Towards photon counting X-ray image sensors B. Dierickx1,2, B. Dupont1, A. Defernez1, P. Henckes1 1
Caeleste CVBA, Generaal Capiaumontstraat 11, 2600 Antwerp, Belgium. Tel. +32478299757
[email protected] 2 Vrije Universiteit Brussel (VUB), Brussels, Belgium
Abstract The advantages of photon counting over charge integration, for medical X-ray imaging, are known. Yet the realization is hindered by technical and economical factors. The question that we try to answer is: what does it take to make a photon counting X-ray sensor? ©2010 Optical Society of America OCIS codes: (110.7440) X-ray imaging; (040.7480) X-rays, soft x-rays, extreme ultraviolet; (030.5260) Photon counting
1 Introduction Present state of the art medical X-ray imagers are all of the charge integrating type. Although in theory photon counting is the superior technique, photon counting X-ray imagers appeared only in a few high-end high added value applications. The key reason for that is that photon counting pixels and detectors are significantly more complex and expensive than integrating detectors. The questions that we try address in this paper are: is it worthwhile to pursue photon counting in medical X-ray, and what does it take to make a photon counting medical X-ray sensor? CERN [1-2] pioneered the possibility of monolithic Si photon/particle counting pixel detectors in nuclear physics, for high energy particles, under which also gamma and X-rays. For applications in medical X-ray one needs to use heavy detector materials, thus leading to direct detectors [3-7] or indirect detectors (scintillators). Hybridization of heavy direct detectors on Silicon poses the question of commercial viability of such devices. Manufacturable solutions may require direct detectors that can be deposited in a layers as amorphous Se, or as sheets as many indirect detectors or scintillators. 2
The Physics of detection
2.1
Direct and Indirect X-ray detection
X photon
X photon
100000
>100V
mean f ree path [af ter http://physics.nist.gov/PhysRef Data/ ]
Typical tissue or specimen thickness
Si (14)
10000
water Se (34) CsI
1000 Visible photons (fluorescence) Excited electron Electron-hole cloud
Charge collection By electric field
100
Realistic detector thickness
10
Si photodiode Si charge sensitive amplifier
Gd2O2S Bi4GeO12
MFP [µm]
Excited electron Electron-hole cloud
CdTe
1 1000
X-ray photon energy [eV] 10000
100000
1000000
Si charge sensitive amplifier
Medical X-ray range
↑Figure 1 Schematic comparison of the principle operation of a direct detector (left) and an indirect detector or scintillator (right). The physics of detection (absorption) is the same: an X-ray photon, by Compton scattering or Photoelectric Effect, excites an inner shell electron of a material atom. This primary electron has a considerable kinetic energy, which it looses by creating a trace of secondary electron-hole pairs. In a direct detector, one collects these electrons (or holes, depending on the applied polarity of the field) by an electrical contact to a charge sensitive amplifier in a ROIC (readout IC) underneath the detector. In an indirect detector, the secondary electrons decay back to their ground state, thereby emitting visible photons in random directions (“fluorescence”). Photons may be absorbed by the photodiode in the ROIC underneath the scintillator, and create there ternary electron-hole pairs that are sensed by the charge sensitive amplifier.
↑Figure 2 Mean free path (≈absorption length) of X-rays vs. photon energy, in the most popular direct detectors (Si, α-Se, CdTe) and scintillators (CsI, Gd2O2S). Also shown is the absorption in water, that is representative for biological tissue. Bi4GeO12 is shown too because this is the most efficiently absorbing scintillator material, consisting of the highest Z element (Bi) and having a high density (7.13 g/cm3).
2.2 Desired detector properties It is clear that indirect detection is much less efficient than direct detection in terms of overall conversion of Xphotons to effectively collected electrons. A medical X-ray detector should be an efficient absorber. This translates to material with high Z-number and high mass density.
10000 charge packet [e-]
All direct detectors have nearly same performance
CdTe InP Se Si
1000
Typical pulse shaper noise floor
Figure 3 Collected charge per X-photon, versus X-photon energy, for few of the most relevant direct and indirect detectors. For direct detectors this is a straightforward derivation. For the scintillator based detectors we made the quite optimistic assumption that the system has an overall quantum efficiency of 50%, from fluorescent light emission to charge collection in the visible light photodiode.
collected charge per X-photon
100000
GaAs CsI BGO 100
Scintillators using ideal 50%QE diodes
ZnSe Gd2O2S LaBr3 NaBiW2O8
10 1
10 X-photon energy 100
[keV]
1000
A second key performance criterion is the efficiency of conversion of X-photons to secondary (or ternary) electrons that are sensed in the readout circuit. Table 1 lists other performance parameters.
Table1: Direct detection versus Scintillation
Desired properties
Preferred materials MTF X-ray to electron conversion
Conversion and collection speed Physical nature
Safety Max detector thickness Energy separation (color X-ray) Main limit on DQE
3
Direct detection High Z (atom number) High specific density Low fluorescence High resistivity semiconductor High carrier mobility CdTe, CdZnTe, InP, Se, … Nearly perfect In the order of 1electron per 5eV Xray
Collection times and peak durations are in the order of few ns ⇒ Monocrystalline hybrids ⇒ Amorphous / polycrystalline deposition High applied bias voltage. Limited by detector material thickness only good Material thickness; thus close to 100% is possible
Indirect detection (Scintillator + Si photodiode) High Z (atom number) High specific density High fluorescence Transparent for visible light optical confinement e.g. by needle-like crystals CsI, LaBr3, Gd2O2S, LuOS, Bi4GeO12, HgI, ... Issues due to optical diffusion Overall eV-to-electrons is at least 10x worse than direct detection Best Light output: 1 visible photon per 12 to 30eV Transmission efficiency in scintillator: 10% to 50% QE of ROIC Si diodes: 40% to 90% CsI and Gd2O2S have 1µs decay time; several other scintillators are faster (10…100ns) ⇒ (poly-) Crystal sheet ⇒ Powder sheet
Limited by internal optical absorption and light diffusion Due to internal optical absorption, large fluctuation on packet size vs. energy Scintillators have a thickness limit related to internal light absorption and light diffusion affecting MTF.
The advantages of photon counting over charge integrating detectors
3.1 Color X-ray Although spectroscopic, multi-energy or “color” X-ray imaging is possible with classic charge integrating digital detectors [8-11], it requires multiple exposures, which may not be acceptable for reason of total dose or motion artifacts. In a photon counting device, each photon can be “weighted” and thus counted per energy range, without a total dose cost. 3.2 Sharpness recovery Information on the coincidence of pulses on neighbor pixels yields the real point of incidence of the X-photon [12].
3.3
Quantum limited performance ←Figure 4 Noise versus signal for X-ray pixels, comparing a charge integrating pixels and a photon counting pixel. Both signal and noise are expressed in “Xphotons”. The analog charge integration system is limited by read noise (here equivalent to 3 X-photons) and an overall achievable analog dynamic range (here 3000:1). A photon counting system does not have these limitations if one disregards counter depth limitation and counting speed saturation. comparator Charge
1000 noise (counter) noise (integration)
noise equivalent X-photons
100
packet Light flash
10
1 1
4
10
100
1000
10000 100000 1000000 signal [X-photons]
Pulse shaper
counter
MUX
reference
Analog V pulse train
Binary pulse train
Detector output
Figure 5 Typical photon counting pixel functional flow, which is reflected in circuit topologies.
Photon counting readout circuit limitations
4.1 Pixel Topology Most published counting pixel topologies [1-7; 12-15] correspond to the scheme in Figure 5. Pixel pitches (100µm to 500µm) depends on pixel complexity and CMOS technology (0.18µm to 0.8µm). With hybrid direct detectors, the charge packet contains several 1000s of electrons (Figure 3). With indirect detection, the pulse shaper becomes a critical part as charge packets are only a few 100 electrons large. 4.2 ,oise Apart from the inherent, device noise (thermal noise, MOSFET 1/f noise etc.) [14], a major designer concern is the electromagnetic interference noise, or the feedback of the digital part of the comparators, counters and the multiplexing to the extremely sensitivity pulse shapers [15]. 4.3 Yield Photon counting architectures have significantly more complex pixels than charge integrating solutions. Whereas passive and active pixels have 1 up to 7 transistors per pixel, a digital photon counting pixel has several hundreds [17], of which the largest part is the digital counter. We expect Si manufacturing yield and circuit design techniques to improve in the medium future to allow such arrays to be manufactured in reasonably large arrays (larger than 1dm2) with good yield. 4.4 Counter speed saturation Counter speed limited by: scintillator decay time, photodiode charge collection time and the readout circuit speed. References [1] R. Ballabriga, M. Campbell, E. H. M. Heijne, X. Llopart, and L. Tlustos, “The Medipix3 Prototype, a Pixel Readout Chip Working in Single Photon Counting Mode With Improved Spectrometric Performance”, IEEE Trans Nuclear Science, vol.54, no.5 (2007) [2] X. Llopart, M. Campbell, R. Dinapoli, D. San Segundo, and E. Pernigotti, “Medipix2: a 64-k Pixel Readout Chip With 55-µm Square Elements Working in Single Photon Counting Mode”, IEEE Trans. Nucl. Sci., vol.49, no.5, Oct.2002 [3] P. Pangaud et al. “XPAD3: A new photon counting chip for X-Ray CT-scanner”, Nuclear Instruments and Methods in Physics Research, Vol 571, Issues 1-2, 2007, Pages 321-324 [4] K. Spartiotis & al.“A photon counting CdTe gamma- and X-ray camera”, Nuclear Instruments and Methods in Physics Research, vol.550, p.267-277, Sept 2005, energy optimization for chest imaging”, Med. Phys. 34 (10), Oct. 2007 [9] T. Asaga, C. Masuzawa, A. Yoshida, et al. ”Dual-energy subtraction mammography”. J Digit Imaging 1995 vol.8 p.70-73 [10] SC Kappadath, CC Shaw. “Dual-energy digital mammography for calcification imaging: noise reduction techniques”, Phys Med Biol. 2008 Oct 7;53(19):5421-43. Epub 2008 Sep 2. [11] B. Dierickx, N. Buls, C. Bourgain, C. Breucq, J. Demey, B. Dupont, A. Defernez, “On the diagnostic value of multi-energy X-ray imaging for Mammography”, European Optical Society symposium, Munchen, 16-18 June 2009 [12] B. Dierickx, B. Dupont, A. Defernez, “X-ray image sharpening by coincidence detection”, IISW, Bergen, 26-28 June 2009 [13] M. Perenzoni, D. Stoppa, M. Malfatti, A. Simoni, “A Multi-Spectral Analog Photon Counting Readout Circuit for X-Ray Hybrid Pixel Detectors”, IMTC 2006, Sorrento, Italy 24-27 April 2006 [14] C. Lotto and P. Seitz, “Charge Pulse Detection with Minimum Noise for Energy-Sensitive Single-Photon X-Ray Sensing”, European Optical Society symposium, Munchen, 15 June 2009 [15] J. Lundgren, S. Abdalla, M. O´Nils, B. Oelmann, “Evaluation of Mixed-Signal Noise Effects in Photon Counting X-Ray Image Sensor Readout Circuits”, 2005
