guide, and position sensitive photo multiplier tube (PMT) [3]. The PMT output is multiplexed down from 12 to 4 signals at the preamplifier and sent to an EPM board [4] shown in Figure 1. The EPM has a custom ASIC which provides CFD and TDC functions along with pulse shaping and amplification of the analog signals [5]. The ASIC’s analog outputs, labeled A, B, C, and D, are then continuously sampled using 10-bit, 100MHz analog to digital converters from Analog Devices [6] and read out by a Xilinx Virtex II Pro FPGA [7]. This signal processing path is shown in Figure 2. II. ACQUISITION METHOD DESCRIPTION The EPM FPGA is configured to collect 16 samples of continuous ADC data when the ASIC CFD triggers. At 100 MHz, this corresponds to 10ns between each ADC sample for a total sampled time of 160ns. For the case of fast LSO pulses, even after shaping, this 160ns window contains all of the event pulse of interest. The 16 samples are taken across the A, B, C, Manuscript received November 23, 2007. A. R. McFarland, D. F. Newport, B. Atkins, S. Siegel, R. Mintzer, and M. Lenox are with Siemens Molecular Imaging, Knoxville, TN 37932 USA (telephone: +1 865 218 1627, e-mail: [email protected]).

1-4244-0923-3/07/$25.00 ©2007 IEEE.

LSO Block

PS-PMT

X1

A

A

10

X2

B

B

10

C

10

D

10

Y1 Y2

ASIC

C D

100Mhz ADC

Xilinx FPGA

CFD Trigger EPM Board

Figure 2

III. ALGORITHM COMPARISION AND PROCESSING The energy calculation of an event pulse is a tradeoff between performance and limited FPGA resources. In the analog domain, many peak sampling or analog integration techniques have been used in the past to calculate energy. The digital equivalent of these can be done in the FPGA, but with more flexibility. As a starting point, a simple peak minus baseline method was used to calculate energy as shown in figure 3, where: Pulse Energy (A,B,C, or D) = N7 – N0 This method provides good results for both energy and position with a total block energy resolution of 20% and a position profile peak-to-valley ratio of 2.2 at the block edge. The method works reasonability well at low count rates.

4262

However, as the count rate increases it becomes more susceptible to pileup events and baseline changes. While the energy spectra shows little change at increased rates, the event positioning, especially for the edge crystals, begins to degrade due to the centroid calculation used for computing the event position. ADC Sampling of PMT Shaped Pulse N+7

N+8

Pulse Peak

Amplitude

Pulse Baseline

N

N+1

N+15

valley ratio of 2.2 at the edge, which is an 18% improvement over the first method. In a multi-pixel block detector such as the Inveon™ detector module, pileup events not only lead to incorrect energy discrimination but also to spatially mispositioned events. As more pileup occurs at higher count rates, the ANGER readout logic found in many block detectors cause the positioning of pileup events to average toward the center region of the block. Mispositioned events can be seen in the Coincidence Efficiencies and Detector Flood images Figures 4 and 5. In Figure 5, a cloud of mispositioned events can be seen in the central detector region. The same effect is seen in the Coincidence Efficiency which leads to mispositioned lines of response.

Time

Figure 3

A digital integration method was also tested whereby the 16-samples were integrated and the normalized baseline was subtracted to find the pulse energy, where: § 15 · ¨¦N ¸ ¸−N PulseEnergy = ¨ 0 0 ¨ 16 ¸ ¨ ¸ © ¹

This method also shows reasonable results with a block energy resolution of again ~20% and a position profile peak to valley ratio of 2.2. While the digital integration does reduce signal noise caused by the 10ns uncertainty of the peak, it is still susceptible to pileup effects especially in the pulse tail. It was found that by integrating around the pulse peak using only N7+ N8 and not using the tail, many events that would be skewed due to pileup occurring in the tail are inherently rejected while still keeping the bulk of the energy information contained in the signal.

Figure 4 Coincidence Efficiency without Pileup Rejection

IV. RESULTS After testing several methods, the following was found to be a good balance between performance and FPGA resources: N7 + N8 2 N + N1 Baseline n = 0 2 PulseEnerg y = Peak − Baseline n Peak =

Figure 5 Detector Flood and Crystal Energy without Pileup Rejection

§ n · ¨ ¦ Baseline ¸ © n −8 ¹ Where : Baseline n < + Threshold 8

This method integrates samples at just the peak and leading baseline. As mentioned, integrating around the peak provides good signal to noise while not being as affected by pileup in the remainder of the pulse. A moving average of the baseline is also used to reject pileup if the current event’s baseline is greater than the moving average value by a set threshold. This method shows improvement in both energy resolution and event positioning over the other methods due to more accurate A, B, C, D energy calculation and pileup rejection, with a total block energy resolution of 18% and position profile peak-to-

4263

The Detector Flood image and Coincidence Efficiency in Figures 6 and 7 preserve the correct spatial position do not have a bias toward the middle of the detector when using the pileup rejection algorithm. An improvement can also be seen in the energy spectra by the reduction in pileup events improving the signal to noise ratio and detector energy resolution.

Figure 8 shows the calculated post pileup event rates match the observed rates collected from an Inveon™ Denicated PET scanner closely and validates that the FPGA implementation performs as expected. Figure 9 shows a close correlation between the observed rates and calculated rates ensuring that the pileup rates modeled in the deadtime correction should be accurate over a wide range of count rates .

Counts per Seconds

Inveon Detector Event Count Rate vs. Activity

_

2,000,000 1,800,000 1,600,000 1,400,000 1,200,000 1,000,000 800,000 600,000 400,000 200,000 0

CFDs Qualified Post Pileup Reject Calculated Post Reject

0

2

4

6

8

10

12

Activity (mCi)

Figure 6 Coincidence Efficiency with Pileup Rejection

Figure 8

Calculated vs. Observed Count Rate per Detector after Pileup Rejection 5 uCi to 10.5 mCi 1.200E+06

Observed Count Rate

1.000E+06 y = 1.0387x 2 R = 0.999 8.000E+05 Observed vs. Calculated Linear (Observed vs. Calculated)

6.000E+05

4.000E+05

2.000E+05

0.000E+00 0.000E+00 2.000E+05 4.000E+05 6.000E+05 8.000E+05 1.000E+06 Calculated Count Rate

Figure 7 Detector Flood and Crystal Energy with Pileup Rejection Figure 9

The effectiveness of the pileup rejection method was checked by calculating the estimated pileup rate and comparing the estimate to the observed rejected rates. The calculation used to estimate pileup must also be performed by the Inveon™ reconstruction software in order to accurately model the system deadtime correction. The probability that two or more events cause pileup event is given by the following equation [8].

V. SUMMARY The Event Processing Module has been designed to meet the challenges of event processing and data capture that the high countrates produced by the latest generation of PET systems demand. Using digital pulse processing methods, the EPM is able to obtain accurate energy and positioning resolution even at count rates well above what would be used in general operation.

x ( nτ ) e − nτ P ( x) =

x!

Where: x = Number of events occurring in the time window n = Counting rate (for the EPM the CFD rate) t = time window event is processed over for a nonparalyzable system

4264

REFERENCES [1] [2] [3] [4]

[5] [6] [7] [8]

D. Newport, et al., “QuickSilver™: a flexible, extensible, and highspeed architecture for multi-modality imaging,” presented at IEEE NSSMIC Conf. 2006, M08-1 B. J. Kemp, et al., “Performance Measurements of the Siemens Inveon Small Animal PET Scanner,” presented at IEEE NSS-MIC Conf. 2006, M14-69 R. A. Mintzer, S. B. Siegel , “Design and Performance of a New Pixellated-LSO/PSPMT Gamma-Ray Detector for High Resolution PET Imaging,” presented at IEEE NSS-MIC Conf. 2007, M18-142 B. Atkins, et al., “A Data Acquisition, Event Processing and Coincidence Determination Module for a Distributed Parallel Processing Architecture for PET and SPECT Imaging,” presented at IEEE NSS-MIC Conf. 2006, M11-62 B. K. Swann, et al., "A custom mixed signal CMOS integrated circuit for high performance PET tomograph front-end applications", IEEE Trans. Nucl. Sci., vol. 50, Issue 4, pp. 909-914, Aug. 2003. Analog Devices, “AD9218”, datasheet available on-lijne at http://www.analog.com Xilinx, “Virtex-II Pro Platform FPGAs”, datasheet available on-line at http://www.xilinx.com Glenn Knoll, “Radiation Detection and Measurement”, John Wiley and Sons Inc. 2000, page 637

4265