- Email: [email protected]
A new modular semi-parallel EIT system for medical application
Biomedical Signal Processing and Control 39 (2018) 416–423
Contents lists available at ScienceDirect
Biomedical Signal Processing and Control journal homepage: www.elsevier.com/locate/bspc
Research paper
A new modular semi-parallel EIT system for medical application Yixin Ma a,∗ , Liwen Miao a , Hui Qin b , Xinyi Chen a , Xiaofan Xiong a , Tao Han a , Peng Qin a , Xiaojun Ji a , Ping Cai a a b
School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University, Shanghai, China Department of Respiratory, Ren Ji Hospital, School of Medicine, Shanghai Jiao Tong University, Shanghai, China
a r t i c l e
i n f o
Article history: Received 10 April 2017 Received in revised form 29 July 2017 Accepted 1 August 2017 Available online 30 August 2017 Keywords: Bioimpedance Compound electrode Electricalimpedance tomography Data acquisition system (DAS) System calibration
a b s t r a c t A modular semi-parallel EIT data acquisition system (SJTU Mk-1) for medical application is newly developed. It consists of one control module and an expandable number of independent frontend modules. The control module generates stimulating signals, intermediates the communication between remote PC and frontends, and synchronize frontends during parallel data acquisition. All frontend modules are closely and symmetrically connected to electrode sensing array, so the length of signal traces can be minimized for better measurement accuracy at higher frequencies. The system can spectrum impedance from 1 kHz to 1 MHz which covers the majority frequency range of medical impedance investigations. The amplitudes of stimulating currents are limited to 0.4 mA with built-in alarm of abnormal current on each frontend module for extra safety protection. Transformers and optoelectronic couplers are used to isolate the human body under test from mains power supply and geological ground. The developed EIT Data Acquisition System (DAS) is natively safe and suitable for medical applications. To maximize number of independent measurement for better spatial resolution of reconstructed image, the sensing array is implemented with compound electrodes. System performance tests at excitation current less than 0.5 mA show that, the Signal-to-Noise Ratio (SNR) of transfer impedance is higher than 70 dB, the amplitude and phase measurement repeatability are better than 0.6% and 1◦ respectively. Initial phantom experiments further demonstrate the imaging capability of the developed EIT DAS for medical application. © 2017 Elsevier Ltd. All rights reserved.
1. Introduction Electrical Impedance Tomography (EIT) is a kind of non-invasive functional imaging technology to diagnose/monitor patients by measuring bioimpedances via externally wore electrode sensing array [1]. Benefited from low cost, fast imaging speed, no radiation hazard and high-sensitivity to functional change or biological activity of tissue, EIT is the only known imaging modality suitable for continuously monitoring lung function on bedside [2,3]. Medical EIT has also been actively investigated for pathological examination of malignant tissues like breast cancer [4,5], monitoring the bladder fullness [6], imaging neuronal depolarization in the brain [7], etc., which shows the prosperous of EIT research. Although some novel integrated circuit chips have been commercialized in recent years for impedance measurement which is very convenient for developing wearable instruments, but they
∗ Corresponding author at: School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University, Shanghai, China. E-mail address: [email protected] (Y. Ma). http://dx.doi.org/10.1016/j.bspc.2017.08.001 1746-8094/© 2017 Elsevier Ltd. All rights reserved.
are generally for lower frequency application only. For example, AD5933/5934 from Analog Devices were first released in 2005, is for impedance measurement up to 100 kHz. AFE4300 from Texas Instruments is a very comprehensive one with bandwidth up to 250 kHz, but its internal integrated signal generator has only 6 bits of Digital to Analog Converter (DAC) which may lead to instability of the impedance measurement. As for medical EIT, there are still physiological properties of tissue remain unknown. The intracellular physiological property can only be observed at high-frequency range [1]. Research at higher frequencies is still necessary and may actually be beneficiary or even has advantage to reveal new phenomenon. Development of broadband high performance DAS system is one important direction of EIT research. If the EIT systems are classified according to their structure, we can see that many EIT systems including Sheffield Mk3.5 EIT/EIS system [8], UCLH Mk 2.5 [9], and some other systems [10,11] at the early EIT research, are all serial systems with only one set of stimulating sources and one channel of voltmeters. This kind of system is relative simple and has low cost. Analog multiplexers are used to sequentially switch the stimulating signals and voltmeters to
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423
different set of excitation electrodes and measurement electrodes. The parasitic capacitances of multiplexers and cables are major error sources when the measurement frequency is 250 kHz and above [12]. The serial EIT systems generally measure impedances at 200 kHz and below. To overcome the parasitic capacitance affects, despite the significant complexity [12], a few research groups have implemented multiple stimulators and voltmeters fully parallel across all excitation channels and measurement channels, therefore no multiplexers are required to switch in between channels. For example, the KHU Mark2 [13] system with frequency range of 10 Hz to 500 kHz, and the Dartmouth system (published in 2008) [14] expanded the working frequency to more than 2 MHz. The significant advantages of the parallel structure are faster data acquisition speed, flexible stimulation strategies, and reduced effects of parasitic capacitances at higher frequencies. The Dartmouth system even eliminated as much as possible analog circuits in the analog frontend unit to further reduce parasitic capacitances and improve impedance measurement performance at higher frequencies. In real medical application, for safety purpose, the excitation currents are generally less than half mA. Accurate calibration of excitation currents on all front end channels may be a big challenge and timing consuming. As a compromise, semi-parallel EIT systems are developed with one pair of stimulators and parallel voltmeters. For example, the Leeds System (published in 2005) [15] developed with electronic components for industrial applications, the KHU Mark1 (published in 2007) [16] and the most recent Dartmouth system (published in 2015) [17] implemented with commercialized expensive National Instrumentation (NI, USA) function modules. The Dartmouth system (2015) is great for research but a bit expensive for commercialization of the EIT system. The four NI modules quoted more than 25k US dollar on NI website (www.ni.comwww. ni.com). All the three systems have their circuit modules packed in a chassis and so, long electrode cables are necessary for signal transmission. The long cables introduce parasitic capacitances and result in degradation of impedance measurement performance at higher frequencies. Started from the motivation to design a medical EIT system [18,19] that has stable performances in the spectrum range from 1 kHz to 1 MHz, with deep consideration of patient safety and ease
417
of hardware upgrades and scalability, a modular semi-parallel EIT Data Acquisition System (DAS), SJTU Mk-1, have been recently developed and presented in this paper. The system structure, the control module, the frontend module, and the compound electrode array are introduced. System performances are calibrated in terms of transfer impedance with resistor rings and water phantom. Excitation current amplitude is set to less than half mA, the same as how it will be used in medical application. The calibration results are presented. 2. Design of the EIT data aqcuisition system 2.1. Architecture design The architecture of the SJTU Mk-1 EIT DAS is described in Fig. 1. It features modular design and minimized lengths of electrode cables. The system has four key function blocks: a control module, a power supply module, sixteen frontend modules, and a ribbon cable with seventeen connecting slots numbered from 1 to 17. Slot 17 is the only one that is on the outer side of the ribbon cable, connecting with the control module. The rest of the 16 slots are installed in the inner side of the ribbon cable and equal spaced for easier time compensation of signal transmission. Time compensation is achieved by careful PCB layouts on control module and frontend modules by making signal trances from different slots the same length. This is very important for accurate phase measurement at higher frequencies. Each frontend module is closely connected with one compound electrode [20], via a pair of short flexible cables for current and voltage signal transmissions. The frontend modules are then plugged into ribbon cable slots 1–16 for power supply, signal transmission, and communication with controller module. Signals are interleaved and shielded by ground lines in the ribbon cable to reduce crosstalk noises. A significant difference of the SJTU Mk-1 EIT DAS system from others is that, instead of all circuits assembled in a compact chassis, all the frontend modules are placed closely around the target to minimize parasitic capacitances of electrode cables. The compound electrode and electrode array is shown in Fig. 2. Compound electrodes allow excitation and measurement at the same location and hence independent measurement number can
Fig. 1. Architecture of the modular design of SJTU Mk-1 EIT DAS.
418
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423
Fig. 2. Compound electrodes and electrode array. (a) Schematics of Compound Electrode (b) Photo of Compound Electrode (c) Photo of Compound Electrode Array for Chest EIT.
be increased. In adjacent measurement mode, 16 compound electrodes give 136 independent measurements while 16 conventional one-piece electrodes can only give 104 independent measurements. This means in theory the 16 compound electrodes can improve image spatial resolution by 23%. Although the compound electrodes are designed to directly plug into the frontend modules, electrode cables of 10 cm long are used for flexibility in adapting the measurement system to different sizes and shapes of human bodies. Every frontend module has an embedded Micro-Controller Unit (MCU) for distributed signal processing. The communication between frontend modules and control module is based on the protocol of CAN bus 2.0B. The CAN bus features minimized electromagnetic radiation and expandable network by interface repeaters. It provides potential for the EIT DAS to accommodate more frontend modules for increased number of electrodes. The modular semi-parallel architecture of the SJTU Mk-1 EIT DAS has the advantage for partial upgradation in the future to make the system state of the art.
2.2. Control module The control module includes an embedded MCU, excitation sources, and a synchronizer, as shown in Fig. 3. The MCU controls a 10-bit DDS (AD9832, Analog Devices, USA) to generate a sine-wave current signal with programmable frequency. The current signal is then converted to voltage signal via a programmable digital potentiometer (AD5160, Analog Devices, USA), and further split into two complementary voltage signals via ADA4922 (Analog Devices, USA), a differential driver, for reduced common mode voltage and increased common mode rejection ratio at the voltmeter circuit. Isolation transformers couple the voltage stimulation signals to frontend circuits so the human body under test is float from geological ground signal and the DC components in the stimulation signals can be automatically eliminated. This is an important effort for medical safety.
The second key function of the control module is to generate synchronized sampling clock for frontend Analog-to-Digital Converters (ADC) based on the stimulation frequency and samples per cycle setting from the user interface. The clock is transmitted to frontend circuits through ribbon cable as complementary clocks driven by LVDS mechanism for noise rejection and low power consumption. The last key function of the control module is data processing and communication. The control module communicates with frontends via reliable CAN Bus. A wired-AND logic circuit functions as the handshaking signal between all frontend modules and the control module. The hardware handshaking signal is pulled down by the frontend controller each time before the injection of stimulating signals, and turns into active high-level if and only if all frontend modules completed their A-to-D conversion. Then the controlling module sequentially read data from frontend modules and further processed to work out impedance. The impedance data are then uploaded to remote computer via Universal Asynchronous Receiver/Transmitter (UART) for image reconstruction. The UART interface is electrically isolated by optoelectronic couplers to prevent noise coupling from remote computer. If the MCU is replaced by a powerful embedded computer, or if the speed is not that crucial, the image reconstruction can be accomplished on the control module as well.
2.3. Frontend module Each frontend module has a voltmeter to measure the voltage between the connected voltage electrode and the last neighboring voltage electrode, and a current meter to measure the current if the connected current electrode is one of the two stimulating electrodes, as shown in Fig. 4. The current is sampled by a precision 20 ohm resistor of 0.1% accuracy and then amplified 100 times before A-to-D conversion. The voltage and current signals are digitized synchronously by a 12-bit dual-channel ADC (ADS2806, TI, USA). ADC results of voltage and current signals are automatically buffered in two independent
Fig. 3. Functional block diagram of the control module.
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423
419
Fig. 4. Functional block diagram of the frontend module.
SRAMs for high speed of data acquisition. The electrode switch connects the current electrode to one of the complementary stimulation signals, or analog ground, or none of them during stimulation, calibration or measurement only. The MCU reads ADC results from SRAM and initially demodulates to get amplitudes and phases of voltage and currents. The phases are relative to the first ADC clock. All frontend modules manipulate signals, conduct A-to-D conversion, read data and demodulate data in parallel to increase the speed of data collection and data processing. The demodulated data are transferred to control module in turns. It should be mentioned that for medical safety, the manipulated current-sampling signals and boundary voltage signals are isolated from the control circuits by analog transformers, and the digital control signals from MCU to electrode switches are electrically isolated by optoelectronic couplers. Besides, if any abnormal current is detected, the microcontroller will switch off all the stimulation circuits immediately. The frontend module is designed for adjacent measurement mode. It can be easily modified for other measurement mode by disconnecting the signal from last module and wire it to analog ground. Then the electrode potential can be in turn measured one by one and voltage across two electrodes can be calculated in software at the control module.
3. System performance Although larger excitation current can improve the impedance measurement performances, to be consistent with medical application purpose, the system calibration was conducted at the same condition as how it will be used in a medical application, i.e. excitation currents are less than 0.5 mA for all frequencies and are far below the limit of most medical safety regulations. Seven frequencies (1 kHz, 20 kHz, 50 kHz, 250 kHz, 500 kHz, 750 kHz, 1 MHz) covered the entire frequency range were selected for performance calibration. Five resistor rings and a cylindrical water phantom were used to calibrate the system, as shown in Fig. 5. Each resistor ring is made of 16 resistors of the same value. The five resistor rings are of 120 , 180 , 220 , 300 , and 430 ±1% respectively. The resistors are soldered and linked in a ring with 16 joint points. On each joint point, a resistor of 330 ±1% is used to mimic electrode contact resistance, which connects the frontend module to one of the 16 joint points of the resistor ring. Fig. 5(a) shows the calibration setup with a resistor ring. Fig. 5(b) shows the calibration setup with a water phantom. A cylindrical phantom of 20 cm diameter, installed with 16 compound electrodes was used in calibration. The water depth was maintained at 9 cm, the water conductivity (335 uS/cm) and temperature (18.8 ◦ C) were measured by a conductivity meter EL30 (Mettler Toledo, Switzerland).
Fig. 5. Calibration setups.(a)Calibration setup with a resistor ring(b)Calibration with a cylindrical phantom.
Transfer impedances are measured in adjacent mode, i.e. stimulation and voltage measurement are all via/from two adjacent electrodes. Linearity, repeatability and Signal-to-Noise Ratio (SNR) of the developed SJTU Mk-1 EIT DAS system were tested in terms of transfer impedance and reported in the following. The performance calibration of transfer impedance is very importance and more meaningful as an EIT system compared to performance evaluation of voltmeter circuit only or other parts of the EIT data acquisition system, because the transfer impedances are at the end used in image reconstruction. 3.1. Data acquisition speed The frame rate of the DAS system is over 20 frame/s (fps) at frequency of 20 kHz and above, but it reduces to 0.5 fps at 1 kHz due to the longer signal sampling time.
420
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423
3.2. Linearity In theory, boundary voltage is proportional to transfer impedance across two pairs of electrodes. Five resistor networks were measured by the developed SJTU Mk-1 EIT system. The measurement results were compared with the reference values measured by a LCR meter (U1733C, Agilent, USA). The linearity of the system is found to be better than 1.1% at frequencies up to 500 kHz, 2.4% at 750 kHz, and 4.5% at 1 MHz. Although strong efforts have been made to reduce parasitic capacitances, the linearity of impedance measurement still became worse with the increase of frequency because the unavoidable remaining parasitic capacitances. The increase of gaps between analog signal tracks to neighboring signal tracks on PCB boards and the usage of analog components with smaller stray capacitances may be helpful to further improve linearity of impedance measurement at higher frequencies.
3.3. Stability Stability or repeatability is an important parameter for any measurement instruments. The stability of SJTU Mk-1 EIT system is evaluated with the 430 resistor ring and a homogeneous water phantom. The resistance between adjacent electrodes of the phantom is 448 at 1 kHz which is comparable with the resistance network. Measurements were repeated 100 times for all seven frequencies. The amplitude stability error (ASE) was evaluated by the maximum Relative Standard Deviation (RSD), defined in Eq. (2) [14], that among all sixteen channels.
N 1 ¯ 2 (Zi − Z)
Fig. 6. Repeatability calibration at selected frequencies. (a) Repeatability calibrated with resistor network (b)Repeatability calibrated with water phantom.
N
RSD =
i=1
Z¯
× 100%
(2)
where, N is the number of repeated measurements, zi is the ith measured resistance, z¯ is the average resistance of all repeated measurements of the transfer impedance Z. Here N = 100, each time Z is measured over 100 cycles and there are 16 samples in each cycle of excitation signal. The phase stability error (PSE) was evaluated by calculating the Standard Deviation (SD) of phase angle of the transfer impedance Z from the same set of data. The calibration results are given in Fig. 6, where the error bar represents the range of respected value among all transfer impedances. The RSD of impedance amplitude is graded on left vertical axis and the SD of impedance phase is graded on right vertical axis. It can be seen that the impedance amplitude fluctuation is less than 0.6% and the impedance phase fluctuation is less than 1◦ at all tested frequencies. The calibration results with water phantom is slightly better than the calibration results with resistor network because the resistor network is soldered on PCB board with wire connections. The wires may introduce magnetic noise from environment and parasitic capacitances to the circuit.
bility test described in section III. C. The SNR was computed from Eq. (3) [14]. N
SNR = 10log10 (
Zi2
i=1 N
¯ (Zi − Z)
)
i=1
where, N is the number of repeated measurements, zi is the ith measured resistance, z¯ is the average resistance of all repeated measurements. Here N is 100. Each transfer impedance is measured over 100 sinewave cyles and the signal sampling frequency is 16 times of excitation signal frequency. The calibration results show that the system SNR is better than 70 dB at the tested frequencies, as shown in Fig. 7. It can also be found that the SNR shows a degradation at frequencies higher than
3.4. Signal-to-Noise ratio (SNR) The evaluation of SNR of the overall impedance measurement was conducted at the same experiment condition as the repeata-
(3)
2
Fig. 7. Signal-to-Noise Ratio (SNR) calibrated with resistor ring.
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423
421
Fig. 8. Phantom experiments with agar lung models. (a) Photo of the lung phantom (b) Some Measured Impedance Amplitudes. For better displaying, the red line and green line are shifted to the right 4 and 8 coordinates respectively to avoid overlapping (c) The amplitdue images reconstructed from time difference data and the imaginary images reconstructed from frequency-difference data. The black lines represent the contours of agar models.. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
250 kHz, which may be contaminated by the switching frequency (about 380 kHz − 500 kHz) of the DC-DC converters on the power supply PCB board. Proper shielding, a better power line filtering circuit and avoiding working frequency close to the switching frequency of DC-DC converters may be helpful to further improve the system SNR, and also the average of repeated measurements can significantly improve system SNR [16]. 4. Phantom expriemtns To verify the performance of the developed EIT system, a lung phantom has been setup as shown in Fig. 8. The saline water is 900 uS/cm (T = 14.4 ◦ C), the height of the water is maintained at 5 cm, the phantom diameter is 20 cm, and two 5.5 cm height of lung-
shaped agar models with conductivity of 1.5 uS/cm are inserted into water to mimic lungs. Reference data was collected when the phantom was filled with saline water only, and then the two agar models were placed into the phantom while some water was taken out to keep the water level unchanged. Finite Element model with 1664 triangle elements was used in image reconstruction and images were reconstructed by normalized one-step sensitivity-based conjugate gradient method [21]. The image reconstruction is coded in MATLAB. The impedance amplitude at 20 kHz and 1 MHz are shown in Fig. 8(b). For clear display, the 20 kHz and 1 MHz impedance amplitude are shifted to the right for 4 and 8 coordinates respectively to avoid overlapping.
422
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423
Time-difference images [14] were reconstructed to get the amplitude of impedance distribution by Eq. (4) [21], and the imaginary part of the impedance distribution was reconstructed from the frequency-difference [13] data compared with the reference data at 20 kHz by Eq. (5) [21]. Zmea − Zref Zref
=−
Im[Zmea ] − Im[Zref ] Zref
˝
=−
∇ ∇ϕ I
Iϕ
(ωε) ˝
d˝
(4)
∇ ∇ϕ I
Iϕ
d˝
(5)
where, ||•|| and Im[•] represent the amplitude and imaginary part of the measured impedance. Zmea is the impedance vector measured on phantom with agar model, and Zref is the reference impedance vector measured on homogenous water phantom. and ϕ represent the reciprocal potential distribution in sensitivity field , when current I and Iϕ are injected into from the reciprocal excitation electrodes. The equations are solved to get the normalized conductivity and permittivity ε distributions. The reconstructed images are shown in Fig. 8(c), where the real boundary of the lung models are outlined in black for comparison. It can be seen from Fig. 8 that the normalized time-difference images are similar at different frequencies while the normalized frequency-difference images have increased contrast at higher frequencies. It confirms the enhanced imaginary information of the target at higher frequencies. The importance of broadband frequency range in observing permittivity property is demonstrated. 5. Conclusion and discussion This article presents a newly developed modular semi-parallel EIT system for medical research and applications. The data acquisition system includes a control module, a power supply module, sixteen frontend modules, and a ribbon cable. The modular design and plug-in assembly mechanism make the system flexible and easily expandable. We can increase/reduce number of frontend modules to adapt the measurement system for increased/reduced number of electrodes based on specific application purpose. The strategy to have the current sampling circuit close to excitation electrode, the minimized analog signal traces and the parallel signal conditioning and acquisition, are key features of the system. These features are important to reduce noise interference, improve measurement performance at higher frequencies and increase data acquisition speed. The system has one pair of matched stimulators, and parallel multiple voltmeters and equipped with compound electrodes. All voltmeters are distributed closely to the object under measurement to minimize cable parasitic capacitance effects. Compared with multi-stimulation sources, the usage of only one pair of matched stimulation sources can greatly reduce calibration load, and also avoid measurement error caused by mismatch of multistimulators. In addition, because the output impedance of constant current source reduces with frequency, the output current is not constant across entire frequency range. In the presented EIT data acquisition system, the synchronized current-sampling with voltage measurement offers additional amplitude and phase correction to make the impedance measurements more accurate. The EIT image reconstruction is an ill-posed problem. Small measurement change may result in large perturbation on reconstructed image. Therefore, the data acquisition system (DAS) of EIT is requested to have good stability and SNR. Generally the SNR has to be better than 60 dB in order to distinguish two different conductivity distributions [14]. Performance calibrated at real application condition (the amplitude of the excitation current is less than 0.5 mA) show that the presented system has good linearity, stability
and more than 70 dB SNR of transfer impedance measurement. The system performance degrades slightly at frequencies higher than 500 kHz, for which we believe the imperfect mechanical assembly, PCB layout and the parasitic capacitance of electronic components may be blamed. Better PCB layout, more reliable electrical connections, and electrical components with smaller stray capacitances should be considered in the future for performance improvement at higher frequencies. As a kind of medical devices, the electrical safety is deeply considered in the presented system. Optoelectronic couplers are used for digital signal isolation and transformers are used for analog signal isolation, therefor the frontend circuits have floating grounds to prevent ground current flowing through human body. The amplitude of excitation current is strictly controlled and monitored to be far less than the medical safety limit, so the human body under test can be safely protected from electrical damage. The fundamental lung-phantom experiments have verified the tomographic ability of the developed SJTU Mk-1 EIT system. Static phantom experiments show clear conductivity image across the frequency range of 1 kHz to 1 MHz. It is also demonstrated that significant imaginary image can be obtained at frequencies of 500 kHz and above. The total cost of the prototype system is less than 2.9k USD which is much less than the 25k USD of the four NI modules used in some published EIT system. The presented prototype system is a little cumbersome, though, by upgrading the control module with wireless communication and replacing the frontend modules with specialized integrated circuit chips, the SJTU Mk-1 system could be made wearable and wireless for daily bedside monitoring, which will be the next target of our research. Acknowledgment This research was financially supported by the National Natural Science Foundation of China (Grant No. 61371017). References [1] R.H. Bayford, Bioimpedance tomography (electrical impedance tomography), Annu. Rev. Biomed. Eng. 8 (2006) 63–91, http://dx.doi.org/10.1146/annurev. bioeng.8.061505.095716. [2] R.H. Bayford, A. Tizzard, Bioimpedance imaging: an overview of potential clinical applications, Analyst 137 (2012) 4635, http://dx.doi.org/10.1039/ c2an35874c. [3] D.T. Nguyen, C. Jin, a Thiagalingam, a L. McEwan, A review on electrical impedance tomography for pulmonary perfusion imaging, Physiol. Meas. 33 (2012) 695–706, http://dx.doi.org/10.1088/0967-3334/33/5/695. [4] R.J. Halter, A. Hartov, S.P. Poplack, R. Diflorio-Alexander, W.A. Wells, K.M. Rosenkranz, R.J. Barth, P.A. Kaufman, K.D. Paulsen, Real-time electrical impedance variations in women with and without breast cancer, IEEE Trans. Med. Imaging 34 (2015) 38–48, http://dx.doi.org/10.1109/TMI.2014.2342719. [5] M.S. Campisi, C. Barbre, A. Chola, G. Cunningham, J. Viventi, Breast cancer detection using high – density flexible electrode arrays and electrical impedance tomography, 36th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. (2014) 1131–1134, http://dx.doi.org/10.1109/embc.2014.6943794. [6] R. Li, J. Gao, Y. Li, J. Wu, Z. Zhao, Y. Liu, Preliminary study of assessing bladder urinary volume using electrical impedance tomography, J. Med. Biol. Eng. 36 (2016) 71–79, http://dx.doi.org/10.1007/s40846-016-0108-1. [7] D.S. Holder, Electrical impedance tomography (EIT) of brain function, Brain Topogr. 5 (1992) 87–93, http://dx.doi.org/10.1007/BF01129035. [8] A.J. Wilson, P. Milnes, A.R. Waterworth, R.H. Smallwood, B.H. Brown, Mk3.5: a modular, multi-frequency successor to the Mk3a EIS/EIT system, Physiol. Meas. 22 (2001) 49–54, http://dx.doi.org/10.1088/0967-3334/22/1/307. [9] A. McEwan, A. Romsauerova, R. Yerworth, L. Horesh, R. Bayford, D. Holder, Design and calibration of a compact multi-frequency EIT system for acute stroke imaging, Physiol. Meas. 27 (2006) S199–S210, http://dx.doi.org/10. 1088/0967-3334/27/5/S17. [10] M. Goharian, M. Soleimani, A. Jegatheesan, K. Chin, G.R. Moran, A DSP based multi-frequency 3D electrical impedance tomography system, Ann. Biomed. Eng. 36 (2008) 1594–1603, http://dx.doi.org/10.1007/s10439-008-9537-5. [11] N. Ranade V, D.C. Gharpure, Design and development of instrumentation for acquiring electrical impedance tomography data, 2nd Int. Symp. Phys. Technol. Sensors, IEEE, 2015 (2015) 97–101, http://dx.doi.org/10.1109/ISPTS. 2015.7220091.
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423 [12] A. McEwan, G. Cusick, D.S. Holder, A review of errors in multi-frequency EIT instrumentation, Physiol. Meas. 28 (2007) S197–S215, http://dx.doi.org/10. 1088/0967-3334/28/7/s15. [13] T.I. Oh, H. Wi, D.Y. Kim, P.J. Yoo, E.J. Woo, A fully parallel multi-frequency EIT system with flexible electrode configuration: KHU Mark2, Physiol. Meas. 32 (2011) 835–849, http://dx.doi.org/10.1088/0967-3334/32/7/S08. [14] R.J. Halter, A. Hartov, K.D. Paulsen, A broadband high-frequency electrical impedance tomography system for breast imaging, IEEE Trans. Biomed. Eng. 55 (2008) 650–659, http://dx.doi.org/10.1109/tbme.2007.903516. [15] M. Wang, Y. Ma, N. Holliday, Y. Dai, R.A. Williams, G. Lucas, A high-performance EIT system, IEEE Sens. J. 5 (2005) 289–298, http://dx.doi. org/10.1109/JSEN.2005.843904. [16] T.I. Oh, E.J. Woo, D. Holder, Multi-frequency EIT system with radially symmetric architecture: KHU Mark1, Physiol. Meas. 28 (2007) S183–S196, http://dx.doi.org/10.1088/0967-3334/28/7/s14. [17] S. Khan, P. Manwaring, A. Borsic, R. Halter, FPGA-based voltage and current dual drive system for high frame rate electrical impedance tomography, IEEE
[18]
[19]
[20]
[21]
423
Trans. Med. Imaging 34 (2015) 888–901, http://dx.doi.org/10.1109/tmi.2014. 2367315. L. Miao, Y. Ma, X. Xiong, X. Chen, T. Han, P. Cai, P. Qin, X. Ji, A new symmetric semi-parallel electrical impedance tomography (EIT) system – i: the design, IEEE Int. Conf. Imaging Syst. Tech., IEEE, 2016 (2016) 19–23, http://dx.doi.org/ 10.1109/IST.2016.7738191. L. Miao, Y. Ma, X. Chen, X. Xiong, P. Cai, P. Qin, X. Ji, T. Han, New symmetric semi-parallel electrical impedance tomography (EIT) system – II: the performance, IEEE Int. Conf. Imaging Syst. Tech., IEEE, 2016 (2016) 283–288, http://dx.doi.org/10.1109/IST.2016.7738238. P. Hua, E.J. Woo, J.G. Webster, W.J. Tompkins, Using compound electrodes in electrical impedance tomography, Biomed. Eng. IEEE Trans. 40 (1993) 29–34, http://dx.doi.org/10.1109/10.204768. M. Wang, Inverse solutions for electrical impedance tomography based on conjugate gradients methods, Meas. Sci. Technol. 13 (2001) 101–117, http:// dx.doi.org/10.1088/0957-0233/13/1/314.
Contents lists available at ScienceDirect
Biomedical Signal Processing and Control journal homepage: www.elsevier.com/locate/bspc
Research paper
A new modular semi-parallel EIT system for medical application Yixin Ma a,∗ , Liwen Miao a , Hui Qin b , Xinyi Chen a , Xiaofan Xiong a , Tao Han a , Peng Qin a , Xiaojun Ji a , Ping Cai a a b
School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University, Shanghai, China Department of Respiratory, Ren Ji Hospital, School of Medicine, Shanghai Jiao Tong University, Shanghai, China
a r t i c l e
i n f o
Article history: Received 10 April 2017 Received in revised form 29 July 2017 Accepted 1 August 2017 Available online 30 August 2017 Keywords: Bioimpedance Compound electrode Electricalimpedance tomography Data acquisition system (DAS) System calibration
a b s t r a c t A modular semi-parallel EIT data acquisition system (SJTU Mk-1) for medical application is newly developed. It consists of one control module and an expandable number of independent frontend modules. The control module generates stimulating signals, intermediates the communication between remote PC and frontends, and synchronize frontends during parallel data acquisition. All frontend modules are closely and symmetrically connected to electrode sensing array, so the length of signal traces can be minimized for better measurement accuracy at higher frequencies. The system can spectrum impedance from 1 kHz to 1 MHz which covers the majority frequency range of medical impedance investigations. The amplitudes of stimulating currents are limited to 0.4 mA with built-in alarm of abnormal current on each frontend module for extra safety protection. Transformers and optoelectronic couplers are used to isolate the human body under test from mains power supply and geological ground. The developed EIT Data Acquisition System (DAS) is natively safe and suitable for medical applications. To maximize number of independent measurement for better spatial resolution of reconstructed image, the sensing array is implemented with compound electrodes. System performance tests at excitation current less than 0.5 mA show that, the Signal-to-Noise Ratio (SNR) of transfer impedance is higher than 70 dB, the amplitude and phase measurement repeatability are better than 0.6% and 1◦ respectively. Initial phantom experiments further demonstrate the imaging capability of the developed EIT DAS for medical application. © 2017 Elsevier Ltd. All rights reserved.
1. Introduction Electrical Impedance Tomography (EIT) is a kind of non-invasive functional imaging technology to diagnose/monitor patients by measuring bioimpedances via externally wore electrode sensing array [1]. Benefited from low cost, fast imaging speed, no radiation hazard and high-sensitivity to functional change or biological activity of tissue, EIT is the only known imaging modality suitable for continuously monitoring lung function on bedside [2,3]. Medical EIT has also been actively investigated for pathological examination of malignant tissues like breast cancer [4,5], monitoring the bladder fullness [6], imaging neuronal depolarization in the brain [7], etc., which shows the prosperous of EIT research. Although some novel integrated circuit chips have been commercialized in recent years for impedance measurement which is very convenient for developing wearable instruments, but they
∗ Corresponding author at: School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University, Shanghai, China. E-mail address: [email protected] (Y. Ma). http://dx.doi.org/10.1016/j.bspc.2017.08.001 1746-8094/© 2017 Elsevier Ltd. All rights reserved.
are generally for lower frequency application only. For example, AD5933/5934 from Analog Devices were first released in 2005, is for impedance measurement up to 100 kHz. AFE4300 from Texas Instruments is a very comprehensive one with bandwidth up to 250 kHz, but its internal integrated signal generator has only 6 bits of Digital to Analog Converter (DAC) which may lead to instability of the impedance measurement. As for medical EIT, there are still physiological properties of tissue remain unknown. The intracellular physiological property can only be observed at high-frequency range [1]. Research at higher frequencies is still necessary and may actually be beneficiary or even has advantage to reveal new phenomenon. Development of broadband high performance DAS system is one important direction of EIT research. If the EIT systems are classified according to their structure, we can see that many EIT systems including Sheffield Mk3.5 EIT/EIS system [8], UCLH Mk 2.5 [9], and some other systems [10,11] at the early EIT research, are all serial systems with only one set of stimulating sources and one channel of voltmeters. This kind of system is relative simple and has low cost. Analog multiplexers are used to sequentially switch the stimulating signals and voltmeters to
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423
different set of excitation electrodes and measurement electrodes. The parasitic capacitances of multiplexers and cables are major error sources when the measurement frequency is 250 kHz and above [12]. The serial EIT systems generally measure impedances at 200 kHz and below. To overcome the parasitic capacitance affects, despite the significant complexity [12], a few research groups have implemented multiple stimulators and voltmeters fully parallel across all excitation channels and measurement channels, therefore no multiplexers are required to switch in between channels. For example, the KHU Mark2 [13] system with frequency range of 10 Hz to 500 kHz, and the Dartmouth system (published in 2008) [14] expanded the working frequency to more than 2 MHz. The significant advantages of the parallel structure are faster data acquisition speed, flexible stimulation strategies, and reduced effects of parasitic capacitances at higher frequencies. The Dartmouth system even eliminated as much as possible analog circuits in the analog frontend unit to further reduce parasitic capacitances and improve impedance measurement performance at higher frequencies. In real medical application, for safety purpose, the excitation currents are generally less than half mA. Accurate calibration of excitation currents on all front end channels may be a big challenge and timing consuming. As a compromise, semi-parallel EIT systems are developed with one pair of stimulators and parallel voltmeters. For example, the Leeds System (published in 2005) [15] developed with electronic components for industrial applications, the KHU Mark1 (published in 2007) [16] and the most recent Dartmouth system (published in 2015) [17] implemented with commercialized expensive National Instrumentation (NI, USA) function modules. The Dartmouth system (2015) is great for research but a bit expensive for commercialization of the EIT system. The four NI modules quoted more than 25k US dollar on NI website (www.ni.comwww. ni.com). All the three systems have their circuit modules packed in a chassis and so, long electrode cables are necessary for signal transmission. The long cables introduce parasitic capacitances and result in degradation of impedance measurement performance at higher frequencies. Started from the motivation to design a medical EIT system [18,19] that has stable performances in the spectrum range from 1 kHz to 1 MHz, with deep consideration of patient safety and ease
417
of hardware upgrades and scalability, a modular semi-parallel EIT Data Acquisition System (DAS), SJTU Mk-1, have been recently developed and presented in this paper. The system structure, the control module, the frontend module, and the compound electrode array are introduced. System performances are calibrated in terms of transfer impedance with resistor rings and water phantom. Excitation current amplitude is set to less than half mA, the same as how it will be used in medical application. The calibration results are presented. 2. Design of the EIT data aqcuisition system 2.1. Architecture design The architecture of the SJTU Mk-1 EIT DAS is described in Fig. 1. It features modular design and minimized lengths of electrode cables. The system has four key function blocks: a control module, a power supply module, sixteen frontend modules, and a ribbon cable with seventeen connecting slots numbered from 1 to 17. Slot 17 is the only one that is on the outer side of the ribbon cable, connecting with the control module. The rest of the 16 slots are installed in the inner side of the ribbon cable and equal spaced for easier time compensation of signal transmission. Time compensation is achieved by careful PCB layouts on control module and frontend modules by making signal trances from different slots the same length. This is very important for accurate phase measurement at higher frequencies. Each frontend module is closely connected with one compound electrode [20], via a pair of short flexible cables for current and voltage signal transmissions. The frontend modules are then plugged into ribbon cable slots 1–16 for power supply, signal transmission, and communication with controller module. Signals are interleaved and shielded by ground lines in the ribbon cable to reduce crosstalk noises. A significant difference of the SJTU Mk-1 EIT DAS system from others is that, instead of all circuits assembled in a compact chassis, all the frontend modules are placed closely around the target to minimize parasitic capacitances of electrode cables. The compound electrode and electrode array is shown in Fig. 2. Compound electrodes allow excitation and measurement at the same location and hence independent measurement number can
Fig. 1. Architecture of the modular design of SJTU Mk-1 EIT DAS.
418
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423
Fig. 2. Compound electrodes and electrode array. (a) Schematics of Compound Electrode (b) Photo of Compound Electrode (c) Photo of Compound Electrode Array for Chest EIT.
be increased. In adjacent measurement mode, 16 compound electrodes give 136 independent measurements while 16 conventional one-piece electrodes can only give 104 independent measurements. This means in theory the 16 compound electrodes can improve image spatial resolution by 23%. Although the compound electrodes are designed to directly plug into the frontend modules, electrode cables of 10 cm long are used for flexibility in adapting the measurement system to different sizes and shapes of human bodies. Every frontend module has an embedded Micro-Controller Unit (MCU) for distributed signal processing. The communication between frontend modules and control module is based on the protocol of CAN bus 2.0B. The CAN bus features minimized electromagnetic radiation and expandable network by interface repeaters. It provides potential for the EIT DAS to accommodate more frontend modules for increased number of electrodes. The modular semi-parallel architecture of the SJTU Mk-1 EIT DAS has the advantage for partial upgradation in the future to make the system state of the art.
2.2. Control module The control module includes an embedded MCU, excitation sources, and a synchronizer, as shown in Fig. 3. The MCU controls a 10-bit DDS (AD9832, Analog Devices, USA) to generate a sine-wave current signal with programmable frequency. The current signal is then converted to voltage signal via a programmable digital potentiometer (AD5160, Analog Devices, USA), and further split into two complementary voltage signals via ADA4922 (Analog Devices, USA), a differential driver, for reduced common mode voltage and increased common mode rejection ratio at the voltmeter circuit. Isolation transformers couple the voltage stimulation signals to frontend circuits so the human body under test is float from geological ground signal and the DC components in the stimulation signals can be automatically eliminated. This is an important effort for medical safety.
The second key function of the control module is to generate synchronized sampling clock for frontend Analog-to-Digital Converters (ADC) based on the stimulation frequency and samples per cycle setting from the user interface. The clock is transmitted to frontend circuits through ribbon cable as complementary clocks driven by LVDS mechanism for noise rejection and low power consumption. The last key function of the control module is data processing and communication. The control module communicates with frontends via reliable CAN Bus. A wired-AND logic circuit functions as the handshaking signal between all frontend modules and the control module. The hardware handshaking signal is pulled down by the frontend controller each time before the injection of stimulating signals, and turns into active high-level if and only if all frontend modules completed their A-to-D conversion. Then the controlling module sequentially read data from frontend modules and further processed to work out impedance. The impedance data are then uploaded to remote computer via Universal Asynchronous Receiver/Transmitter (UART) for image reconstruction. The UART interface is electrically isolated by optoelectronic couplers to prevent noise coupling from remote computer. If the MCU is replaced by a powerful embedded computer, or if the speed is not that crucial, the image reconstruction can be accomplished on the control module as well.
2.3. Frontend module Each frontend module has a voltmeter to measure the voltage between the connected voltage electrode and the last neighboring voltage electrode, and a current meter to measure the current if the connected current electrode is one of the two stimulating electrodes, as shown in Fig. 4. The current is sampled by a precision 20 ohm resistor of 0.1% accuracy and then amplified 100 times before A-to-D conversion. The voltage and current signals are digitized synchronously by a 12-bit dual-channel ADC (ADS2806, TI, USA). ADC results of voltage and current signals are automatically buffered in two independent
Fig. 3. Functional block diagram of the control module.
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423
419
Fig. 4. Functional block diagram of the frontend module.
SRAMs for high speed of data acquisition. The electrode switch connects the current electrode to one of the complementary stimulation signals, or analog ground, or none of them during stimulation, calibration or measurement only. The MCU reads ADC results from SRAM and initially demodulates to get amplitudes and phases of voltage and currents. The phases are relative to the first ADC clock. All frontend modules manipulate signals, conduct A-to-D conversion, read data and demodulate data in parallel to increase the speed of data collection and data processing. The demodulated data are transferred to control module in turns. It should be mentioned that for medical safety, the manipulated current-sampling signals and boundary voltage signals are isolated from the control circuits by analog transformers, and the digital control signals from MCU to electrode switches are electrically isolated by optoelectronic couplers. Besides, if any abnormal current is detected, the microcontroller will switch off all the stimulation circuits immediately. The frontend module is designed for adjacent measurement mode. It can be easily modified for other measurement mode by disconnecting the signal from last module and wire it to analog ground. Then the electrode potential can be in turn measured one by one and voltage across two electrodes can be calculated in software at the control module.
3. System performance Although larger excitation current can improve the impedance measurement performances, to be consistent with medical application purpose, the system calibration was conducted at the same condition as how it will be used in a medical application, i.e. excitation currents are less than 0.5 mA for all frequencies and are far below the limit of most medical safety regulations. Seven frequencies (1 kHz, 20 kHz, 50 kHz, 250 kHz, 500 kHz, 750 kHz, 1 MHz) covered the entire frequency range were selected for performance calibration. Five resistor rings and a cylindrical water phantom were used to calibrate the system, as shown in Fig. 5. Each resistor ring is made of 16 resistors of the same value. The five resistor rings are of 120 , 180 , 220 , 300 , and 430 ±1% respectively. The resistors are soldered and linked in a ring with 16 joint points. On each joint point, a resistor of 330 ±1% is used to mimic electrode contact resistance, which connects the frontend module to one of the 16 joint points of the resistor ring. Fig. 5(a) shows the calibration setup with a resistor ring. Fig. 5(b) shows the calibration setup with a water phantom. A cylindrical phantom of 20 cm diameter, installed with 16 compound electrodes was used in calibration. The water depth was maintained at 9 cm, the water conductivity (335 uS/cm) and temperature (18.8 ◦ C) were measured by a conductivity meter EL30 (Mettler Toledo, Switzerland).
Fig. 5. Calibration setups.(a)Calibration setup with a resistor ring(b)Calibration with a cylindrical phantom.
Transfer impedances are measured in adjacent mode, i.e. stimulation and voltage measurement are all via/from two adjacent electrodes. Linearity, repeatability and Signal-to-Noise Ratio (SNR) of the developed SJTU Mk-1 EIT DAS system were tested in terms of transfer impedance and reported in the following. The performance calibration of transfer impedance is very importance and more meaningful as an EIT system compared to performance evaluation of voltmeter circuit only or other parts of the EIT data acquisition system, because the transfer impedances are at the end used in image reconstruction. 3.1. Data acquisition speed The frame rate of the DAS system is over 20 frame/s (fps) at frequency of 20 kHz and above, but it reduces to 0.5 fps at 1 kHz due to the longer signal sampling time.
420
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423
3.2. Linearity In theory, boundary voltage is proportional to transfer impedance across two pairs of electrodes. Five resistor networks were measured by the developed SJTU Mk-1 EIT system. The measurement results were compared with the reference values measured by a LCR meter (U1733C, Agilent, USA). The linearity of the system is found to be better than 1.1% at frequencies up to 500 kHz, 2.4% at 750 kHz, and 4.5% at 1 MHz. Although strong efforts have been made to reduce parasitic capacitances, the linearity of impedance measurement still became worse with the increase of frequency because the unavoidable remaining parasitic capacitances. The increase of gaps between analog signal tracks to neighboring signal tracks on PCB boards and the usage of analog components with smaller stray capacitances may be helpful to further improve linearity of impedance measurement at higher frequencies.
3.3. Stability Stability or repeatability is an important parameter for any measurement instruments. The stability of SJTU Mk-1 EIT system is evaluated with the 430 resistor ring and a homogeneous water phantom. The resistance between adjacent electrodes of the phantom is 448 at 1 kHz which is comparable with the resistance network. Measurements were repeated 100 times for all seven frequencies. The amplitude stability error (ASE) was evaluated by the maximum Relative Standard Deviation (RSD), defined in Eq. (2) [14], that among all sixteen channels.
N 1 ¯ 2 (Zi − Z)
Fig. 6. Repeatability calibration at selected frequencies. (a) Repeatability calibrated with resistor network (b)Repeatability calibrated with water phantom.
N
RSD =
i=1
Z¯
× 100%
(2)
where, N is the number of repeated measurements, zi is the ith measured resistance, z¯ is the average resistance of all repeated measurements of the transfer impedance Z. Here N = 100, each time Z is measured over 100 cycles and there are 16 samples in each cycle of excitation signal. The phase stability error (PSE) was evaluated by calculating the Standard Deviation (SD) of phase angle of the transfer impedance Z from the same set of data. The calibration results are given in Fig. 6, where the error bar represents the range of respected value among all transfer impedances. The RSD of impedance amplitude is graded on left vertical axis and the SD of impedance phase is graded on right vertical axis. It can be seen that the impedance amplitude fluctuation is less than 0.6% and the impedance phase fluctuation is less than 1◦ at all tested frequencies. The calibration results with water phantom is slightly better than the calibration results with resistor network because the resistor network is soldered on PCB board with wire connections. The wires may introduce magnetic noise from environment and parasitic capacitances to the circuit.
bility test described in section III. C. The SNR was computed from Eq. (3) [14]. N
SNR = 10log10 (
Zi2
i=1 N
¯ (Zi − Z)
)
i=1
where, N is the number of repeated measurements, zi is the ith measured resistance, z¯ is the average resistance of all repeated measurements. Here N is 100. Each transfer impedance is measured over 100 sinewave cyles and the signal sampling frequency is 16 times of excitation signal frequency. The calibration results show that the system SNR is better than 70 dB at the tested frequencies, as shown in Fig. 7. It can also be found that the SNR shows a degradation at frequencies higher than
3.4. Signal-to-Noise ratio (SNR) The evaluation of SNR of the overall impedance measurement was conducted at the same experiment condition as the repeata-
(3)
2
Fig. 7. Signal-to-Noise Ratio (SNR) calibrated with resistor ring.
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423
421
Fig. 8. Phantom experiments with agar lung models. (a) Photo of the lung phantom (b) Some Measured Impedance Amplitudes. For better displaying, the red line and green line are shifted to the right 4 and 8 coordinates respectively to avoid overlapping (c) The amplitdue images reconstructed from time difference data and the imaginary images reconstructed from frequency-difference data. The black lines represent the contours of agar models.. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
250 kHz, which may be contaminated by the switching frequency (about 380 kHz − 500 kHz) of the DC-DC converters on the power supply PCB board. Proper shielding, a better power line filtering circuit and avoiding working frequency close to the switching frequency of DC-DC converters may be helpful to further improve the system SNR, and also the average of repeated measurements can significantly improve system SNR [16]. 4. Phantom expriemtns To verify the performance of the developed EIT system, a lung phantom has been setup as shown in Fig. 8. The saline water is 900 uS/cm (T = 14.4 ◦ C), the height of the water is maintained at 5 cm, the phantom diameter is 20 cm, and two 5.5 cm height of lung-
shaped agar models with conductivity of 1.5 uS/cm are inserted into water to mimic lungs. Reference data was collected when the phantom was filled with saline water only, and then the two agar models were placed into the phantom while some water was taken out to keep the water level unchanged. Finite Element model with 1664 triangle elements was used in image reconstruction and images were reconstructed by normalized one-step sensitivity-based conjugate gradient method [21]. The image reconstruction is coded in MATLAB. The impedance amplitude at 20 kHz and 1 MHz are shown in Fig. 8(b). For clear display, the 20 kHz and 1 MHz impedance amplitude are shifted to the right for 4 and 8 coordinates respectively to avoid overlapping.
422
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423
Time-difference images [14] were reconstructed to get the amplitude of impedance distribution by Eq. (4) [21], and the imaginary part of the impedance distribution was reconstructed from the frequency-difference [13] data compared with the reference data at 20 kHz by Eq. (5) [21]. Zmea − Zref Zref
=−
Im[Zmea ] − Im[Zref ] Zref
˝
=−
∇ ∇ϕ I
Iϕ
(ωε) ˝
d˝
(4)
∇ ∇ϕ I
Iϕ
d˝
(5)
where, ||•|| and Im[•] represent the amplitude and imaginary part of the measured impedance. Zmea is the impedance vector measured on phantom with agar model, and Zref is the reference impedance vector measured on homogenous water phantom. and ϕ represent the reciprocal potential distribution in sensitivity field , when current I and Iϕ are injected into from the reciprocal excitation electrodes. The equations are solved to get the normalized conductivity and permittivity ε distributions. The reconstructed images are shown in Fig. 8(c), where the real boundary of the lung models are outlined in black for comparison. It can be seen from Fig. 8 that the normalized time-difference images are similar at different frequencies while the normalized frequency-difference images have increased contrast at higher frequencies. It confirms the enhanced imaginary information of the target at higher frequencies. The importance of broadband frequency range in observing permittivity property is demonstrated. 5. Conclusion and discussion This article presents a newly developed modular semi-parallel EIT system for medical research and applications. The data acquisition system includes a control module, a power supply module, sixteen frontend modules, and a ribbon cable. The modular design and plug-in assembly mechanism make the system flexible and easily expandable. We can increase/reduce number of frontend modules to adapt the measurement system for increased/reduced number of electrodes based on specific application purpose. The strategy to have the current sampling circuit close to excitation electrode, the minimized analog signal traces and the parallel signal conditioning and acquisition, are key features of the system. These features are important to reduce noise interference, improve measurement performance at higher frequencies and increase data acquisition speed. The system has one pair of matched stimulators, and parallel multiple voltmeters and equipped with compound electrodes. All voltmeters are distributed closely to the object under measurement to minimize cable parasitic capacitance effects. Compared with multi-stimulation sources, the usage of only one pair of matched stimulation sources can greatly reduce calibration load, and also avoid measurement error caused by mismatch of multistimulators. In addition, because the output impedance of constant current source reduces with frequency, the output current is not constant across entire frequency range. In the presented EIT data acquisition system, the synchronized current-sampling with voltage measurement offers additional amplitude and phase correction to make the impedance measurements more accurate. The EIT image reconstruction is an ill-posed problem. Small measurement change may result in large perturbation on reconstructed image. Therefore, the data acquisition system (DAS) of EIT is requested to have good stability and SNR. Generally the SNR has to be better than 60 dB in order to distinguish two different conductivity distributions [14]. Performance calibrated at real application condition (the amplitude of the excitation current is less than 0.5 mA) show that the presented system has good linearity, stability
and more than 70 dB SNR of transfer impedance measurement. The system performance degrades slightly at frequencies higher than 500 kHz, for which we believe the imperfect mechanical assembly, PCB layout and the parasitic capacitance of electronic components may be blamed. Better PCB layout, more reliable electrical connections, and electrical components with smaller stray capacitances should be considered in the future for performance improvement at higher frequencies. As a kind of medical devices, the electrical safety is deeply considered in the presented system. Optoelectronic couplers are used for digital signal isolation and transformers are used for analog signal isolation, therefor the frontend circuits have floating grounds to prevent ground current flowing through human body. The amplitude of excitation current is strictly controlled and monitored to be far less than the medical safety limit, so the human body under test can be safely protected from electrical damage. The fundamental lung-phantom experiments have verified the tomographic ability of the developed SJTU Mk-1 EIT system. Static phantom experiments show clear conductivity image across the frequency range of 1 kHz to 1 MHz. It is also demonstrated that significant imaginary image can be obtained at frequencies of 500 kHz and above. The total cost of the prototype system is less than 2.9k USD which is much less than the 25k USD of the four NI modules used in some published EIT system. The presented prototype system is a little cumbersome, though, by upgrading the control module with wireless communication and replacing the frontend modules with specialized integrated circuit chips, the SJTU Mk-1 system could be made wearable and wireless for daily bedside monitoring, which will be the next target of our research. Acknowledgment This research was financially supported by the National Natural Science Foundation of China (Grant No. 61371017). References [1] R.H. Bayford, Bioimpedance tomography (electrical impedance tomography), Annu. Rev. Biomed. Eng. 8 (2006) 63–91, http://dx.doi.org/10.1146/annurev. bioeng.8.061505.095716. [2] R.H. Bayford, A. Tizzard, Bioimpedance imaging: an overview of potential clinical applications, Analyst 137 (2012) 4635, http://dx.doi.org/10.1039/ c2an35874c. [3] D.T. Nguyen, C. Jin, a Thiagalingam, a L. McEwan, A review on electrical impedance tomography for pulmonary perfusion imaging, Physiol. Meas. 33 (2012) 695–706, http://dx.doi.org/10.1088/0967-3334/33/5/695. [4] R.J. Halter, A. Hartov, S.P. Poplack, R. Diflorio-Alexander, W.A. Wells, K.M. Rosenkranz, R.J. Barth, P.A. Kaufman, K.D. Paulsen, Real-time electrical impedance variations in women with and without breast cancer, IEEE Trans. Med. Imaging 34 (2015) 38–48, http://dx.doi.org/10.1109/TMI.2014.2342719. [5] M.S. Campisi, C. Barbre, A. Chola, G. Cunningham, J. Viventi, Breast cancer detection using high – density flexible electrode arrays and electrical impedance tomography, 36th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. (2014) 1131–1134, http://dx.doi.org/10.1109/embc.2014.6943794. [6] R. Li, J. Gao, Y. Li, J. Wu, Z. Zhao, Y. Liu, Preliminary study of assessing bladder urinary volume using electrical impedance tomography, J. Med. Biol. Eng. 36 (2016) 71–79, http://dx.doi.org/10.1007/s40846-016-0108-1. [7] D.S. Holder, Electrical impedance tomography (EIT) of brain function, Brain Topogr. 5 (1992) 87–93, http://dx.doi.org/10.1007/BF01129035. [8] A.J. Wilson, P. Milnes, A.R. Waterworth, R.H. Smallwood, B.H. Brown, Mk3.5: a modular, multi-frequency successor to the Mk3a EIS/EIT system, Physiol. Meas. 22 (2001) 49–54, http://dx.doi.org/10.1088/0967-3334/22/1/307. [9] A. McEwan, A. Romsauerova, R. Yerworth, L. Horesh, R. Bayford, D. Holder, Design and calibration of a compact multi-frequency EIT system for acute stroke imaging, Physiol. Meas. 27 (2006) S199–S210, http://dx.doi.org/10. 1088/0967-3334/27/5/S17. [10] M. Goharian, M. Soleimani, A. Jegatheesan, K. Chin, G.R. Moran, A DSP based multi-frequency 3D electrical impedance tomography system, Ann. Biomed. Eng. 36 (2008) 1594–1603, http://dx.doi.org/10.1007/s10439-008-9537-5. [11] N. Ranade V, D.C. Gharpure, Design and development of instrumentation for acquiring electrical impedance tomography data, 2nd Int. Symp. Phys. Technol. Sensors, IEEE, 2015 (2015) 97–101, http://dx.doi.org/10.1109/ISPTS. 2015.7220091.
Y. Ma et al. / Biomedical Signal Processing and Control 39 (2018) 416–423 [12] A. McEwan, G. Cusick, D.S. Holder, A review of errors in multi-frequency EIT instrumentation, Physiol. Meas. 28 (2007) S197–S215, http://dx.doi.org/10. 1088/0967-3334/28/7/s15. [13] T.I. Oh, H. Wi, D.Y. Kim, P.J. Yoo, E.J. Woo, A fully parallel multi-frequency EIT system with flexible electrode configuration: KHU Mark2, Physiol. Meas. 32 (2011) 835–849, http://dx.doi.org/10.1088/0967-3334/32/7/S08. [14] R.J. Halter, A. Hartov, K.D. Paulsen, A broadband high-frequency electrical impedance tomography system for breast imaging, IEEE Trans. Biomed. Eng. 55 (2008) 650–659, http://dx.doi.org/10.1109/tbme.2007.903516. [15] M. Wang, Y. Ma, N. Holliday, Y. Dai, R.A. Williams, G. Lucas, A high-performance EIT system, IEEE Sens. J. 5 (2005) 289–298, http://dx.doi. org/10.1109/JSEN.2005.843904. [16] T.I. Oh, E.J. Woo, D. Holder, Multi-frequency EIT system with radially symmetric architecture: KHU Mark1, Physiol. Meas. 28 (2007) S183–S196, http://dx.doi.org/10.1088/0967-3334/28/7/s14. [17] S. Khan, P. Manwaring, A. Borsic, R. Halter, FPGA-based voltage and current dual drive system for high frame rate electrical impedance tomography, IEEE
[18]
[19]
[20]
[21]
423
Trans. Med. Imaging 34 (2015) 888–901, http://dx.doi.org/10.1109/tmi.2014. 2367315. L. Miao, Y. Ma, X. Xiong, X. Chen, T. Han, P. Cai, P. Qin, X. Ji, A new symmetric semi-parallel electrical impedance tomography (EIT) system – i: the design, IEEE Int. Conf. Imaging Syst. Tech., IEEE, 2016 (2016) 19–23, http://dx.doi.org/ 10.1109/IST.2016.7738191. L. Miao, Y. Ma, X. Chen, X. Xiong, P. Cai, P. Qin, X. Ji, T. Han, New symmetric semi-parallel electrical impedance tomography (EIT) system – II: the performance, IEEE Int. Conf. Imaging Syst. Tech., IEEE, 2016 (2016) 283–288, http://dx.doi.org/10.1109/IST.2016.7738238. P. Hua, E.J. Woo, J.G. Webster, W.J. Tompkins, Using compound electrodes in electrical impedance tomography, Biomed. Eng. IEEE Trans. 40 (1993) 29–34, http://dx.doi.org/10.1109/10.204768. M. Wang, Inverse solutions for electrical impedance tomography based on conjugate gradients methods, Meas. Sci. Technol. 13 (2001) 101–117, http:// dx.doi.org/10.1088/0957-0233/13/1/314.