Development of a large-area CMOS-based detector for real-time x-ray imaging

Transcription

1 Development of a large-area CMOS-based detector for real-time x-ray imaging Sung Kyn Heo a, Sung Kyu Park a, Sung Ha Hwang a, Dong Ak Im a, Jari Kosonen a, Tae Woo Kim a, Seungman Yun b, Ho Kyung Kim b* a Sensor Business Division, E-WOO Technology, Co., Ltd., Bora, Giheung, Yongin , South Korea b School of Mechanical Engineering, Pusan National University, Busan , South Korea ABSTRACT Complementary metal-oxide-semiconductor (CMOS) active pixel sensors (APSs) with high electrical and optical performances are now being attractive for digital radiography (DR) and dental cone-beam computed tomography (CBCT). In this study, we report our prototype CMOS-based detectors capable of real-time imaging. The field-of-view of the detector is cm. The detector employs a CsI:Tl scintillator as an x-ray-to-light converter. The electrical performance of the CMOS APS, such as readout noise and full-well capacity, was evaluated. The x-ray imaging characteristics of the detector were evaluated in terms of characteristic curve, pre-sampling modulation transfer function, noise power spectrum, detective quantum efficiency, and image lag. The overall performance of the detector is demonstrated with phantom images obtained for DR and CBCT applications. The detailed development description and measurement results are addressed. With the results, we suggest that the prototype CMOS-based detector has the potential for CBCT and real-time x-ray imaging applications. Keywords: Active pixel sensor, CMOS, detective quantum efficiency, real-time imaging 1. INTRODUCTION Recent developments in complementary metal-oxide-semiconductor (CMOS) imaging sensors have widely gained attention in digital radiographic applications because of their high-speed readout, low noise and high-spatial resolution. The arbitrary two-dimensional (D) addressability for readout and low-power consumption have been recognized as unique advantages against charge-coupled devices (CCDs). 1 In CMOS active pixel sensors (APSs), the term, "active" is referred to as the incorporation of active amplifiers within each pixel element, e.g., source-follower input transistor, 1 compared to the passive pixel sensor (PPS) design typically used in the conventional amorphous silicon-based flat-panel detectors and CCD-based detectors. Although the PPS design can be also achieved by the CMOS process, it may suffer from higher readout noise and lower readout speed than the APS design. CMOS APS typically consists of more than three transistors in each pixel element and those would be the reset, row-select, and source-follower transistors. In this study, we have developed a CMOS APS of which each pixel element has an N-well photodiode and four transistors as shown in Fig. 1(a). The additional transistor is the gain shaper transistor for improving the signal-to-noise ratio performance. The pixel-to-pixel pitch is 00 μm and the active pixel array format is pixels, which provides a field-of-view of ~ cm. The typical readout speed is 30 frames per second (fps). The output bit-depth is 14 bits. The detailed description on the development of CMOS APS is addressed with the measurement results of the electrical performance such as readout noise and full-well capacity in this manuscript. * hokyung@pnu.edu; phone ; fax Medical Imaging 010: Physics of Medical Imaging, edited by Ehsan Samei, Norbert J. Pelc, Proc. of SPIE Vol. 76, 763T 010 SPIE CCC code: /10/$18 doi: / Proc. of SPIE Vol T-1

2 Fig. 1. (a) Schematic diagram describing pixel architecture and (b) the layout of pixel. A prototype detector was completed by being coupled with a CsI:Tl scintillator. By measuring the mean pixel signal and variance as a function of x-ray input dose, we evaluated the signal and noise characteristics as well as the quantumlimited operation range. In addition, the imaging characteristics of the protoype detector were evaluated in terms of characteristic curve, modulation transfer function (MTF), noise-power spectrum (NPS), detective quantum efficiency (DQE) and image lag under the International Electrotechnical Commission (IEC) RQA5 imaging condition.. MATERIALS AND METHODS.1 CMOS APS The detector was designed in standard 0.18 µm CMOS technology. The detector has four transistors in each pixel described in Fig. 1(a). One is used for the pixel reset, where the pixel integration start level is set, the second transistor is common source, the third transistor is row selection (muxing) and the last one is gain shaping. The detector is a CMOS APS array with 00 µm pixel size. The detector is operated in usual rolling shutter way and its readout speed is 30 fps.. CMOS APS operation principle After the pixel integration start level is set by reset transistor control the image is captured. During the image capture the pixel voltage level rises because of the light signal affecting pn-junction of pixel photodiode. Acquired charge signal is proportional to the incoming light intensity in the pixel. After the integration is finished the sensor is read pixel by pixel by using the row and column transistors for muxing only one pixel from the array at time. The pixel current is converted Fig.. (a) Readout direction in one channel and (b) the timing for sensor. Proc. of SPIE Vol T-

3 to voltage signal and then digital signal by analog-to-digital (AD) conversion. Then several digital processing steps are used to produce the final output image. The rows go from top to bottom and column from right to left as shown in Fig. (a). Column readout is parallel for each horizontal 7-pixel segment to increase the maximum frame rate. Before actual readout, the detector is started and reset with the start sequence like Fig. (b). After the end row clock signal the column readout will be started internally in the ASIC (application specific integrated circuit). During 1-cycle row-change time, the readout (IOUT) will not produce any image signal and the zero data there should be removed in the image processing..3 Electrical performance evaluation The pixel leakage currents mainly consist of the photodiode and the reset transistor leakages 3. Also the common source transistor gate leakage could cause problems in submicron designs. The leakage currents should be small enough for successful charge integration and readout. Major of the integrated charge should be in the pixel when the pixels are read. Leakage currents causes also fixed pattern noise, since due to manufacturing defects, the leakage in the pixels is different. Pixel with high leakage due to photodiode or reset transistor defect can be seen in the raw image at different level from the surrounding pixels. Defected pixels can have high leakage or even short circuit, which prevents successful charge integration and signal readout. The sensitivity of image sensor is highest when the signal is as close to the overflow as possible in the maximum signal case 4. The sensitivity can be adjusted by photodiode selection (Nwell/Psub. N+/Psub, P+/Nwell) or by adjusting the junction depth of the particular diode 5. Junction depth adjustment is usually not available in standard CMOS process and thus the standard photodiode sensitivities are usually low for some applications. Pixel sensitivity can be adjusted also by the common source transistor dimensions (W/L-ratio). High W/L-ratio increases the pixel capacitance, and photodiode sensitivity in terms of the ratio of output voltage to input pixel voltage. Sensitivity is also increased by QE (quantum efficiency). For the photodiodes QE is highest when capacitance is lowest, since low capacitance photodiodes have large depletion width. Also the surface absorption before the depletion region should be minimized in the photodiode manufacturing by keeping the un-depleted surface layer as thin as possible. The dominant noise source of CMOS APS is reset noise 6. The pixel reset noise Q n,reset, which is caused by the thermal noise when the pixel-reset switch is off, is defined as Q n, reset kb T C pix =, (1) where, k B is the Boltzmann constant, C pix is the pixel capacitance, and T is absolute temperature of CMOS chip. The reset ktc noise was measured to lower than the expectation given by (1). This is because of the large fall time of reset line. The reset-control-pulse waveform causes damping in the high frequency noise components. Thus ktc noise is usually lower than the given (1). Leakage currents also induce noise in a pixel. This shot noise is defined as Q n, shot leakage = Ileak ( Tinteg + Tread ) qe,. () The square root of the number of electrons due to the leakage currents are integrated during the integration and readout time. The detector readout causes also noise because of the thermal noise in the common source, row and column transistors. Usually this noise is much lower than the previous noise mechanisms so that the achieved signal-to-noise ratio will not decrease significantly by the readout. Signal crosstalk between the photodiodes and EMI (electromagnetic interference) can also increase noise in some conditions. Crosstalk appears in the pixels when the signal is integrated or when the pixels are reset in one row at a time. Photodiodes have relatively high capacitve coupling to each other. During the integration the coupling capacitance interferes the normal integration. This could be avoided by covering the photodiodes with transparent conductors, such as ITO (Indium Tin Oxide) or ZnO (Zinc Oxide), and by connecting this layer to power supply (Ground or Vdd)..4 X-ray performance evaluation The detector is made of CMOS APS which is coupled with scintillator of CsI:Tl. To perform the physical x-ray characterization, a beam quality RQA5 was chosen from the IEC60-1 which requires a tube voltage of 70kV with 1mmAl filtration, giving rist to the half-value layer of 7.1 mmal. The x-ray spectrum from a 0.6-mm focal spot-sized x- ray tube (E75X, TOSHIBA ELECTRON TUBES & DEVICES CO., LTD.) was used under RQA5 condition. The x- Proc. of SPIE Vol T-3

4 ray dose measurement was performed with a calibrated Victoreen Nero max 8000 equipment. The detector was investigated in terms of the characteristic curve, variance analysis, MTF, NPS, DQE and image lag. To evaluate the detector response, uniform images were acquired at different x-ray input doses. The average value of a center area of 5 5 cm in output uniform images as a function of x-ray dose means the detector response and calls the characteristic curve of the detector. In digital x-ray imaging system, the mean pixel value is directly proportional to input x-ray dose on the detector surface for a specified x-ray spectrum 7. All image quality metrics of x-ray imaging system including MTF, NPS and DQE, are applicable only to linear range of the detector system. The pre-sampling MTF was measured by the slanted-edge method described in IEC A slanted 1-mm-thick tungsten edge phantom was placed on the detector surface aligned with x-ray beam center. The slanted degree was approximately 3 ~ 5. It determines the number of over-sampling. Over-sampled edge-response function was acquired from edge images. To obtain line-spread function, edge-response function was differentiated with a standard centraldifference algorithm 8. The MTF was then obtained by fast Fourier transform (FFT) of the line-spread function. The MTF was normalized to its value at zero frequency. The data for MTF were gain-offset corrected. The pre-sampling MTF should be fully finely sampled to avoid aliasing error due to finite pixel spacing. The NPS may be thought of as the variance of image intensity divided among the various frequency components of the image. The measurement of NPS was performed according to IEC60-1. In principle, the NPS measurement is very simple. At a particular x-ray dose, it was computed by taking the square of the magnitude of a D fast Fourier transform of a region-of-interest in number of flat-field images (gain-offset corrected). To avoid variation and erroneous of the result, the NPS was calculated with at least 16 million data. The linearized data in a region of pixels which cover the center region of x-ray field were evaluated for NPS. One-dimensional (1D) -NPS was extracted by radially averaging the D -NPS assuming circular symmetricity 9. To analyze the components of noise, the signal variance of the images which were used for NPS calculation was evaluated 10. The variance of image is the sum of three components which are additive noise σ e, photon noise σ p and fixed-pattern noise σ b and described as b p e a σ = σ + σ + σ = α K + β K + σ, (3) where K a is incident x-ray dose. The additive noise is mainly from electronic noise of the detector and should be independent of x-ray dose. The photon noise should be proportional to x-ray dose and fixed-pattern noise should be vary square of x-ray dose. The DQE is defined with the calculated MTF and NPS as SNR DQE = SNR out in a e ( G MTF) =, (4) NPS K q where, q, K a and G denote x-ray quantum fluence per unit dose, the measured input x-ray dose on detector surface and the linearized gain (sensitivity) of the detector, respectively. The value of x-ray quantum fluence was taken from IEC The x-ray imaging system based on the scintillator and photodiode arrays is known to have temporal artifact which decays over time. These temporal artifacts are attributed mainly to residual signals. The residual signals are due to delays in light generation in the scintillator or incomplete reset of signals. The residual signal from one image affects the next subsequent images. Lag effect was calculated by measuring the relative response or contrast to x-ray impulse on the detector. Images were subsequently acquired with sec intervals after x-ray impulse. The lag effect is defined as final initial a S final Safter lag =. (5) S S where S initial, S final and S after are the signal levels at equilibrium with x-ray impulse, the signal levels at equilibrium without x-ray exposure and the signal in the # th frame after exposure, respectively. Proc. of SPIE Vol T-4

5 Fig. 3. The characterization of dark current of detector was presented in number of electrons considering gain conversion factor. (a) The output dark current at different input pixel voltage was plotted. The line is linear fit curve describing linearity of data. (b) The r.m.s. of dark current at different input pixel voltage was plotted. The noise of dark current is linearly increasing until range of input voltage 1. V and decreasing after 1. V. 3. RESULTS 3.1 Electrical performance evaluation The electrical performance of the detector was evaluated in terms of full-well capacity, readout signal-and-noise curve and leakage currents. The electrical characterization parameters are summarized in table I. The performance of dark current of the detector was evaluated in terms of the output dark current and the r.m.s. of dark current as a function of input pixel voltage. The output dark current data and r.m.s. of dark current with different input voltage (Vpix) shown in Fig. 3 have been plotted in the unit of electrons when full-well capacitance was 10 7 electrons. The gain conversion factor of the detector was 610 electrons per ADU (analog-to-digital unit) at 14-bit depth. The output dark current data shape slightly in sigmoid curve but almost linear. In very low level of dark and just before saturation level, the curve is slightly drifted. Due to the signal saturation, the r.m.s of dark current is drastically decreased when the input voltage exceeds 1.4 V. The simulated pixel noise without ktc component corresponded to 17 μv and ktc noise was 40 μv. Considering the pixel capacitance.65 pf and the photodiode bias voltage, the simulated total noise is 711 electrons. On the contrary, the experimentally measured electrical noise of detector was 909 electrons, which implies that there is another source of additive noise from the readout board, such as active chips and power supply. The additive Performance parameter Value Units Quantum efficiency 55 % Fill factor 67 % Full well capacitance 10 M electrons Gain conversion factor 610 electrons/adu Electronic noise 910 Electrons Leakage current 4.0 fa Table I. Summary of electrical characterization parameter. noise from readout board was measured to 698 electrons. Therefore, the experimental result of CMOS pixel noise is 58 electrons and this is less than the simulation result estimated by (1). The leakage current of the detector was 4.0 fa which is the simulation result under 7 C condition. The best way to check the fixed-pattern noise and bad pixel due to leakage current is to evaluate leakage-current map. To generate the Proc. of SPIE Vol T-5

6 Fig. 4. (a) Leakage current map from subtracted dark frames which were acquired at different integration time showed leakage current was accumulated inverse-direction of readout order. (b) 3D surface plotted of leakage current map gave easy to recognize of characterization of leakage current. leakage-current map of the detector, we acquired two sets of dark data with different integration times. The control of integration time was easily performed by varying delay time of the readout pulse signals. We obtained the leakagecurrent map by subtracting the dark frame with lower integration time (0.033 sec, 30fps) from that with higher integration time (0.066 sec, 15fps). As shown in Fig. 4 (a) we could find the leakage current becomes higher in opposite direction of the readout order. The fixed-pattern noise due to leakage current is more visible in Fig. 4 (b). 3. X-ray performance evaluation The response performance of the detector is described in Fig. 5 (a) in terms of characteristic curve. The data were analyzed by first-order regression method. The curve shows that the pixel response with respect to incident x-ray dose is linear, but slightly drifts in the range of very low signal. The CMOS APS responds linearly under moderate illumination but has small nonlinearity under low light condition 11. We assumed three components of noise, which are fixed-pattern noise, photon noise, additive noise, as described in (3). The offset data were subtracted before variance analysis. The curve was analyzed with second-order polynomial to evaluate each component of noise as shown in Fig 5 (b). The fitting equation is also plotted on the graph. The term of additive noise σ e which is mainly due to electrical noise is almost same with noise of dark data, measured in dark current analysis. Over x-ray dose level of 3.75 μgy, variance of image decreases compared with the variance model of the detector. Measured pre-sampling MTF of detector is displayed in Fig. 6 and is 19% at the frequency of.5 lp/mm. The data were fitted with Lorenzian function. Fig. 5. (a) Characteristic curve of detector as a function of incident x-ray exposure with linear fitting curve. (b) Variance modeling of the detector with fitting curve. Proc. of SPIE Vol T-6

7 Fig. 6. Pre-sampling MTF curve of detector. The line is Lorenzian fit curve of MTF data. Fig.7 shows the measured 1D NPS of detector for different x-ray doses. The 1D NPS was extracted by radially averaging of D NPS assuming circular symmetricity. All the curves of NPS data have the same shape, but differ by a scaling factor that is proportional to x-ray dose. Due to the increased quantum noise, the noise spectral density increases as dose increases. Due to the decreased signal variance over x-ray dose level of 3.75 μgy as shown in Fig. 5 (b), the noise spectral density hardly increases in the range of 3.75 ~ 6.07 μgy. DQE of the detector is plotted in Fig. 8 with various x-ray doses. The data are dependent on x-ray dose. Over the x-ray dose level of 3.75 μgy, DQE data are significantly increased. Improvement in NPS over the x-ray dose level of 3.75 μgy and higher output signal of the detector increased the DQE performance. Lag effects were performed using the method as described by (5). The data for lag-effect test were acquired with 30 fps readout speed, which is the maximum speed of the detector, under x-ray input dose of.7 μgy and thus the time interval between each sequential frame is sec. The offset data were subtracted before lag-effect test. Fig. 9 (a) shows measured lag effects of the detector as a function of time after the end of x-ray exposure. Lag effect of first frame is % and then gradually decreases until 0.58 sec which is 16 th frame after the end of x-ray exposure. After 1 sec from the end of x-ray exposure, lag effect is 0.018%. Signal increase due to the lag effect shown in Fig. 9 (b) is less than 5 ADU during 1 sec of x-ray exposure. We acquired clinical images such as the projections of skull and hand phantoms. The image of hand phantom shown in Fig. 10 (a) was acquired in the x-ray exposure condition of 60kV and the irradiation dose of 4.1 μgy. At the x-ray exposure condition of 85 kv and the dose of 6.01 μgy, the head phantom image in Fig. 10 (b) was acquired. For those images, only flat-field correction was applied without any image enhancement procedures. CBCT-reconstructed data Fig. 7. Measured 1D NPS at different incident x-ray doses. Proc. of SPIE Vol T-7

8 Fig. 8. Measured DQE at different incident x-ray doses. were also acquired with the prototype detector and are demonstrated in Fig. 11. The data for CBCT were acquired with 30-fps readout speed. 4. DISCUSSION AND CONCLUSIONS The electrical and x-ray imaging performances of the prototype CMOS APS detector have been presented. The electrical characterization performed in terms of full-well capacitance, dark current analysis and leakage-current map. The photodiode of the detector was linearly responded to the input pixel voltage. Due to very low-level of photodiode leakage current, successful charge integration and readout were achieved. But some fixed-pattern noise due to leakage current was appeared in the leakage-current map. Low electrical noise (910 electrons) with high speed readout (30fps), high full well capacitance (10 7 electrons) and good quantum efficiency of the photodiode (44% at 540~560 nm) were key benefits of the developed detector. The selected N-well diode as the photodiode and the depth of pn-junction made the optimal pixel response. The x-ray performance was evaluated in terms of characteristic curve, variance analysis, MTF, NPS and DQE. The detector was linearly responded well and its sensitivity with respect to x-ray dose in linear range was electrons/μgy. Considering variance analysis, the component of fixed-pattern noise was quite high and was suspected due to leakage currents. Over 10 %-level of MTF result at high frequencies (~.5 lp/mm) would be suitable for dental CBCT and fluoroscopic applications. DQE(0) of the detector was varying from 35 to 60 % proportional to x- ray dose. The image lag of the first frame, which was acquired at sec after the end of x-ray exposure, is %. This is very low lag effect and the signal of image was just increased less than 5ADU within 1 sec. Therefore, the developed detector hardly has image lag. Fig. 9. The data for lag test were acquired 30 fps readout speed. (a) Measured lag effect under x-ray dose level of.7μgy with respect to time after end of x-ray exposure. (b) Signal mean value during x-ray exposure Proc. of SPIE Vol T-8

9 Fig. 10. X-ray images of human modeling phantom. (a) Hand phantom was acquired in 60kV x-ray energy. (b) Head phantom was acquired in 85kV x-ray energy. Fig. 11. CBCT reconstructed image of head phantom with developed detector. (a) Axial view and (b) cross axial view of reconstructed data. (c) 3D rendered image of reconstructed data. In this study we have investigated the design, physical performance and feasibility of the developed CMOS APS detector for medical applications, especially for CBCT and fluoroscopy. The performance of low noise and high sensitivity with high frame rate (~30fps) and no lag was well adaptable for CBCT and fluoroscopic applications. With pseudo-clinical images as shown in Fig. 10 and 11, we suggest that the detector has the potential in medical applications. However, the detailed analysis of the performance, such as the reason of variance decrease and DQE drop, was not investigated. Future work includes the analysis of parameters affecting image quality. ACKNOWLEDGEMENTS This work was supported by a Grant-in-Aid for Strategy Technology Development Programs from the Korea Ministry of Knowledge Economy (No ). Proc. of SPIE Vol T-9

10 REFERENCES [1] [] [3] [4] [5] [6] [7] [8] [9] [10] [11] Blue, A., Liang, H. X., Clark, A., Prydderch, M., Turchetta, R., and Apeller, R., Empirical electro-optical and x-ray performance evaluation of CMOS active pixel sensor for low dose, high resolution x-ray medical imaging, Med. Phys. 34(1), (007). Tian, H, et al., Active pixel sensors fabricated in a standard 0.18 μm CMOS technology, Proc. SPIE 4306, (001). Kwon, H. I., Kang, I. M., Park, B. -G., Lee, J. D., and Park, S. S. The analysis of dark signals in the CMOS APS imagers from characterization of test structures, IEEE Trans. Electron Devices, 51(), (004). Tarek, L., Stephan, B., Hogler, K., Frank, M., Peter, R., Konstantin, S., Michael, S., and Markus, B., Sensitivity of CMOS based imagers and scaling perspectives, IEEE Trans. Electron Devices, 47(11), (000). Lee, J. S., Hornsey, R. I., and Renshaw, D., Analysis of CMOS photodiodes Part II: Lateral photoresponse, IEEE Trans. Electron Devices, 50(5), (003). Tian, H., Fowler, B., and Gamal, A. E., Analysis of temporal noise in CMOS photodiode active pixel sensor, IEEE J. Solid-State Circuits, 36, (001). Cunningham, I. A., Image quality metrics for digital systems, Ch 3, in Handbook of Medical Imaging: vol 1 Physics, SPIE Press, Bellingham, 161- (000). Samei, E. M., Flynn, M., and Reinmann, D. A., A method for measuring the presampled MTF of digital radiographic systems using an edge test device, Med. Phys. 5, (1998). Aufrichtug, R., Su, Y., Cheng, Y., and Granfors, P. R., Measurement of the noise power spectrum in digital x-ray detector, Proc. SPIE 430, (001). Burgess, A., On the noise variance of a digital mammography system, Med. Phys. 31, (004). Tian, H., Liu, H., Lim, S. H., Kleinfelder, S., Gamal, A. E., Active pixel sensor fabricated in standard 0.18 um CMOS technology, Proc. SPIE 4306, (001). Proc. of SPIE Vol T-10