Amorphous Silicon Detectors

8.19

Amorphous Silicon Detectors

W Zhao, State University of New York at Stony Brook, Stony Brook, NY, USA JA Rowlands, Thunder Bay Regional Research Institute, Thunder Bay, ON, Canada ã 2014 Elsevier B.V. All rights reserved.

8.19.1 8.19.1.1 8.19.1.2 8.19.2 8.19.2.1 8.19.2.1.1 8.19.2.1.2 8.19.2.1.3 8.19.2.2 8.19.2.2.1 8.19.2.2.2 8.19.2.2.3 8.19.2.2.4 8.19.3 8.19.3.1 8.19.3.2 8.19.3.2.1 8.19.3.2.2 8.19.3.2.3 8.19.3.3 8.19.3.3.1 8.19.3.3.2 8.19.4 8.19.4.1 8.19.4.2 References

Amorphous Silicon Technology Is Driven by Consumer Large Area Flat-Panel Displays Amorphous Silicon Thin Films Process for Making Amorphous Silicon Integrated Circuit Principle of Operation for Amorphous Silicon Flat-Panel Imagers Indirect Detectors x-Ray scintillator Design variations in amorphous silicon optical-sensing elements Pixel x-ray sensitivity for indirect AMFPI Direct Detectors Detector structure Design considerations for HV protection of TFTs Pixel x-ray sensitivity for direct AMFPIs Other x-ray photoconductors Evaluation of Imaging Performance Image Correction Spatial Frequency Domain Image Quality Metrics Spatial resolution: MTF Noise power spectra Detective quantum efficiency Temporal Performance of Different x-Ray Detector Technologies Temporal performance of indirect AMFPIs Temporal performance of direct AMFPI Emerging Detector Technology Increasing Gain Decreasing Electronic Noise: APSs

Glossary AMFPI (active matrix flat-panel imager) It is a large-area x-ray imaging detector based on amorphous silicon technology. Direct detection Detection of x-rays using photoconductive materials to convert x-ray energy directly to electronic charge. DQE (detective quantum efficiency) It is defined as the ratio between the signal-to-noise ratio (SNR) squared at the output of a detector to that at the input. It reflects the detector’s ability to utilize the incident x-ray photons and is

This chapter reviews the physics of different types of amorphous silicon detectors and their imaging performance. Emphasis is placed on the detector technologies and the important factors affecting the imaging performance for diagnostic x-ray imaging.

8.19.1 Amorphous Silicon Technology Is Driven by Consumer Large Area Flat-Panel Displays The rapid advancement and expansion of amorphous silicon (a-Si) manufacturing facility in the past two decades have been

Comprehensive Biomedical Physics

315 316 316 316 317 317 317 319 319 319 320 321 321 322 323 323 323 323 324 325 325 326 327 327 327 328

used as the gold standard for comparing the physical performance of different x-ray imaging detectors. Indirect detection Detection of x-rays using scintillators to convert x-ray energy to optical photons, which are subsequently converted to electronic charge using optical sensors. MTF (modulation transfer function) Describes the ability of an image sensor to detect information at different spatial frequencies. NPS (noise power spectrum) Plots noise power density of images as a function of spatial frequency.

driven by the increasing consumer demand for active matrix liquid crystal displays (AMLCDs). This exponential growth in production has increased the yield and driven down the cost, which led to the widespread availability of this technology for a number of applications. The underlying technology for AMLCDs is large-area integrated circuits based on a-Si semiconductor thin film deposited uniformly across glass substrates with sizes up to 2.2
2.5 m. The thin films of metal layer, insulator, and a-Si semiconductor are delineated using a photolithographic process to make a two-dimensional (2D) array

http://dx.doi.org/10.1016/B978-0-444-53632-7.00620-1

315

316

Amorphous Silicon Detectors

of thin film transistors (TFTs). This technology provides a unique opportunity for making monolithic large-area sensors for x-ray imaging, which requires sizes of up to 43
43 cm. Coupling traditional x-ray detection materials such as phosphors or photoconductors with a large-area active matrix readout structure forms the basis of x-ray active matrix flat-panel imagers (AMFPIs).

8.19.1.1

Amorphous Silicon Thin Films

The majority of today’s active matrix TFT arrays are fabricated from hydrogenated amorphous silicon (a-Si:H). Research into the properties of a-Si grew from the interest in chalcogenide amorphous semiconductors pursued in the 1950s and 1960s. The development of photovoltaic devices in the late 1970s was led by the discovery of the beneficial effects of introducing hydrogen to occupy many of the dangling bonds present in a-Si and the ability to control the electrical properties of the material by doping. Complex integrated circuits using doped a-Si:H, such as TFTs, were first demonstrated in the early 1980s, which led to the development of the current large-area active matrix TFT arrays which have electrical properties suitable for display and sensor applications. The most common approach for the deposition of a-Si:H thin films is plasma-enhanced chemical vapor deposition, which deposits a-Si:H onto a glass substrate from silane gas (SiH4) in a plasma reactor at temperatures of 250  C. The morphology of layers fabricated in such a manner is one of short-range order and long-range disorder, hence the use of the term amorphous. This disordered structure is responsible for the exceptional radiation damage tolerance that a-Si:H exhibits (Boudry and Antonuk, 1996). It is also the source of charge trapping, associated with the so-called metastable states created by this morphology. Such amorphous materials have many similarities to perfect crystals, for instance, the presence of energy states throughout the material, that is, the existence of bands and, more surprisingly, the presence of a forbidden gap. The long-range disordered nature of a-Si:H results in the creation of states in this bandgap that would be forbidden in a perfect crystal. These bandgap states control many of the electrical properties of a-Si:H. Variation of the deposition parameters (e.g., temperature or plasma frequency) can change the morphology of the material and the defect density in the bandgap. This results in a variety of materials such as nano-, micro-, and polycrystalline silicon which differ in the range over which the crystallite structure remains ordered before breaking down into the disordered structure characteristic of amorphous materials. These morphological differences greatly affect the electrical properties of the final material, for example, the mobility of the charge carriers when an external electric field is applied. Good-quality amorphous silicon has carrier mobilities of 1 and 0.003 cm2 V1 s1 for electrons and holes, respectively. These are several orders of magnitude less than the mobilities (1300 and 500 cm2 V1 s1, respectively, for electrons and holes) typically obtained with crystalline Si. Other properties, such as the creation of metastable states under illumination and current flow, can also cause temporal variations in the electrical behavior of the amorphous layers. During the deposition process, different gases may be introduced into the plasma to control the level of doping in the a-Si:H layer in

much the same way doping impurities control the electrical properties of standard crystalline silicon components. (For more details on amorphous silicon science and technology, refer to the monographs by Street 1991), (Kanicki 1992), and (Kuo 2004).

8.19.1.2 Process for Making Amorphous Silicon Integrated Circuit The process of forming the active matrix array is similar to techniques developed for crystalline silicon integrated circuits. It entails deposition and photolithographic etching of thin layers of metals, insulators, and a-Si:H with different doping concentrations. Depending on the complexity of the pixel design, as few as four or as many as a dozen different layers may be required to complete an array.

8.19.2 Principle of Operation for Amorphous Silicon Flat-Panel Imagers A large-area active matrix consists of a 2D array of a-Si:H TFTs, each of which functions as an electronic switch (Rowlands and Yorkston, 2000). Depending on the x-ray detection materials, AMFPIs are divided into two main categories: ‘direct’ and ‘indirect’ detection. As shown in Figure 1(a), direct detection AMFPIs employ a uniform layer of x-ray sensitive photoconductor, for example, amorphous selenium (a-Se), to convert incident x-rays directly to charge, which are collected by the pixel electrodes. (Zhao and Rowlands, 1995). Figure 1(b) shows that indirect AMFPI uses an x-ray scintillator such as structured cesium iodide (CsI) to convert x-ray energy to optical photons, which are then converted to charge by integrated photodiodes (PDs) at each pixel of the TFT array (Antonuk et al., 1992). Both direct and indirect AMFPIs use the same readout scheme: the scanning control circuit turns on the TFTs one row at a time, and transfers image charge from the pixels to external chargesensitive amplifiers, which are shared by all the pixels in the same column. The readout rate is dictated by the pixel RC time constant, where R is the on-resistance of the TFT and C is the capacitance for the pixel-sensing element. For complete readout of charge, the TFT needs to be ON for at least Ton ¼ 5RC. With nominal values of R ¼ 4 MO and C ¼ 1 pF, each row of the AMFPI would require 20 ms to read out. In addition to the time required for charge transfer, the parallel to serial conversion of the digitized signal also necessitates an overhead. Hence, a detector with 1024
1024 pixels can be read out in real time (i.e., 30 fps (frames per second)). A faster readout rate, for example, 60 fps, may be possible by binning the pixels, that is, switching on more than one pixel at a time to reduce the image matrix. Both direct and indirect conversion AMFPIs have been commercialized for a wide variety of clinical x-ray imaging applications, including radiography, fluoroscopy, and conebeam computed tomography (CBCT). The direct method has the advantages of higher image resolution and simpler TFT array structure that can be manufactured in a standard facility for AMLCDs. The indirect method has the advantage of higher atomic number of Cs and I compared to a-Se.

Amorphous Silicon Detectors

x-Rays

317

x-Rays

HV bias electrode Structured Csl

Amorphous selenium

l

ol ntr

nin

an

Sc

Mult

S

tro

on

o gc

c ing

n

an

G D

Sc

Storage capacitor

TFT

TFT Pixel electrode

iple

xer

Mul

tiple

(a)

Photodiode

xer

(b)

Figure 1 Diagram showing the concept of AMFPI with direct and indirect x-ray conversion: (a) direct detector uses an x-ray photoconductor (e.g. a-Se) to convert x-rays directly to charge; and (b) indirect detector uses a phosphor screen or structured scintillator to first convert x-rays to optical photons, which are then converted to charge by an integrated photodiode at each pixel of the detector.

Charged surface

Structured CsI columns +

+

+

+

+

+

+

1.0

+

QE (a-Si PD)

+

(a)

Columnar CsI

(b)

Photoconductor

Figure 2 Image resolution of indirect and direct conversion x-ray detection materials: (a) columnar structures of CsI(Tl) helps channeling light in the forward direction, thus providing better resolution than powder phosphor screens; (b) The applied electric field in photoconductors draws x-ray generated image charge directly to surfaces without lateral spread.

8.19.2.1

Indirect Detectors

8.19.2.1.1 x-Ray scintillator Indirect AMFPIs have been manufactured by major medical imaging equipment vendors for different clinical applications. The most advanced form of detector construction uses thallium (Tl)-doped CsI with columnar (needle-like) structure as the x-ray scintillator. As shown in Figure 2(a), the columns in CsI help channel light photons, which are generated by x-ray interaction, in the forward direction. Although the light guidance is not as perfect as in fiber optics with smooth walls, columnar CsI provides much better imaging performance than powder phosphor screens (Rowlands and Yorkston, 2000; Zhao et al., 2004). The CsI (Tl) used in AMFPI is less hygroscopic than the CsI (Na) layers used in x-ray image intensifiers (XRIIs), and its optical emission spectrum (green) is a better match to the spectral response of a-Si PDs, as shown in Figure 3 (Rowlands and Yorkston, 2000). For lower cost detectors, which are mainly used for general radiography, powder phosphors such as gadolinium oxysulfide (GOS) have also been incorporated. There are also GOS screens optimized specifically for AMFPIs, which differ from those used in screen films in the sizes and spatial distribution of phosphor grains.

Quantum efficiency

0.8

0.6 Scintillator emission

0.4 CsI (Na)

0.2

0.0 300

CsI (Tl)

400

500 600 Wavelength (nm)

700

800

Figure 3 Optical quantum efficiency of a-Si photodiode as a function of wavelength of light. Plotted in comparison are photon emission spectra for three types of x-ray scintillators: structured CsI (Na) and CsI (Tl) (graph reproduced from Antonuk LE, Boudry J, Huang W, et al. (1992) Demonstration of megavoltage and diagnostic x-ray imaging with hydrogenated amorphous silicon arrays. Medical Physics 19: 1455–1466.)

8.19.2.1.2 Design variations in amorphous silicon optical-sensing elements An active matrix array designed specifically for medical imaging requires not only the TFT but also an image charge-sensing element. A number of factors can influence the choice of pixel design, some related to imaging performance and others not. Perhaps the most important nonimaging factor is the fabrication yield (the fraction of devices that is useful). The desire to increase the yield, and hence reduce the cost of the array, pressures the designer to reduce the complexity of the pixel elements; the number of pixel elements; and the number of mask layers required for fabrication. This provides impetus to compromise the individual elements by fabrication of both the switching and sensing/storage elements at the same time even

318

Amorphous Silicon Detectors

1.0

x-Rays

ITO

(a)

a-Si:H photodiode

p I n

TFT

Quantum efficiency

0.8

Scintillator

0.6

0.4

0.2

x-Rays

0.0

Scintillator

Passivation

a-Si (b)

TFT

n Insulator

Bias a-Si MIS-type sensor

Figure 4 Cross section of different PD configuration.

if this yields suboptimal performance. While these modifications may improve device yield, it can be at the cost of reduced imaging performance of the final detector. Intellectual property considerations can also influence the pixel design, as can experience with a particular fabrication process. For example, flat-panel display manufacture typically does not require pþ deposition or thick (i.e., 2 mm) intrinsic layer processes. These additional processes may make the fabrication of a-Si:H PDs difficult at foundries designed specifically for display manufacture. Consequently, in practice, optimal imaging performance of the pixel elements may not be the ultimate consideration in the design of the array for use as a flat-panel x-ray detector. The most common type of a-Si optical-sensing element is the a-Si PD. As shown in Figure 4(a), each a-Si PD has a p–i–n multilayer structure, where the thin p- and n-doped layers are reverse biased to block injection of electrons and holes, respectively, from the bias electrodes into the intrinsic layer. The top ITO bias electrode is connected to a negative potential, typically 5 V. Light photons enter the intrinsic layer through the passivation layer, the ITO, and the p layer. Each of these layers absorbs some light, thereby reducing the absorption in the intrinsic layer, hence quantum efficiency (QE) of the PD. Charge generated in the p- and n-layers does not contribute to the signal due to the very short drift lengths of minority carriers. Light absorbed in the intrinsic layer generates electron– hole pairs, which are driven to the n- and p-contacts. Figure 3 shows the QE as a function of photon wavelength in the visible range for a 1.5-mm-thick p–i–n PD at  5 V reverse bias. The periodicity is due to the interference effects in the top passivation layer, and the material and thickness for this layer can be chosen to have antireflective properties for certain wavelengths. The lower QE at longer wavelengths is due to the decrease in absorption coefficient of intrinsic a-Si, and the decrease in QE at shorter wavelengths is due to increased absorption in the p-doped layer. Figure 3 also shows the emission spectra for typical x-ray phosphors. It can be seen

600 mm columnar CsI 1000 mm a-Se 0

20

40

100 60 80 x-Ray energy (keV)

120

140

Figure 5 Quantum efficiency (QE) Z as a function of x-ray photon energy for two materials (a-Se and columnar CsI) used in x-ray imaging detectors: Thickness of layer is 1000 mm for a-Se and 600 mm for CsI.

that the spectral response of a-Si:H PD is well matched to the emission from these phosphors. Another type of a-Si:H photosensing element has been developed using an insulating layer in place of the p-doped layer in a p–i–n PD. This is called a metal–insulator–semiconductor (MIS) structure (Kameshima et al., 1998) and is shown schematically in Figure 4(b). Incident light passes through the n-layer resulting in the movement of holes as the main component of the signal current. These holes accumulate at the interface between the i-layer and the insulator, inducing charge in the lower metal electrode that is connected to the TFT. A refresh cycle is necessary with this design to remove the photogenerated holes at this interface after the signal has been read out. This is achieved by reversing the polarity of the bias voltage applied to the upper metal contact. The front n-layer is placed in a forwardbiased condition and electrons are injected into the a-Si:H layer to neutralize the photogenerated holes. This design has the advantage of not requiring a p-doped a-Si:H layer and is thus compatible with the fabrication process for TFT arrays in AMLCDs, which only requires n-doping. Furthermore, in this design the MIS sensor was fabricated at the same time as the TFT, resulting in fewer lithographic processing steps. In both approaches described above, the optical-sensing element is built side by side with the switching element. Although the manufacturing complexity is relatively low by minimizing the number of photolithographic processes, the fraction of pixel area occupied by the sensing element (fill factor) is not optimal. Figure 6(a) shows the micrograph of the pixel designs for an indirect AMFPI with 127 mm pixel size (Weisfield et al., 2004). The TFT is at one corner of the pixel. The gate lines, data lines, and the PD occupy the rest of the space. This is the typical design used in most commercial indirect AMFPI. Since the space taken by the TFT and the lines does not change as a function of the pixel pitch d, the fill factor fp decreases rapidly as d decreases (Rowlands and Yorkston, 2000). The value of fp is 0.57 for d ¼ 127 mm. Advanced sensor structures have been proposed to increase the fill factor. One method is shown in Figure 6(b), where the PD is built on top of the other circuit elements (e.g., TFT, gate line and data line) at each pixel (Powell et al., 1997; Weisfield et al., 2004). With this

Amorphous Silicon Detectors

Gate line TFT

Data line

57% FF

Bias line

(a)

Bias line over Gate line

TFT Data line

85% FF

absorbed x-ray; (3) optical QE of the a-Si PD; and (4) pixel fill factor. The thickness of CsI, dCsI, used in indirect AMFPI varies depending on the clinical application. For the relatively high x-ray energy (>70 kVp) used in clinical CBCT, dCsI is typically 600 mm (Jaffray and Siewerdsen, 2000). Columnarstructured CsI layers have lower density compared with single crystals. The packing density could vary depending on the deposition procedures; however, the widely quoted value is 75% which results in a density r of 3.38 g cm3. Shown in Figure 5 is the QE
of a 600 mm CsI layer as a function of x-ray photon energy. The k-edge of Cs (35 keV) and I (33 keV) creates a boost for
over the energy range typically used for CBCT. With RQA5 spectrum (70 kVp Tungsten spectrum with 21 mm of added Al filtration),
with the above detector parameters is 0.84. As shown in Figure xx, the optical QE of a-Si PDs is 0.7 for the green light emitted from CsI (Tl). The reported conversion gain of CsI for indirect AMFPI is 25 eV photon1, which results in gc ¼ 2000 for a 50 keV x-ray photon. The fill factor fp depends on the pixel pitch d and the pixel design of the a-Si PD-TFT array. With all factors considered, the overall gain of an indirect AMFPI is 1000 electrons per 50 keV x-ray photon with a nominal fill factor of fp ¼0.7. The main disadvantage of AMFPI compared to crystalline Si detector technology, such as complementary metal oxide semiconductor (CMOS) or CCD, is its higher electronic noise, which is dominated by the noise associated with the pixel reset and the charge amplifier (Weisfield and Bennett, 2001). The nominal value for the pixel electronic noise is 1500 electrons, which is higher than the number of light photons captured by the PD for each absorbed x-ray in CsI. The excessive electronic noise would lead to degradation in low-dose imaging performance, especially for high spatial frequency information, which is already compromised by the image blur in CsI.

8.19.2.2 (b)

Figure 6 Micrograph showing the top view of a single pixel of two different indirect flat-panel designs: (a) side-by-side TFT and photodiode; and (b) photodiode on top of TFT. Reproduced from Weisfield RL, Yao W, Speaker T, Zhou K, Colbeth RE, and Proano C (2004) Performance analysis of a 127-micron pixel large-area TFT/photodiode array with boosted fill factor. Proceedings of SPIE 5368: 338–348.

design the fill factor for the same pixel size of 127 mm is increased to fp ¼ 0.85. In principle, the fill factor can be increased to unity by building a continuous a-Si:H PD with common bias electrode on top of the entire TFT array, and some preliminary work has been conducted to demonstrate its feasibility. The main challenges for this approach are the increased line capacitance and the image blooming effect due to charge crosstalk between pixels (Weisfield et al., 2004).

8.19.2.1.3 Pixel x-ray sensitivity for indirect AMFPI The overall pixel x-ray sensitivity of an indirect AMFPI depends on four factors: (1) x-ray QE
of the scintillator; (2) inherent x-ray to optical photon conversion gain gc, that is, the number of optical photons emitted from the scintillator for each

319

Direct Detectors

8.19.2.2.1 Detector structure In the direct detection approach, the energy absorbed in the photoconductor through x-ray interaction is converted directly to charge, as shown in the conceptual side view in Figure 7(a). Consequently, the sensing element is typically a pixel charge collection electrode, which is made using a thin metal layer and extended to cover as large an area as possible within a pixel. These pixel electrodes are connected electrically to a charge-storage capacitor and the TFT, as shown in the schematic diagram of Figure 7(b). A continuous electrode is applied to the upper surface of the photoconductor to allow the application of an external bias voltage. The fabrication of the storage capacitor only requires the standard metal and insulator layers with standard AMLCD manufacturing processes, thus simplifying production and lowering cost (den Boer et al., 1998). The operational requirements of the photoconductive layer are similar to those of the PD in the indirect approach in that a bias voltage must be applied across the thickness of the photoconductor to facilitate the separation and collection of the charge carriers produced by x-ray interaction. The most highly developed x-ray photoconductor is a-Se which is currently used in all commercial direct AMFPIs. To maintain an internal field

320

Amorphous Silicon Detectors

+HV

x-Rays +HV

Se

+

-

Photoconductor + VP

TFT

+

Pixel electrode

(a)

CP

TFT (b)

+HV

+HV Insulator

-

-

-

Se

Se

+

+

TFT

VP

TFT

VP

-

+

+

CP

CP

Voltage regulator

(c)

Vbias

(d)

-HV

+HV Se

+ +

-

Se

Top gate

VP

-

VP

+

TFT

+

CP

CP TFT (e)

(f)

Figure 7 Schematic drawings showing the different detector configurations for direct conversion AMFPI: (a) side view of single pixel; (b) schematic diagram for the structure shown in (a); (c) detector implemented with an additional insulator layer on top of a-Se; (d) Schematic diagram of a pixel with additional voltage regulation element to prevent HV damage; (e) Pixel design using the top gate of the same TFT for HV protection; (f) Using negative potential on the top bias electrode.

of 10 V mm1 within a 1000-mm-thick layer of a-Se, a bias voltage of 10 000 V must be applied. This high voltage (HV) necessitates blocking contacts to prevent charge injection. One type of thin blocking layers contains a large number of deep traps. These traps, when filled, permit a rapid decrease in the field at the interface between a-Se and metal contact, resulting in minimal charge injection (Kabir et al., 2008).

8.19.2.2.2 Design considerations for HV protection of TFTs Without careful design it is possible for the HV bias of a-Se, for example, 10 000 V, to appear across the TFT and storage capacitor of each pixel. This would result in permanent damage to the active matrix array, for example, through dielectric breakdown of storage capacitor and/or gate insulator of the TFT.

Amorphous Silicon Detectors

When HV is first applied to the panel, the voltage drop is distributed between the pixel storage capacitance and the a-Se pixel capacitance, which are in series, as shown in Figure 7(a). This configuration is a potential divider with the larger potential occurring across the smaller capacitance. Fortunately, the capacitance of a 1000-mm-thick layer of a-Se is small (2.5 pFcm2) so that an individual pixel of 200
200 mm has a capacitance of 1 fF. If the storage capacitance is designed to be 1 pF, the potential at the pixel electrode is only 10 V with the other 9990 V dropped across the a-Se layer. Thus, the components on the active matrix panel can be protected in a purely passive manner from the application of the high potential. However, if the TFT array is left without scanning, dark current or signal current from a-Se will cause the potential on the pixel electrode to rise toward the HV bias. Unless some preventive measures are taken, this voltage increase will eventually damage the active matrix circuit. Two general approaches have been implemented to prevent the voltage at the pixel electrode reaching damaging values, which is usually a few tens of volts with typical a-Si:H TFT designs. The first approach, as shown in Figure 7(c), is to include an extra dielectric layer between a-Se and the HV bias electrode (Lee et al., 1995, 1996). In this approach the potential drop is redistributed across the dielectric layer rather than in the storage capacitor as the potential across the a-Se layer collapses. However, this protection results in increased readout complexity which has to eliminate the trapped image at the a-Se/dielectric layer before subsequent exposures. This is achieved by removal of the applied bias voltage and flooding the detector with light to generate charge in a-Se that flows in the opposite direction to that generated by the x ray. The requirements of a refresh cycle make this approach to HV damage protection incompatible with fluoroscopic (i.e., real-time readout) applications. The second approach is to maintain real-time imaging capability while providing HV protection. The key pixel design is to include means to drain away charge on the pixel if the pixel potential exceeds a predetermined safe value. Three possible designs have been proposed: (1) To include an extra circuit element at each pixel in parallel with the storage capacitor for voltage regulation, as shown in Figure 7(d). This device can be made using the standard TFT process, with a separate bias line to provide a path to bleed excess charge from the pixel electrode (Tsukamoto et al., 1999). (2) To modify the TFT design by incorporating a second (top) gate, which is connected to the pixel electrode as shown in Figure 7(e) (Lehnert and Zhao, 2006; Zhao et al., 1998). When the pixel potential approaches damaging levels, the top gate will increase channel current to drain away the excess charge. Since the excess charge is bled away along the readout (data) lines, the timing of the electronic scan needs special consideration to avoid corruption of image information of pixels sharing the same readout line with overexposed pixels. (3) To apply a negative bias on the top electrode of a-Se to ensure that the ordinary TFT will start to conduct when the pixel potential reaches the threshold value, as shown in Figure 7(f) (Polischuk et al., 1999). This approach requires the same consideration for proper timing of the electronic scan because excess charge from a single pixel could potentially corrupt the image data from all the pixels sharing the same data line. Although the simplicity of this design is attractive, the reversed

321

a-Se structure makes it more susceptible to charge trapping (Manouchehri et al., 2008). This is because the electrons, which have much lower mobility than holes, have to travel a longer distance before reaching the pixel electrodes. As can be seen, the basic design of the sensing/storage element for a direct detector is straightforward and compatible with the manufacturing process for TFT arrays in AMLCDs. However, additional consideration for detector design is required to ensure the safe operation of TFT array in the presence of the HV bias. With the rapid advancement in TFT manufacturing process and increased yield for AMLCDs, the design complexity associated with direct AMFPIs is no longer a concern.

8.19.2.2.3 Pixel x-ray sensitivity for direct AMFPIs The most highly developed x-ray photoconductor is a-Se, which is being used in all commercial direct AMFPIs. Because of its lower atomic number than CsI, as shown in Figure 5, the thickness dSe is 1000 mm for most clinical applications except mammography, where dSe of 200 mm provides essentially complete absorption of mammographic spectral energies. The density of a-Se is 4.27 g cm3, lower than that for crystalline selenium. For an RQA 5 x-ray spectrum,
is 0.77 for dSe ¼1000 mm. The x-ray to charge conversion gain of a-Se is inversely proportional to the energy required togenerate an electron–hole pair, W. The value for W in most single-crystal solid-state photoconductors, for example, Si, Ge, and CdTe, follows Klein’s relation, that is, 3 times the bandgap energy. In a-Se, W depends on the electric field ESe. The nominal value is W¼ 50 eV at ESe ¼10 V mm1 (Rowlands et al., 1992). The geometric fill factor for direct AMFPI is high because the pixel electrode is built on top of the TFT and the gate and data lines. In addition, the image charge collection in a-Se is governed by the electric field. Because the field lines in the gap between pixels bend toward the pixel electrodes, the image charge created in this region can also be collected (Pang et al., 1998). This leads to an effective fill factor of unity, which has been confirmed experimentally from direct AMFPIs (Zhao et al., 2003). Compared to indirect AMFPI, a-Se direct detectors have approximately the same x-ray conversion gain (1000 electrons per incident 50 keV x-ray photon) and electronic noise. Hence, they share the same advantages and limitations in low-dose imaging performance as the indirect AMFPI. One of the advantages of the direct AMFPI compared to the indirect one is the ability to make smaller pixels because of its simpler array structure (no need for the PDs) and the unity fill factor which is independent of pixel size.

8.19.2.2.4 Other x-ray photoconductors Despite the overwhelming success of a-Se in AMFPI, it has two shortcomings: One is the high electric field required to achieve a reasonable conversion gain, for example, W ¼ 50 eV at 10 V mm1 and the other is the rather low atomic number, Z ¼ 34, which requires very thick layers (e.g., 1000 mm) for high QE at diagnostic energies (100 keV). Other photoconductors with higher atomic numbers and/or conversion gains have been investigated. Table 1 lists a few examples and a comparison of their properties with a-Se. Some of these photoconductors, for example, Cd(Zn)Te, HgI2, and PbI2, were first investigated in their single-crystalline forms as nuclear radiation detectors. Cd(Zn)Te single crystals are currently under intensive investigation for single photon counting detectors in nuclear medicine (SPECT and PET) imaging and computed tomography (CT). For application in

322

Table 1

Amorphous Silicon Detectors

Material properties of several x-ray photoconductors

Material

Z

D (g cm3)

E (V mm1)

Eg (eV)

W (eV)

Idark (nAcm1)

a-Se

34

4.27

2.2

PbI2

82/53

42 20 5*

0.01 0.1 2

PbO TlBr HgI2

82/16 81/35 80/53

4 1 1–2

1.9 2.7 2.1

48/52

0.5

1.5

9–20 6.5* 5* 8–20 4.5–5*

1 10

Cd(Zn)Te

6.2* 3.1 9.8 7.56 6.4* 3.2–5.7 6.2*

10 30 2

2.3

9–50

Values for thick film materials were quoted if available; otherwise, * represents single-crystal values. Z is the atomic number of the x-ray absorbing elements, D is the density, E is the applied electric field, Eg is the bandgap energy, W is the energy required to release an electron–hole pair, and Idark is the dark current at the given E. All materials (with the exception of a-Se) are polycrystalline films, and their properties are generally less known than those for single crystals, and the material properties vary widely with deposition methods.

Table 2

Pixel size (mm) Range of detector exposures (mR) Detector mean exposure (mR) Detector area (cm
cm) Typical x-ray spectrum (kVp) Frame rate (frames s1)

General radiography

Fluoroscopy

Mammography

130–200 0.03–10 0.3 35
43 70–120 0.2 5 (tomosynthesis)

400–500 0.0001–0.01 0.001 Up to 40
40 70 30

70–100 1–1000 12 Up to 24
30 28 0.05 2 (tomosynthesis)

large-area AMFPI, polycrystalline thin films have been formed through physical vapor deposition (PVD) (Shah et al., 1999). Polycrystalline PbO thin films have been used as the photoconductive target in an optical vidicon since the 1950s (Heijne et al., 1954). A large-area (800 diameter) x-ray vidicon was made in 1956 using a 150-mm-thick layer of PbO, which has a p–i–n structure (Jacobs, 1956). It was difficult to manufacture because PbO reacts with ambient air, causing dark current to increase and x-ray sensitivity to decrease. More recently, PbO thick films have been deposited on direct conversion-type TFT arrays to make prototype flat-panel detectors (Simon et al., 2005). TlBr thick films have also been used for prototype large-area vidicons, which was cooled with a Peltier cooler to reduce the dark current to a reasonable level (Ouimette et al., 1998). Polycrystalline films of Cd(Zn)Te, PbI2, and HgI2 have also been developed and investigated intensively for making direct AMFPIs (Adachi et al., 2000; Kang et al., 2005; Street et al., 2002a; Tokuda et al., 2001). In addition to PVD, particle-inbinder screen printing method has also been developed as a cost-effective alternative for depositing x-ray photoconductors (Choi et al., 2007; Du et al., 2008). Studies have shown that the trap-free limit of W of polycrystalline thick films approaches Klein’s theoretical values (Street et al., 2002b). However under moderate bias field (1 V mm1), which keeps dark current at a manageable level, charge trapping of the slower charge carriers is a serious problem. It results in reduced sensitivity (i.e., higher effective W), decreased modulation transfer function (MTF) at high spatial frequencies, lower detective QE (DQE), and ghosting artifacts (Kabir, 2008; Kabir and Kasap, 2002). This problem needs to be overcome before commercialization for regular clinical use.

8.19.3

Evaluation of Imaging Performance

Several international standards (IEC and AAPM Task group) have adopted image quality metrics expressed in the spatial frequency domain to evaluate the image quality of projection x-ray images. These image quality metrics, which include MTF, noise power spectrum (NPS), and DQE, are reviewed here. Published measurements of commercial detectors will be used as examples for discussion. AMFPIs have been developed for a wide range of clinical imaging applications, which include general radiography, mammography, radiography and fluoroscopy (R/F) mixed mode operation. They have also been used extensively in CBCT and digital tomosynthesis, where rapid image acquisition is required. Table 2 lists the typical range of AMFPI detector design parameters for existing AMFPIs. The maximum exposure encountered in radiography and mammography is the unattenuated radiation, and the minimum is behind the bone of the thickest part of the body, or the dense breast tissue in mammography. Mammography has the highest requirement for spatial resolution. R/F mixed mode detectors have the highest requirement for detector dynamic range and readout rate. To accommodate the different pixel size requirement in fluoroscopy and radiography, pixel binning is implemented. Because of the flexibility in acquisition speed and spatial resolution, this type of detector is usually used for CBCT application. Table 3 shows example detector parameters and operational modes for both direct and indirect AMFPIs developed in R/F applications. These detector parameters will be used as examples in our following discussion of imaging.

Amorphous Silicon Detectors

Table 3

323

Examples of AMFPI detector parameters used in R/F and CBCT

Detector type

Indirect

Indirect

Direct

Maker/model x-Ray detection material thickness Pixel pitch (mm) Fill factor Detector active area (cm
cm) Detector matrix Readout rate

Varian/Paxscan 4030CB CsI (Tl) 0.6 mm 194
194 0.7 40
30 2048
1536 60 fps (4
4 binning) 30 fps (2
2 binning) 7.5 fps (full res.)

Varian/Paxscan 2020 CsI (Tl) 0.6 mm 194
194

Shimadzu/Safire/Anrad/FPD 9 or 14 a-Se 1.0 mm 150
150 1 22
22 1472
1472 30 fps (1024
1024 ROI)

8.19.3.1

Image Correction

Before quantitative evaluation of image quality can be performed, projection images acquired by FPI need to be corrected for imperfection due to detector nonuniformity and defects. Defect pixels are unavoidable during fabrication of the active matrix. Due to the large number of pixels, even a 0.1% defect rate could result in 9000 bad pixels in a 3000
3000 pixel AMFPI. In addition, there is nonuniformity between pixels due to several reasons: (1) nonuniformity in active matrix which results in variation in TFT characteristics; (2) variation in the thickness of x-ray detector material; and (3) gain nonuniformity between different charge amplifier channels. This necessitates image correction through postprocessing. The standard method is an offset and gain nonuniformity correction followed by a defect pixel replacement. An offset (or dark) image is obtained without x-ray exposure and subtracted from each x-ray image. In order to reduce the effect of electronic noise, the average of several dark images is usually used. Since there is temporal drift in offset due to device instability, offset images are constantly updated between x-ray examinations. The gain correction is performed by dividing the offset subtracted image by a gain table, which is obtained during a calibration procedure. During calibration, the detector is exposed to uniform radiation. By averaging several x-ray images, the gain of each pixel can be determined. Defect pixels are identified by setting a lower threshold of x-ray sensitivity based on the pixel statistics, and the result is stored in a defect map. After gain correction, the bad pixels are replaced by the average values of neighboring good pixels. The gain table and bad pixel map are much more stable compared to offset; hence, in commercial detectors the calibration procedure only needs to be repeated once a month or even less frequently.

8.19.3.2

Spatial Frequency Domain Image Quality Metrics

8.19.3.2.1 Spatial resolution: MTF The spatial resolution of projection images is quantified by MTF. MTF is defined as the Fourier transform (FT) of the point spread function (PSF). This concept applies to a linear system that is shift invariant. In practice, MTF is usually measured in two orthogonal directions using FT of the line spread function. For AMFPI, the shift-invariance condition is violated because the digital detectors are undersampled, which makes

20
20 1024
1024 60 fps (2
2 binning) 30 fps (full res)

the PSF position dependent. For a digital detector with pixelsensing element width a ¼ 140 mm and pixel pitch d ¼ 150 mm, the detector aperture response is a sinc function with the first zero at f ¼ 1/a ¼ 7.1 cycles/mm. The Nyquist frequency of pixel sampling is fN ¼ 1/(2d) ¼ 3.3 cycles mm1. This means that a digital detector is always undersampled except when the frequency response of the x-ray detection material is very poor. In order to apply MTF to a digital detector, the concept of presampling MTF is usually used. It describes the frequency response of the detector before sampling occurs. The standard experimental techniques adopted by IEC for measuring the presampling MTF is the slanted edge method. Figure 8(a) shows the measured presampling MTF of an indirect AMFPI (Varian Paxscan 4030) with different detector pixel binning (Tognina et al., 2004). In full resolution with pixel pitch of 192 mm, the presampling MTF is dominated by the image blur in CsI, which is 600 mm thick. It is important to note that even with such thick CsI layers, the detector is undersampled with MTF ¼ 0.2 at the Nyquist frequency of fNY ¼ 2.6 cycles mm1. With 2
2 and 4
4 pixel binning, which is often used in fluoroscopy and CBCT image acquisition to increase readout speed, the aperture function of the larger binned pixels becomes the dominant factor for spatial resolution. The presampling MTF of direct AMFPI, on the other hand, is only limited by the pixel aperture function because there is essentially no image blur in the x-ray photoconductor. Figure 8(b) shows the measured MTF for an a-Se-based AMFPI with 150 mm pixel size (FPD14, Anrad Corp.; Hunt et al., 2004). The presampling MTF of the detector has its first zero at 6.1 cycles mm1, indicating an effective fill factor of unity for the 150 mm pixels. Image blur has been observed in some a-Se detectors due to charge trapping and recombination in the bulk of very thick a-Se layers or near the pixel electrode interface. This is a source of presampling blur and could contribute to 10–20% drop in presampling MTF at the Nyquist frequency depending on the material properties and thickness of a-Se (Hunt et al., 2004; Zhao et al., 2003).

8.19.3.2.2 Noise power spectra The noise in a projection image can be characterized in the spatial frequency domain using NPS, which is the FT of the autocorrelation of a flat-fielded x-ray image. The inherent stochastic (Poisson) noise of incident x-rays is white, that is, no spatial correlation. Image blur in an AMFPI detector could lead to spatial correlation of noise, which results in a

324

Amorphous Silicon Detectors

dimensional variable (in additional to the two spatial dimensions x and y), and determine the 2D NPS after correction of lag effect (Friedman and Cunningham, 2010); (2) measure 2D spatial domain NPS by eliminating the temporal effect, that is, at low frame rate where temporal correlation is negligible. For accurate measurement of 2D NPS, fixed pattern in projection images must be removed through offset and gain correction. Then a region of interest (ROI) I(x,y) is selected from each flat-fielded image with its mean value subtracted before FT to obtain NPS (IEC62220-1, 2004):

1.0 4x4 2x2 1x1

0.8

MTF

0.6

0.4

NPSðu; vÞ ¼ 0.2

0.0

0

1

2

3

4

5

E dx dy D 2 jFT½Iðx; yÞ  Iðx; yÞÞj Nx Ny

[1]

where hi represents the ensemble average, Nx and Ny are the number of elements in the x and y directions, respectively, and dx and dy are the pixel pitch in each direction.

Spatial frequency (cycles mm-1)

(a)

8.19.3.2.3 Detective quantum efficiency The overall imaging performance of an x-ray detector is best represented by its detective DQE. It is defined as the ratio between the SNR squared at the output of the detector and that at the input, which is equal to the number of x-ray photons per unit area q0:

1.0 1x1 2x2

0.8

DQE ¼

MTF

0.6

[2]

SNRout2 is also known as the number of noise equivalent quanta. Hence DQE describes the efficiency of the detector in utilizing the incident x-rays, and its upper limit is the QE
of the detection material. In order to describe the ability of the detector in transferring information with different frequency content, DQE is usually measured as a function of spatial frequency f using:

0.4

0.2

0.0 0 (b)

SNR out 2 q0

2

4

6

8

10

12

14

Spatial frequency (cycles mm-1)

Figure 8 MTF of commercial AMFPI detectors: (a) Indirect FPI with 192 mm pixel size and 600 mm thick columnar-structured CsI. Reproduced from Tognina CA, Mollov I, Yu JM, et al. (2004) Design and performance of a new a-Si flatpanel imager for use in cardiovascular and mobile C-arm imaging systems. Proceedings of SPIE 5368: 648–656; (b) Direct FPI with 150 mm pixel size and 1000 mm thick a-Se. Reproduced from Hunt DC, Tousignant O, and Rowlands JA (2004) Evaluation of the imaging properties of an amorphous selenium-based flat-panel detector for digital fluoroscopy. Medical Physics 31: 1166–1175.

high-frequency drop of NPS. When presampling NPS has frequency components above the Nyquist frequency of detector sampling, aliasing of NPS occurs, leading to an increase in NPS. Noise aliasing in direct conversion AMFPI results in an NPS that is essentially white. NPS of projection images can be measured experimentally using flat-fielded x-ray images under uniform x-ray exposures. For accurate measurement of NPS, the spatio-temporal behavior of detectors needs to be taken into account. Temporal performance of AMFPIs (to be discussed later) such as image lag could lead to noise correlation between frames, resulting in a reduction in NPS if NPS analysis is performed using a temporal sequence of x-ray images. Two methods have been used to account for this factor: (1) measure the spatio-temporal NPS by adding time domain as a third

DQEðf Þ ¼

k0 MTFðf Þ2 q0 NPSðf Þ

[3]

where k0 is the pixel x-ray response of the detector at a given x-ray exposure and NPS(f) is the NPS, which is the FT of the autocorrelation function of the detector at the same exposure level. Any additional noise source in an imaging system (e.g., detector electronic noise) increases NPS(f) from the x-ray quantum noise of
q0 and degrades the DQE. Because MTF(f) always decreases as a function of f, added noise (which is usually white) will cause DQE(f) to decrease with increasing f. Because it reflects both the signal and noise transfer of an imaging system, DQE(f) is regarded as the gold standard for performance comparison between different detectors. Dose dependence of DQE(f) is another important imaging performance criterion. The DQE of AMFPIs at very low exposures could be degraded due to the readout electronic noise, and it has been recognized as a major disadvantage compared to the more established x-ray detectors such as XRII, which has internal signal gain. Figure 9(a) shows the DQE(f) of Varian Paxscan 2020 detector in both full resolution and 2
2 binning readout modes (Tognina et al., 2004). It shows that at detector entrance exposure of 5.3 mR (46 nGy), DQE(0) of the detector for an RQA 5 spectrum is 0.7, which is approaching the theoretical limit of DQE(0) ¼ ɳAS, where AS is the Swank factor of CsI describing the added noise due to variation in

Amorphous Silicon Detectors

325

1.0 x-Rays

2x2 3.5 nGy 2x2 46 nGy 1x1 5.2 nGy 1x1 46 nGy

0.8

DQE

0.6

0.4

0.2 x-Ray signal

0.0 0.0

0.4

0.8

1.2

1.6

2.0

2.4

2.8

Spatial frequency (cycles mm-1)

Subsequent dark image

Figure 10 Conceptual images showing lag of an x-ray imaging system. Lag is defined as the residual signal from the detector’s previous exposure to radiation. It is manifested as an enhanced signal in a subsequent dark image (acquired without x-rays).

1.0 0.56 uR 4.4 uR 9.9 uR

0.8

x-Rays

x-Rays

DQE

0.6

0.4

0.2 x-Ray signal

0.0 0.0

0.4

0.8

1.2

1.6

Spatial frequency (cycles mm-1) Figure 9 (a) DQE for an indirect AMFPI with pixel size of 192 mm in both full resolution and 2
2 binning operation. DQE degradation due to electronic noise of the AMFPI is evident at detector entrance exposure of 3–4 nGy. Reproduced from Tognina CA, Mollov I, Yu JM, et al. (2004) Design and performance of a new a-Si flat-panel imager for use in cardiovascular and mobile C-arm imaging systems. Proceedings of SPIE 5368: 648–656; (b) DQE for a direct AMFPI with pixel size of 150 mm and 1000 mm thick a-Se layer. The DQE in 2
2 binning mode shows degradation for exposure < 4 uR. Reproduced from Hunt DC, Tousignant O, and Rowlands JA (2004) Evaluation of the imaging properties of an amorphous selenium-based flat-panel detector for digital fluoroscopy. Medical Physics 31: 1166–1175.

x-ray to optical photon conversion gain (Swank, 1973). However, as the exposure decreases to 0.65 mR (5.2 nGy), DQE(0) drops to below 0.6 due to the degradation effect of added electronic noise. The magnitude of this effect is more pronounced at high frequencies because the electronic noise is white, whereas the x-ray quantum noise decreases with frequency. Figure 9(b) shows the measured DQE of direct conversion AMFPI (FPD 14, Anrad). Similar degradation in DQE at low exposures is observed. However, the magnitude of DQE drop at high spatial frequency is similar to that for DQE(0). This is because the x-ray quantum noise in a-Se is virtually white.

Subsequent uniform exposure

Figure 11 Conceptual images showing ghosting of an x-ray imaging detector. Ghosting is defined as the change in x-ray sensitivity as a result of the detector’s exposure to radiation. It can only be seen with subsequent x-ray exposures.

8.19.3.3 Temporal Performance of Different x-Ray Detector Technologies Temporal imaging characteristics of AMFPI can be separated into two categories: lag and ghosting. As shown in Figure 10, lag is the carryover of image charge generated by previous x-ray exposures into subsequent image frames. It is manifested as changes in dark images, that is, readout of the detector without an x-ray exposure. As shown in Figure 11, ghosting is the change of x-ray sensitivity, or gain, of the detector as a result of previous exposures to radiation. It can only be seen with subsequent x-ray exposures. Both lag and ghosting could lead to image artifacts in projection and reconstructed images in CBCT. An overview of the physical mechanism for and the measurement of lag and ghosting will be provided here for both indirect and direct AMFPIs.

8.19.3.3.1 Temporal performance of indirect AMFPIs The lag and ghosting of indirect AMFPIs can be attributed to three sources of mechanisms: (1) charge trapping and release in a-Si PD; (2) after-glow from the CsI scintillator; and (3) incomplete readout of charge from the pixel to the charge

Amorphous Silicon Detectors

8.19.3.3.2 Temporal performance of direct AMFPI Lag and ghosting in a-Se AMFPIs are due to charge trapping in the bulk of the a-Se layer or at the interface between a-Se and the pixel electrodes (Zhao et al., 2002, 2005a). Comparison of temporal performance of complete AMFPI and a-Se samples (without TFT readout) showed that the dominant factors are

10 30 fps 15 fps

8

Lag (%)

amplifiers (when Ton < 5RC; Overdick et al., 2001). During x-ray exposure, the a-Si PD is biased with an electric field in order for image charge to be collected efficiently. Electrons in a-Si have better transport properties; therefore, most a-Si PDs are negatively biased at the light-entrance side. When electrons move toward pixel electrodes, they could be captured by localized state (traps) in the a-Si material, and released at a later time, for example, during the subsequent image frames. Lag has been investigated extensively under different imaging conditions, for example, detector exposure and frame rate (Siewerdsen and Jaffray, 1999). Figure 12(a) shows the relative signal intensity measured from an indirect AMFPI (Varian 2020) with the x-ray exposure delivered to frame zero, whose signal is set as the reference level (100%; Tognina et al., 2004). It shows that the first frame lag depends on the frame rate and ranged between 2% and 10% depending on operational conditions and entrance exposure. The time required for trapped charge to be released depends on the energy depth of the traps. Shallow traps are responsible for short-term lag, and deep traps for long-term residual signal, which could be visible tens of minutes after exposures, and the magnitude of long-term lag depends on the degree of pixel saturation and frame time (Siewerdsen and Jaffray, 1999). Usually, lag is more severe at higher exposures when the electric field across a-Si PDs nears collapses due to pixel saturation. This is because charge is more likely to be trapped under low electric field. Ghosting of indirect AMFPIs has been observed as an increase in x-ray sensitivity after the detector is exposed to radiation (Overdick et al., 2001). Figure 12(b) shows the relative x-ray sensitivity of indirect AMFPIs as a function of exposure, which exhibits a 2% increase in x-ray sensitivity even at 10 s after x-ray exposure of 20 mGy. This is because charge trapped in a-Si due to previous radiation exposure fills the traps and reduces the probability of further charge trapping in subsequent exposures, whereas a ‘rested’ (i.e., no recent history of radiation exposure) detector experiences reduction in x-ray sensitivity due to charge trapping. To alleviate ghosting due to this mechanism, reset light exposure has been implemented, where short pulses (100 ms) of light delivered between x-ray exposures would generate charge to fill the traps in a-Si PD, thereby minimizing the probability of charge trapping during x-ray exposure (Overdick et al., 2001). As shown in Figure 12(b), the longer the reset light duration (RLD), the lower the sensitivity ghost. This was also found to improve the x-ray sensitivity of AMFPI, compared to that without reset light, by 8% for RLD >100 ms. An alternative method developed to overcome lag and ghosting caused by charge trapping is to put a-Si PDs in the forward-bias condition for a short period between two subsequent exposures. Forward bias causes a large number of charge carriers injected from the bias electrodes of a-Si PD and fills the traps before the next x-ray exposure (Mollov et al., 2008).

6

4

2

0

0

5

(a)

10 Frame number

15

20

1.020

Relative gain at 20 uGy

326

1.015

1.010

1.005

1.000 (b)

RLD = 0 RLD = 10 µs RLD = 50 µs RLD = 250 µs

10

20

30 40 50 Time after exposure (s)

60

Figure 12 Measured lag for: (a) indirect AMFPI (Varian 2020) with frame rates of 15 and 30 fps. Reproduced from Tognina CA, Mollov I, Yu JM, et al. (2004) Design and performance of a new a-Si flat-panel imager for use in cardiovascular and mobile C-arm imaging systems. Proceedings of SPIE 5368: 648–656; (b) ghosting of indirect AMFPI (Trixel dynamic FPI), Reproduced from Overdick M, Solf T, and Wischmann H (2001) Temporal artifacts in flat-dynamic x-ray detectors. Proceedings of SPIE 4320: 47–58.

the charge trapping and recombination in the a-Se layer (Tousignant et al., 2005). The drift mobility of electrons in a-Se is only 1/50 of that of holes, and there are a large number of deep electron traps, which could capture electrons for up to several hours. Trapped electrons enhance the electric field near the positive bias electrode, and increase injection of holes, which is manifested as lag, that is, elevated dark signal, after radiation exposure. Figure 13(a) show lag measurements from a real-time a-Se AMFPI (FPD14, Anrad Corporation), as well as an a-Se layer identical to that used in AMFPI but without TFT readout (Tousignant et al., 2005). The first frame lag (30 fps) of the AMFPI depends on the radiation exposure, and increases from 1.7% at 48 mR to 3.9% at 384 mR. Ghosting in a-Se detectors is manifested as a reduction in x-ray sensitivity. This is due to the recombination between previously trapped electrons in the bulk of a-Se and the x-ray generated

Amorphous Silicon Detectors

8.19.4

4.0 FPI with 48 µR FPI with 328 µR a-Se sample with low dose

3.5 3.0 Lag (%)

2.0 1.5 1.0 0.5 0

2

4

(a)

6 8 10 12 14 Frame number at 30 fps

16

18

8.19.4.1

Normalized x-ray sensitivity

33 mR min-1 1.00

0.95

0.90

Without With charge recombination

0.80 0 (b)

10

20 30 40 50 Irradiation time (min)

Increasing Gain

For direct AMFPIs, charge conversion gain can be increased by increasing the electric field in a-Se, or to develop other x-ray photoconductors with higher gain, such as the materials discussed in Section 8.19.2.2.4. Detector concepts involving avalanche multiplication gain in a-Se photoconductor have also been proposed for both direct (Lee et al., 2004; Wronski and Rowlands, 2008) and indirect AMFPIs (Zhao et al., 2005b). Increasing the gain by a factor of 10 may allow AMFPI to maintain high DQE (x-ray quantum noise limited) at the lowest exposure used in fluoroscopy (with a single x-ray absorbed per pixel, even with the current level of electronic readout noise ( 1500 e rms) (Zhao et al., 2005b). The challenge with increased conversion gain in the sensing element is not to sacrifice dynamic range or add new sources of image degradation (e.g., blur or noise).

1.05

0.85

Emerging Detector Technology

As discussed previously, one of the major challenges for AMFPI in fluoroscopy is the degradation of DQE at low exposures due to electronic noise. Recently many new AMFPI detector concepts have been proposed to overcome this limitation. The strategies can be divided into two categories: one is to increase the x-ray to image charge conversion gain so that the x-ray quantum noise can overcome the electronic noise (Street et al., 2002a; Zhao et al., 2005b) and the other is to decrease the electronic noise by incorporating amplification using two or more TFTs at each pixel, which is also referred to as active pixel sensor (APS; El-Mohri et al., 2009; Karim et al., 2003).

2.5

0.0

327

60

70

Figure 13 Temporal performance of a-Se FPI. Adapted from Tousignant O, Demers Y, Laperriere L, Mani H, Gauthier P, and Leboeuf J (2005) Spatial and temporal image characteristics of a real-time large area a-Se x-ray detector. Proceedings of SPIE 5745: 207–215: (a) lag for FPI at two different exposures, as well as for a matching a-Se layer (without TFT readout) at the lower exposure; (2) ghosting, i.e. measurement of relative x-ray sensitivity, as a function of radiation exposure.

free holes (Fogal et al., 2004). It increases as a function of radiation dose and decreases with increasing electric field ESe. Figure 13(b) shows the quantitative measurements of ghosting in the same a-Se AMFPI (FPD14), where the relative x-ray sensitivity is measured as a function of time after x-ray has been delivered at a rate of 33 mR min1. It shows that the x-ray sensitivity continues to decrease with accumulation of exposure. Sensitivity recovery, or ghosting erasure, can be achieved through charge recombination technique. It has been shown previously that injection of holes into the bulk of a-Se between subsequent exposures accelerates recovery of x-ray sensitivity because the trapped electrons are neutralized through recombination with holes (Mahmood et al., 2009; Zhao and Zhao, 2005; Zhao et al., 2005a). This approach is different from the ghost erasure method used for indirect AMFPIs, where saturation of electron traps through injection of charge carriers was shown to be the effective mechanism.

8.19.4.2

Decreasing Electronic Noise: APSs

The readout electronic noise of AMFPI detectors has been analyzed and characterized extensively, and it was found that the fundamental limiting factors are the pixel reset noise and the charge amplifier noise, which increase with the data line capacitance (Weisfield and Bennett, 2001). The resistance of the data line also plays an important role depending on the charge amplifier design. As a result, it is difficult to reduce the electronic noise to <1500 e. The noise due to the readout electronics can be reduced through the incorporation of pixel amplification with two or more a-Si or poly-Si TFTs, so that the data line resistance and capacitance will no longer limit the performance. Furthermore, pixel amplification permits correlated double sampling of the pixel potential, which could eliminate pixel reset noise at the cost of increased readout time. Noise reduction to  500 e has been demonstrated for APS prototype sensors (El-Mohri et al., 2009), which would allow the x-ray quantum noise-limited DQE performance to extend to lower exposures than in existing AMFPI. With the advancement in very large scale integrated circuit, recently there has been a steady increase in the effort of making wafer-scale c-Si CMOS image sensors for x-ray imaging (Heo et al., 2011; Scheffer, 2007). Due to the limited wafer size (mostly 800 diameter due to cost considerations), the largest monolithic ‘tile’ dimension is approximately 12
12 cm (or rectangular tiles with comparable total surface area; Bohndiek

328

Amorphous Silicon Detectors

et al., 2009). The majority of wafer-scale CMOS sensors are APSs, which have amplification circuit at each pixel using three or more transistors (Farrier et al., 2009). They have the following advantages over a-Si FPI: (1) Pixel amplification permitting nondestructive readout and lower electronic noise (100–300 e rms); (2) faster readout speed; and (3) smaller pixel size (30– 40 mm for breast imaging and microCT). To make large area detectors, several CMOS tiles may be butted side by side with minimal dead zone (e.g., less than one pixel wide) between them. With each CMOS tile three-side buttable, tiled CMOS detectors with sizes up to 29
23 cm have been made (Naday et al., 2010). They have potential applications in digital breast tomosynthesis and dental CBCT. With future cost reduction and increase in yield, tiled wafer-scale CMOS APSs are expected to expand their applications.

References Adachi S, Hori N, Sato K, et al. (2000) Experimental evaluation of a-Se and CdTe flatpanel x-ray detectors for digital radiography and fluoroscopy. Proceedings of SPIE 3977: 38–47. Antonuk LE, Boudry J, Huang W, et al. (1992) Demonstration of megavoltage and diagnostic x-ray imaging with hydrogenated amorphous silicon arrays. Medical Physics 19: 1455–1466. Bohndiek SE, Blue A, Cabello J, et al. (2009) Characterization and testing of LAS: A prototype ‘large area sensor’ with performance characteristics suitable for medical imaging applications. IEEE Transactions on Nuclear Science 56: 2938–2946. Boudry JM and Antonuk LE (1996) Radiation damage of amorphous silicon, thin-film, field-effect transistors. Medical Physics 23: 743–754. Choi C-W, Kyun C, Kang S-S, and Nam S-H (2007) Comparison of compound semiconductor radiation films deposited by screen printing method. Proceedings of SPIE 6510: 651042. den Boer W, Aggas S, Byun YH, et al. (1998) Thin-film transistor array technology for high-performance direct-conversion x-ray sensors. Proceedings of SPIE 3336: 520–528. Du H, Antonuk LE, El-Mohri Y, et al. (2008) Investigation of the signal behavior at diagnostic energies of prototype, direct detection, active matrix, flat-panel imagers incorporating polycrystalline HgI2. Physics in Medicine and Biology 53: 1325–1351. El-Mohri Y, Antonuk LE, Koniczek M, et al. (2009) Active pixel imagers incorporating pixel-level amplifiers based on polycrystalline-silicon thin-film transistors. Medical Physics 36: 3340–3355. Farrier M, Graeve Achterkirchen T, Weckler GP, and Mrozack A (2009) Very large area CMOS active-pixel sensor for digital radiography. IEEE Transactions on Electron Devices 56: 2623–2631. Fogal B, Kabir MZ, O’Leary SK, Johanson RE, and Kasap SO (2004) X-ray-induced recombination effects in a-Se-based x-ray photoconductors used in direct conversion x-ray sensors. Journal of Vacuum Science and Technology A 22: 1005–1009. Friedman SN and Cunningham IA (2010) A spatio-temporal detective quantum efficiency and its application to fluoroscopic systems. Medical Physics 37: 6061–6069. Heijne L, Schagen P, and Buining H (1954) An experimental photoconductive camera tube for television. Philips Technical Review 16: 23. Heo SK, Kosonen J, Hwang SH, Kim TW, Yun S, and Kim HK (2011) 12-inchwafer-scale CMOS active-pixel sensor for digital mammography. Proceedings of SPIE 7961. Hunt DC, Tousignant O, and Rowlands JA (2004) Evaluation of the imaging properties of an amorphous selenium-based flat panel detector for digital fluoroscopy. Medical Physics 31: 1166–1175. IEC62220-1 (2004) Medical electrical equipment – Characteristics of digital X-ray imaging devices – Part 1: Determination of the detective quantum efficiency. Jacobs HBJ (1956) Large-area photoconductive x-ray pickup-tube performance. Electrical Engineering 75: 158. Jaffray DA and Siewerdsen JH (2000) Cone-beam computed tomography with a flatpanel imager: Initial performance characterization. Medical Physics 27: 1311–1323.

Kabir MZ (2008) Effects of charge carrier trapping on polycrystalline PbO x-ray imaging detectors. Journal of Applied Physics 104: 074506. Kabir MZ and Kasap SO (2002) Charge collection and absorption-limited sensitivity of x-ray photoconductors: Applications to a-Se and HgI2. Applied Physics Letters 80: 1664. Kabir MZ, Manouchehri F, Mahmood SA, et al. (2008) Presented at the Proceedings of SPIE, San Diego, CA, USA (unpublished). Kameshima T, Kaifu N, Takami E, Morishita M, and Yamazaki T (1998) Novel large-area MIS-type x-ray image sensor for digital radiography. Proceedings of SPIE 3336: 453–462. Kang Y, Antonuk LE, El-Mohri Y, et al. (2005) Examination of PbI2 and HgI2 photoconductive materials for direct detection, active matrix, flat-panel imagers for diagnostic x-ray imaging. IEEE Transactions on Nuclear Science 52: 38–45. Kanicki J (1992) Amophous & Microcrystalline Semiconductor Devices Vol. II: Materials and Device Physics. Boston: Artech House Publishers. Karim KS, Nathan A, and Rowlands JA (2003) Amorphous silicon active pixel sensor readout circuit for digital imaging. IEEE Transactions on Electron Devices 50: 200–208. Kuo Y (ed.) (2004) Thin Film Transistors Materials and Processes. In: Amorphous Silicon Thin Film Transistors, vol. 1. New York: Kluwer Academic Publishers. Lee DLY, Cheung LK, and Jeromin LS (1995) New digital detector for projection radiography. Proceedings of SPIE 2432: 237–249. Lee DLY, Cheung LK, Palecki EF, and Jeromin LS (1996) Discussion on resolution and dynamic range of Se-TFT direct digital radiographic detector. Proceedings of SPIE 2708: 511–522. Lee DLY, Storti G, and Golden KP (2004) Presented at the Proceedings of SPIE, Bellingham, WA, USA (unpublished). Lehnert J and Zhao W (2006) Presented at the Proceedings of SPIE, San Diego, CA, USA (unpublished). Mahmood SA, Kabir MZ, Tousignant O, and Mani H (2009) Ghosting and its recovery mechanisms in multilayer selenium detectors for mammography. Proceedings of SPIE 7258: 725860–725911. Manouchehri F, Kabir MZ, Tousignant O, Mani H, and Devabhaktuni VK (2008) Time and exposure dependent x-ray sensitivity in multilayer amorphous selenium detectors. Journal of Physics D: Applied Physics 41. Mollov I, Tognina C, and Colbeth R (2008) Photodiode forward bias to reduce temporal effects in a-Si based flat panel detectors. Proceedings of SPIE 6913: 69133S. Naday S, Bullard EF, Gunn S, et al. (2010) Optimised breast tomosynthesis with a novel CMOS flat panel detector. 10th International Workshop on Digital Mammography, IWDM 2010, 16 June 2010–18 June 2010Lecture Notes in Computer Science, vol. 6136, pp. 428–435. Ouimette DR, Nudelman S, and Aikens R (1998) A new large area X-ray image sensor. Proceedings of SPIE 3336: 470–476. Overdick M, Solf T, and Wischmann H (2001) Temporal artifacts in flat-dynamic x-ray detectors. Proceedings of SPIE 4320: 47–58. Pang G, Zhao W, and Rowlands JA (1998) Digital radiology using active matrix readout of amorphous selenium: Geometrical and effective fill factors. Medical Physics 25: 1636–1646. Polischuk B, Rougeot KWH, Debrie A, et al. (1999) Presented at the Proceedings of SPIE (unpublished). Powell MJ, Glasse C, French ID, Franklin AR, Hughes JR, and Curran JE (1997) Amorphous silicon photodiode-thin film transistor image sensor with diode on top structure. MRS Proceedings 467: 863–868. Rowlands JA, DeCrescenzo G, and Araj N (1992) X-ray imaging using amorphous selenium: Determination of x-ray sensitivity by pulse height spectroscopy. Medical Physics 19: 1065–1069. Rowlands JA and Yorkston J (2000) Flat panel detectors for digital radiography. In: Kundel HL, Van Metter RL, and Beutel J (eds.) Medical Imaging. Physics and Psychophysics, Psychophysics, vol. 1, pp. 223–328. Bellingham, WA: SPIE. Scheffer D (2007) A wafer scale active pixel CMOS image sensor for generic X-ray radiology. In: Medical Imaging 2007: Physics of Medical Imaging, 18 February–22 February, SPIE, vol. 6510. Shah KS, Bennett P, Dmitriyev Y, et al. (1999) PbI2 for high resolution digital x-ray imaging. Proceedings of SPIE 3770: 164–171. Siewerdsen JH and Jaffray DA (1999) A ghost story: Spatio-temporal response characteristics of an indirect-detection flat-panel imager. Medical Physics 26: 1624–1641. Simon M, Ford RA, Franklin AR, et al. (2005) Analysis of lead oxide (PbO) layers for direct conversion x-ray detection. IEEE Transactions on Nuclear Science 52: 2035–2040.

Amorphous Silicon Detectors

Street RA (1991) Hydrogenated Amorphous Silicon. Cambridge; New York: Cambridge University Press. Street RA, Ready SE, Melkhov L, Ho J, Zuck A, and Breen B (2002) Presented at the Proceedings of SPIE (unpublished). Street RA, Ready SE, Van Schuylenbergh K, et al. (2002b) Comparison of Pbl2 and Hgl2 for direct detection active matrix x-ray image sensors. Journal of Applied Physics 91: 3345. Swank RK (1973) Absorption and noise in x-ray phosphors. Journal of Applied Physics 44: 4199–4203. Tognina CA, Mollov I, Yu JM, et al. (2004) Design and performance of a new a-Si flatpanel imager for use in cardiovascular and mobile C-arm imaging systems. Proceedings of SPIE 5368: 648–656. Tokuda S, Adachi S, Sato T, et al. (2001) Experimental evaluation of a novel CdZnTe flat-panel x-ray detector for digital radiography and fluoroscopy. Proceedings of SPIE 4320: 140–147. Tousignant O, Demers Y, Laperriere L, Mani H, Gauthier P, and Leboeuf J (2005) Spatial and temporal image characteristics of a real-time large area a-Se x-ray detector. Proceedings of SPIE 5745: 207–215. Tsukamoto A, Yamada S, Tomisaki T, et al. (1999) Development and evaluation of a large-area selenium-based flat-panel detector for real-time radiography and fluoroscopy. Proceedings of SPIE 3659: 14–23. Watanabe M, Mochizuki C, Kameshima T, et al. (2001) Development and evaluation of a portable amorphous silicon flat-panel x-ray detector. Proceedings of SPIE 4320: 103–114. Weisfield RL and Bennett NR (2001) Electronic noise analysis of a 127-mu m pixel TFT/ photodiode array. Proceedings of SPIE 4320: 209–218. Weisfield RL, Yao W, Speaker T, Zhou K, Colbeth RE, and Proano C (2004) Performance analysis of a 127-micron pixel large-area TFT/photodiode array with boosted fill factor. Proceedings of SPIE 5368: 338–348.

329

Wronski MM and Rowlands JA (2008) Direct-conversion flat-panel imager with avalanche gain: Feasibility investigation for HARP-AMFPI. Medical Physics 35: 5207–5218. Yamazaki T, Tamura T, Nokita M, Okada S, Hayashida S, and Ogawa Y (2004) Performance of a novel 43-cm
43-cm flat-panel detector with CsI:Tl scintillator. Proceedings of SPIE 5368: 379–385. Zhao W, DeCrescenzo G, Kasap SO, and Rowlands JA (2005a) Ghosting caused by bulk charge trapping in direct conversion flat-panel detectors using amorphous selenium. Medical Physics 32: 488–500. Zhao W, DeCrescenzo G, and Rowlands JA (2002) Investigation of lag and ghosting in amorphous selenium flat-panel x-ray detectors. Proceedings of SPIE 4682: 9–20. Zhao W, Ji WG, Debrie A, and Rowlands JA (2003) Imaging performance of amorphous selenium based flat-panel detectors for digital mammography: Characterization of a small area prototype detector. Medical Physics 30: 254–263. Zhao W, Law J, Waechter D, Huang Z, and Rowlands JA (1998) Digital radiology using active matrix readout of amorphous selenium: Detectors with high voltage protection. Medical Physics 25: 539–549. Zhao W, Li D, Reznik A, et al. (2005b) Indirect flat-panel detector with avalanche gain: fundamental feasibility investigation for SHARP-AMFPI (scintillator HARP active matrix flat panel imager). Medical Physics 32: 2954–2966. Zhao W, Ristic G, and Rowlands JA (2004) X-ray imaging performance of structured cesium iodide scintillators. Medical Physics 31: 2594–2605. Zhao W and Rowlands JA (1995) X-ray imaging using amorphous selenium: Feasibility of a flat panel self-scanned detector for digital radiology. Medical Physics 22: 1595–1604. Zhao B and Zhao W (2005) Temporal performance of amorphous selenium mammography detectors. Medical Physics 32: 128–136.