A versatile and reproducible multi-frequency Electrical Impedance Tomography system 

A highly versatile EIT system, nicknamed the ScouseTom, has been developed. The system allows control over current amplitude, frequency, number of electrodes, injection protocol and data processing. A Keithley 6221 current source is used, along with a 24-bit EEG system for voltage recording. Custom PCBs interface with a PC to control the measurement process, electrode addressing and triggering external stimuli. The performance of the system has been characterised using resistor phantoms in experiments representative of human scalp recordings, with an overall SNR of 77.5 dB (n=343), stable across a four hour recording and across frequencies from 20 Hz to 20 kHz. The ScouseTom was used successfully in four modes of operation: time difference, triggered averaging, multi-frequency EIT and impedance spectrum measurements, in experiments investigating stroke and evoked potentials in both rat and human recordings. The experimental procedure is controlled by software and is readily adaptable to new paradigms. Where possible, commercial or open-source components have been used, to minimise the cost and complexity in reproduction. All of the hardware designs and software for the system have been released under an open source licence, encouraging contributions and allowing for rapid replication.


Electrical Impedance Tomography

Electrical Impedance Tomography (EIT) is a medical imaging technique which reconstructs the internal conductivity of an object from boundary voltage measurements (Metherall 1996). Existing clinical uses of EIT include imaging the lung (Frerichs 2000), liver (YOU 2009) and breast (Halter), with commercial EIT systems now available for clinical use. Brain EIT in other active field of research, with development in applications including epilepsy (Vongerichten 2016, Fabrizi 2006), acute stroke (Dowrick 2016), traumatic brain injury (Manwaring 2013, Fu 2014), evoked potentials (EPs) (Aristovich 2016) and activity in peripheral nerves (citation not found: KirillStockholm).

EIT Hardware

A typical EIT system consists of a current source, a voltage measurement unit, switching circuitry to address multiple electrodes, and a controller to automate the measurement process. A sinusoidal current of a defined amplitude and frequency is injected through a pair of electrodes, with voltages measured at all electrodes. Typically, a single ‘measurement’ is defined as the demodulated amplitude of the voltage at a single electrode, averaged over a chosen number of sinewave periods. These voltage measurements can be obtained either serially, or in parallel on all electrodes. The process is repeated for a number of different pairs of injection electrodes, where the particular sequence is referred to as the ‘protocol’. The complete data set consists of n voltage measurements, equal to the number of injection pairs multiplied by the total number of electrodes. While it is possible for a single frame of data to be used to reconstruct a conductivity profile of the target object, it is more common for to use the difference between two separate frames, producing an image of the change in conductivity.

Existing EIT systems include the KHU (Wi 2014), fEITER (McCann 2011), Dartmouth EIT System (Khan 2015), Xian EIT system (Xuetao 2005), Swisstom Pioneer Set (Swisstom AG, Switzerland) and a number of systems previously developed at UCL (Oh 2011) (McEwan 2006). Where available, the key features of each system have been summarised in table (\ref{TableEITSystems}). Values for noise and output impedance have been taken from the relevant literature for each device where available. The testing method reported for each device varies, as such some caution should be used when comparing figures for different systems.

\label{TableEITSystems}Existing EIT Systems
System Electrodes
Frames per
Max Output
KHU 16-64 Parallel 11-500k 100 80dB-120dB 5.75M
fEITER 32 Parallel 10k 100 90dB -
Dartmouth 32 Parallel 1k-1M 100 80dB-100dB 100k
Pioneer Set
32 Parallel 50k-250k 50 - -
Xian >16 Sequential 1.6k - 380k - >80dB 2M
UCL MK2.5 32 Sequential 20-256k 26 80dB 1M

With the exception of those developed at UCL, the systems described were built primarily for EIT imaging in the torso, with little emphasis placed on capturing the data required for EIT imaging of the brain. While the optimal measurement paradigms for EIT of the head and fast-neural activity have not yet been established, the technical requirements for a range of experimental paradigms are known (table \ref{table_requirements}). In general, the technical implementations of existing EIT systems make translation to brain EIT applications difficult. In particular, EIT systems typically perform all demodulation in hardware, transmitting only the amplitude averaged over a chosen number of periods of the carrier frequency. Therefore, it is not possible to obtain a continuous impedance signal or record additional signals such as EEG, both of which are necessary for EIT of fast-neural activity (Aristovich 2016). Further, performing the signal processing in hardware severely limits the ability to adjust parameters such as carrier frequency, measurement speed, and filter bandwidth to meet the different application requirements.

\label{table_requirements}Example of experimental requirements
Mode Frequency Signal Bandwidth Amplitude Extra Data Electrodes
Stroke/Head Injury 100Hz-10kHz Frequency dependent >1mA Voltage drift 32
Evoked Potentials 2kHz 2kHz 50uA Action potentials 128
Epilepsy 2kHz
Slow - 10Hz
Fast - 2kHz
50uA EEG/ECoG 32-128
Nerve 5kHz 3kHz 50uA Action Potentials 32


The motivation behind this work was to develop an EIT system that could be used primarily for imaging in the head and brain, while offering the maximum versatility, such that it could be easily reconfigured for different experimental methods. The following criteria were established:

  • Arbitrary control over current amplitude, up to 10 mA and frequency in the kHz range. The current levels required for EIT can vary from tens of \(\mu\)A, for epicortical recordings in rats (Oh 2011), to several mA when using human scalp electrodes (Tidswell 2001).

  • Parallel recording of all voltages for processing of data ’offline’, allowing for additional signals, such as EEG/ECoG to be recorded alongside the EIT signal.

  • Noise, frame rate and performance characteristics comparable to existing EIT systems. Target resistance changes of \(\approx\)0.1 % have been previously identified for EIT recordings of fast neural activity related to neuronal depolarization and scalp recordings of epilepsy (Oh 2011, Fabrizi 2006). This requires stable current injection, \(<\)0.1% noise, and voltage recording with accuracy of \(\approx\)100nV.

  • Variable electrode count, up to a maximum of 256 electrodes.

  • Ability to synchronise injection/recording with external triggers (whisker stimulation, visual, auditory). The phase of injected current should be randomised with respect to the stimulation, to minimise the phase related artefacts in EP recordings (Aristovich 2015).

  • Reconfigurable modes of operation, to allow for new functionality to be introduced as a later date. Ideally, this should be achievable through software or firmware changes only.

  • Easily reproducible. Currently, construction of a non-commercial EIT system can take several months and existing publications on EIT systems typically lack enough detail to allow replication. A system which can be easily replicated using a mixture of off the shelf equipment, alongside open source software and hardware designs, will significantly reduce the workload and allow new systems to be assembled in a matter of weeks.

EIT applications

Four main data collection strategies were identified for EIT experiments fig \ref{Modes}. The most common method of data collection in EIT is time difference, where multiple data sets are obtained at different points in time, i.e. before and after the introduction of some perturbation, and the change in impedance over time is reconstructed. Such is the prevalence of this method, it can be considered the ’standard’ mode of operation, being used in the vast majority of successful experiments in the field (Bayford 2012).

The second method is time difference imaging, using an external triggered stimulus to allow for coherent averaging. Data collection is synchronised to an external stimulus (e.g. whisker stimulator in rats, or visual EPs in humans) which is also recorded. This method, first developed by Oh et al. (2011) is used when impedance signals have low signal to noise ratio (SNR) and millisecond duration. Current is continuously injected, while the stimulus is repeated, generating multiple evoked potentials at different time points. Through coherent averaging, an impedance signal is obtained at each sample interval, yielding millisecond resolution.

In certain applications such as stroke (Romsauerova 2006) or breast imaging (Halter), where it is not possible to obtain a reference frame, the change in tissue impedance over frequency is instead reconstructed. To collect Multi frequency EIT data a complete injections protocol is repeated for a number of frequencies, commonly 10 or 20, sufficient to capture the differences in tissue spectra. The ordering of frequencies can be fixed, or randomised and the injection period of each particular frequency can be specified. Composite waveforms, consisting of multiple frequencies such as those implemented in the existing KHU (Wi 2014) and UCL (McEwan 2006) systems, can also be generated. These methods are substantially less robust to modelling and instrumentation errors than standard EIT methods, and thus place more stringent requirements on the algorithms and hardware (Ahn 2011) (Malone 2014).

The final data collection method is a frequency sweep. While not directly used for EIT imaging, this method has previously been used to assist EIT experiments, allowing for impedance characterisation in-situ taking the place of a stand-alone impedance analyser (Gabriel 2009), and selecting an optimal measurement frequency (Vongerichten 2013). The injection frequency is incremented over a given range and step size, with measurements taken at each point. Both increasing and decreasing sweeps are possible, as is randomised ordering.

\label{Modes} Modes of operation of the ScouseTom EIT system. Standard: Initial dataset collected at time T0, used as reference for subsequent datasets. Triggered: Current continuously injected simultaneous to multiple stimuli, with full dZ signal obtained through coherent averaging. MultiFrequency Data sets collected at multiple frequencies, with initial frequency F0 used as reference for subsequent measurements at higher frequencies. Characterisation Injection frequency is incremented across total range of interest

Experimental Design

Experiments were initially carried out on a resistor phantom to characterise the system across a range of frequencies. Subsequently, the system was utilised in all four EIT applications previously described, and the performance evaluated in each case. In some instances, the data were collected as part of other studies, Where the main aim was successfully imaging the impedance change of interest. In these cases additional analysis of the data was required to extract the relevant performance characteristics.

For the initial resistor phantom measurements, direct comparisons with systems reported in the literature are difficult to perform as the exact testing conditions and analyses are not always explicitly stated. Therefore, phantoms and measurement paradigms which represent realistic usage in brain EIT applications were chosen. Specifically, the injected current complied with IEC 60601-1 (Commission 2002) safety limits, the number of measurements and the averaging time was chosen to match those of real experiments. All voltages, rather the subset with highest SNR, were considered in subsequent analysis, with the exception of those with sufficiently low amplitude which are routinely neglected during reconstructions (Packham 2012). Thus the results represent the realistic best case performance of the system in a animal or human experiment, as opposed to the maximum achievable on a test bench. The characteristics of interest in these experiments were the noise and drift in measurements over time, and the reciprocity error (RE), common metric for accuracy of EIT systems (Wi 2014). With a purely resistive test object, the amplitude of the measured voltages should be frequency independent. In reality, the components any system will exhibit some variations with frequency, necessitating some degree of calibration in post processing (Wi 2014, McEwan 2006). To study the frequency dependence of the ScouseTom, changes in amplitude across frequency were compared to the theoretical changes resulting from the anti-aliasing filters in two commercial EEG systems.

Data was collected in healthy subjects using scalp EEG electrodes to represent the expected resting noise during clinical scalp EIT recordings (Fabrizi 2006, Fabrizi 2007, Romsauerova 2006). Additionally, the suitability of the system for imaging impedance changes resulting from stroke during a feasibility study by Dowrick et al. (2016) was assessed. The system was then utilised in measurements of evoked activity in the rat somatosensory cortex, using an established methodology (Oh 2011, Aristovich 2016). Multi-frequency data were collected as part of a larger study in stroke patients in collaboration with University College London Hospital (UCLH). The noise performance of the system was evaluated across a frequency range previously used in a simulation study (Malone 2014a), with a subset of frequencies recorded for longer time period to assess the degradation in data quality in real clinical measurements compared to those from a laboratory. Finally, the system was used in the impedance sweep mode of operation in a study by Dowrick et al. (2015), characterising the impedance of healthy and ischaemic rat brain in vivo.

System Design

The ScouseTom EIT system, figure \ref{STOverview} comprises a commercial current source and EEG amplifier, alongside custom switching/control circuitry and software. The system allows for software control over all aspects of the EIT measurement setup: protocol, injection duration, frequency, amplitude. Time and frequency difference EIT data can be collected, at frequencies between 5Hz and 20kHz, with a frame rate \(>\) 100 per second achievable.

\label{STOverview} Overview of the ScouseTom system. The two bespoke PCBs (highlighted) control programming of the current source, multiplex the current injection channels and randomise the stimulation trigger with respected to the phase of the injected current. Voltages are recorded in parallel by the EEG system and stored on the PC for offline processing.

Current source

The Keithley 6221 current source (Keithley Instruments, Inc.) is able to produce stable, low noise AC currents in the range 2 pA - 100 mA, up to a frequency of 100 kHz; has a rated output impedance up to 1 G\(\Omega\) supports external triggering and phase marking, and is controllable via serial, USB and Ethernet connections. The versatility of the system as a whole is largely as a result of this wide operating range of the Keithley 6221. This range covers the majority of EIT applications, with the exception of those operating at high frequencies \(>\) 100 kHz, such as breast (Khan 2015).

Voltage recording

\label{table_requirements}The requirements for voltage recording are parallel data collection, low noise and the ability to save data for offline processing. EEG amplifiers offer an effective off the shelf solution with high performance systems such as the BioSemi ActiveTwo (Biosemi, Netherlands), actiCHamp (Brainproducts GmbH, Germany) and g.tec HIamp (g.tec medical engineering GmbH, Austria) offering 24-bit resolution and a channel count up to 256. Each system offers a PC GUI for saving data to disk, data streaming over TCP/IP and the option to write custom software to interface with the device. These specifications come at the expense of maximum bandwidth, which in practice is limited by hardware anti-aliasing filters with typical cut off frequencies a tenth or a fifth of the sampling rate, table \ref{tableeeg}. While this is sufficient for all brain EIT applications, table , it may preclude use of the system in applications such as lung ventilation, which typically employ frequencies at 50 kHz or above (Frerichs 2000). For the experimental work presented here, either the BioSemi ActiveTwo (Biosemi, Netherlands) or actiCHamp (Brainproducts GmbH, Germany) system was used for voltage recording.

\label{tableeeg}Summary of EEG amplifier specifications
BioSemi actiCHamp g.tec HIamp
Max. Sampling Rate 16kHz 100kHz 38.4kHz
Resolution 24-bit 24-bit 24-bit
Max. Channel Count 256 256 256
Input Range +/- 262mV +/- 400mV +/- 250mV
Anti-aliasing 3kHz 20kHz/8kHz\({}^{1}\) 19.2kHz

1. Anti-aliasing filter dependent on actiCHamp hardware version.

Controller and switch network

The system controller is based upon the Arduino development platform (Arduino Due - Arduino LLC) and two bespoke PCBs: a controller board or ”shield” and a switch network board (highlighted in figure \ref{STOverview}). The controller shield contains the circuitry required for isolating the communication with other components of the system, namely RS232 and trigger-link connections with the current source, TTL with the EEG systems, and SPI connection to the switch networks. Due to the modular nature of the design and the variation of use cases, each connection was separately isolated from the mains supply, Figure \ref{STBlocks}, to ensure correct isolation regardless of experimental setup.

The current source is programmed via RS232 by the controller, with external triggering and zero-phase marker signals transmitted via a trigger-link cable. The external triggering enables the phase of the injected current to be reset upon switching, or shorter injections to be retriggered by an external pulse. The pulses controlling the stimulation for use in Triggered-Averaging mode are sent by the Arduino controller, with user programmable width from 1.5 \(\mu\)s to \(>\) 10 S. The time between these pulses is also set by the user, but the precise timing is modified by the controller following the suggestions in (Aristovich 2015) to to occur at random time with respect to the phase of the injected current. This is achieved by offsetting the trigger pulse by a random phase with respect to the Zero-Phase Marker received from the Keithley 6221. Alongside this TTL trigger, a simultaneous stimulation pulse with a programmable range of 3 to 18V can be sent for direct electrical stimulation.

Coded pulses sent by the controller are recorded by the EEG system and used as reference during data processing. Pulses are sent to indicate the start and stop of current injection, the switching of electrode pairs, changing of injected frequency, stimulation triggers, and the out-of-compliance status of the current source.

The switch networks comprise two individual series of daisy chained ADG714 CMOS switches (Analog Devices, USA), one each for the source and sink connections from the current source. The switch networks themselves can be daisy chained together, enabling current injection between any two electrodes from a total of 128. The switchboard is battery powered, and communicates with the controller through a digitally isolated SPI bus, figure \ref{STBlocks}. The possible time between switching electrode injection pairs ranges from 100 \(\mu\)S to approximately 70 minutes for 128 channels, which is more than sufficient to meet the frame rate requirements all use cases.

\label{STBlocks} Overview of system architecture

Controller software

As the Arduino platform is not hardware specific, porting the software to another device with different architecture is comparatively simple, and does not constrain the system to a specific board in future iterations. Currently, the PC software for serial communication with the controller is written in MATLAB (The MathWorks Inc.) but the commands can be easily replicated in another language.

Data processing software

In conventional EIT systems, demodulation of the AC signal is performed in hardware, and only the averaged value for each short injection is transmitted and stored in the PC; the user has little control over the parameters used during processing. Whilst it is possible to alter the firmware for research devices such as the KHU (Wi 2014) and UCLH (McEwan 2006), technical constraints such as on-board RAM and limited processing time impact the versatility of these systems.

As the ScouseTom system stores the continuous modulated voltages captured via an EEG system, the demodulation and data processing must be implemented in PC software, currently implemented in MATLAB after data collection. The software employs Zero-phase IIR band pass filtering and the Hilbert transform to produce the envelope and phase of the amplitude modulated signal. The simultaneous EEG signal is obtained by low pass filtering the original signal. This also allows the user control over all the parameters during demodulation and averaging, demodulation method, centre frequency, bandwidth, filter type etc. This versatility is necessary to process data in EP studies, when the system is used in Triggered mode, as the measurement time is orders of magnitude longer than that of conventional EIT, and the synchronised EEG signal is essential.

Electrode connectors

Standard 9 and 37 D-Sub connectors are used for internal connections within the system. Custom connectors and PCBs were created to allow connection to a range of common electrode interfaces, including EEG electrodes, depth electrodes and Omnetics (Omnetics Connector Corp. MN, USA) devices. Bespoke connectors and cabling can be added to the system for use with non-standard electrode arrays.


Resistor phantom - system characterisation

To assess the noise and drift characteristics of the system, experiments were performed on the Cardiff phantom (Griffiths 1995), configured as a purely resistive load. Voltages were recorded with the BioSemi EEG system, in a configuration used during recordings on the human scalp. A 100 \(\mu\)A injection current at 2khz was used, for 100 ms per measurement, over the course of four hours. The injection electrode protocol was one previously designed for recordings in the head, with 34 different injection pairs and 16 measurement electrodes, for a total of 544 voltage measurements (Fabrizi 2009). This was reduced to 363 after rejection of measurements on injection electrodes and boundary voltages below 250 \(\mu\)V. The injection protocol took was repeated every 3.4 seconds over the course of four hours, resulting in a total of 4235 frames.

As with other EIT systems (Oh 2007) the noise was considered across three time intervals: 100 frames (approximately five minutes) representing a typical short term measurement, and one and four hours. These longer recordings replicate the use of the system in long term time difference recordings in monitoring applications (Fu 2014) (Adler 2012) as well as in triggered mode during EP studies (Aristovich 2016). To evaluate the drift in performance over time, the change in voltages over the four hour recording was also calculated, as well as the SNR for each block of 100 frames. The reciprocity error (RE), where the ratio of the impedance measured with the voltage and current electrode pairs swapped, was calculated using the same 100 frames, and expressed as an absolute error in percentage.

Resistor phantom - frequency response


The same resistor phantom was used to investigate the frequency response of the system, using two EEG systems. Data was collected at 15 frequencies between 20 Hz and 2 kHz using the BioSemi ActiveTwo and 33 frequencies between 20 Hz and 20 kHz for the actiCHamp system. To reduce the data collection time, six injection pairs from the protocol used previously were chosen, resulting in a total of 64 voltage measurements for analysis. The time per measurement was frequency dependent, equivalent to 32 sine-wave periods for frequencies below 200 Hz and 64 periods for those above. The whole protocol was repeated for 10 frames, with the frequency order randomised within each frame. The current amplitude was 100 \(\mu\)A for all frequencies. The voltages recorded were normalised with respect to the amplitude at 20 Hz, and expressed as a percentage. To assess the noise frequency dependence of the noise performance, the SNR across all frequencies was also calculated.

Experimental data

To evaluate the system experimentally and clinically, experiments were performed using each of the four modes of operation. Certain methods were common across some experiments which are detailed here.

Scalp recordings - preparation

All recordings on the human scalp were performed using 32 EEG electrodes (EasyCap, Germany), using the configuration described by Tidswell et al. (2001), which includes 21 locations from the EEG 10-20 standard (Jasper 1958) and 11 additional electrodes. For these experiments, the locations were updated to match the nearest equivalents in either the 10-10 or 10-5 extensions (Oostenveld 2001). Each electrode site was first cleaned with surgical spirit, then abraded using Nuprep gel (Weaver and Co., USA), with the electrode finally affixes using Elefix conductive paste (Nihon Kohden, Japan).

EIT reconstruction

EIT images were reconstructed using the same methodology used in other studies by the UCL group (Dowrick 2016, Aristovich 2016, Aristovich 2014). First the ‘forward problem’ and sensitivity matrix was calculated using the Parallel EIT Solver (Jehl 2014), in a c. 4 millions element tetrahedral mesh. The linear inverse solution was obtained using zeroth-order Tikhonov regularisation, with the hyperparameter \(\lambda\) chosen through cross-validation. The conductivity changes were reconstructed in a lower resolution hexahedral mesh of approximately 100,000 elements, and we were subsequently post-processed using a noise based correction to produce images of significance values (Aristovich 2016).

Time difference - imaging haemorrhage

’Standard’ time difference EIT data were recorded in an anesthetised rat, from 40 spring-loaded gold plated electrodes placed on the skull in an even distribution. A model of intracerebral haemorrhage was used wherein 50 \(\mu\)l of autologous blood was injected via a cannula into the brain at a rate of 5 \(\mu\)l per minute \cite{Dowrick_2016}. Given the resistance of blood and grey matter at 1 Khz is approximately 1.4 \(\Omega{\text{m}}^{-1}\) and 10 \(\Omega{\text{m}}^{-1}\) respectively \cite{Gabriel_2009}, this produced up to a seven-fold increase in conductivity localised around the injection site. A current of 100\(\mu\)A at 2kHz was injected, using a protocol of c. 60 injection pairs, with each injection lasting 1 s. The specific protocol was adapted for each individual experiment, dependent upon the quality of electrode contacts. A complete frame of EIT data was recorded every minute, over a total of 30 minutes, including 10 minutes prior to the injection of blood. Time difference images were reconstructed with respect to the baseline, for every frame recorded after starting the injection of blood. This procedure was repeated in a total of 7 rats.

Time difference - scalp recordings


To assess the performance of the system in a clinically realistic scenario, EIT recordings were made in 10 healthy volunteers in a seated position. Current of 1.2 kHz and 160 \(\mu\)A was injected for 54 ms, equivalent to 64 sinewave periods, between 31 pairs of electrodes, resulting in a complete frame every 1.6 seconds. The injection pairs were chosen to maximise both the number of independent measurements and the overall magnitude of the voltages \cite{Malone2014a}. A complete frame consisted of a total of 930 measurements, 540 of which were considered in subsequent analysis, after rejecting measurements below 1 mV. A total of 60 frames were recorded over the course of 20 minutes for each subject.

Triggered averaging - rat somatosensory cortex


The ScouseTom system was utilised in measurements of evoked activity using epicortical electrode arrays in anesthetised rats, repeating methodology from previous studies (Aristovich 2016) (Vongerichten 2016). Arrays of platinised stainless-steel electrodes embedded in silicon, with 0.6 mm diameter contacts spaced 1.2 mm apart were placed directly onto the surface of the brain, targeting the somatosensory cortex. Whilst previous experiments had employed a single grid of 30 electrode contacts on a single side of the brain, in this study two grids of 57 electrodes were used, one on each hemisphere, for a total of 114. The activity was induced via electrical stimuli to the forepaw, delivered in 10 mA pulses of 1 ms duration at a rate of 2 Hz, triggered by the ScouseTom controller. Current was injected at 1.7 kHz with 50 \(\mu\)A amplitude, for 30 seconds across a pair of electrodes for a total of 60 stimuli, with voltages on 114 electrodes recorded in parallel using the ActiChamp system with 25 kHz sampling rate. This process was repeated for c. 50 pairs of injection electrodes over the course of 25 minutes to produce a complete data set of c. 7000 voltages. The signals were demodulated with 1 kHz bandwidth, yielding 2 ms time resolution, and coherent averaging was performed on 60 500 ms trials centered around the time of stimulation. EIT images were reconstructed at each 2ms time point, using 10 ms prior to stimulation as the baseline. To allow for a comparison to previous studies with previous systems, the noise was also calculated after demodulation using a reduced bandwidth of 250 Hz (Oh 2011).

Multifrequency - scalp recordings

Multifrequency EIT data was collected as part of clinical trial in collaboration with the Hyper Acute Stroke unit (HASU) at University College London Hospital (UCLH). The injection protocol comprised the same 31 pairs used in previous scalp recordings, section \ref{methodsTD}. Current was injected at 17 frequencies evenly spaced across the usable range of the BioSemi system, i.e. from 5 Hz to 2 kHz. As with previous multifrequency experiments, section \ref{MethodsRPFreq}, the length of each injection was frequency dependent to reduce the total recording time. Current was injected for the equivalent of 32 periods at carrier frequencies below 200 Hz, 64 periods between 200 and 1 kHz, and 128 periods above 1 kHz. The amplitude of injected current varied with frequency as per the guidelines in IEC 60601 (Commission 2002), except frequencies below 200 Hz which, as in previous studies (McEwan 2006), were reduced by half to ensure the current was not perceptible and to avoid saturation of the EEG amplifier due to the larger contact impedance at these frequencies. The protocol was repeated three times, taking a total of 20 minutes to complete. A further dataset using only three frequencies, 200 Hz, 1.2 kHz and 2 kHz, was also recorded on each patient, to better evaluate the frequency dependence of the noise and drift performance of the system. In this case, a total of 60 frames were collected over the course of 25 minutes. Data were collected in 23 patients, for a total of 31 datasets with 17 frequencies and 29 longer duration, 3 frequency recordings. After rejection of negligible channels, each dataset comprised 540 voltages per frequency.

Impedance spectrum characterisation - ischaemic rat brain

Impedance spectrum measurements were made with the ScouseTom system (Dowrick 2015), in the kHz range on healthy (n=112 voltage measurements in 4 rats) and ischaemic (n=56 in 2 rats) rat brain. Using an 30 contact epicortical electrode array, section \ref{MTrig}. A current of 100\(\mu\)A was injected through a single pair of electrodes, located on opposite corners of the array. The frequency of injected current was increased in 5Hz intervals from 1Hz to 100Hz, 10Hz intervals from 100Hz to 1000Hz and 50Hz intervals between 1000Hz and 3000Hz, for a total of 136, with a minimum of 50 periods of the waveform recorded at each frequency. This sweep was repeated in ascending and descending order and finally with random ordering. Voltages from each electrode were averaged together at each frequency. The relative change in impedance, rather than the absolute value, was calculated, by comparing the voltages at each frequency.

Data presentation

Noise and drift were calculated using all measurements except those with negligible boundary voltages, which were removed before processing. Unless stated otherwise, all data is presented as mean \(\pm\) standard deviation across all measurements. Where applicable, the noise in repeated measurements is presented as three values, to aid comparisons to other systems in the literature: first as the standard deviation in absolute voltage, then as the ratio of the mean to standard deviation both as a percentage, and Signal-to-Noise ratio in dB. In accordance with previous studies, SNR was calculated differently for the triggered averaging experiment in the rat cortex. Instead the SNR was the ratio of the peak voltage change \(\delta z\) following stimulation to the standard deviation before stimulus (Oh 2011). The experimental setup for each experiment is summarised in table \ref{MSummary}.

\label{MSummary}Summary of experiments performed using the ScouseTom system
Experiment Current \(\mu\)A Frequencies (no.) EEG System Measurements Frames
100 2 kHz BioSemi 363 4235
20 Hz - 2 kHz (15)
& 20 Hz - 20 kHz (15)
BioSemi &
64 10
Time Difference Haemorrhage 100 2 kHz BioSemi 418-1381
10 (baseline)
6-50 (stroke)
Scalp 160 1.2 kHz BioSemi 540 60
50 1.7 kHz actiCHamp 5088
(500 ms signal)
Multifrequency Scalp 45 - 280 5 Hz - 2 kHz (17) BioSemi 540 3
Long term
90 - 280 200 Hz - 2 kHz (3) BioSemi 540 60
100 1 Hz - 3 kHz (136) BioSemi 112 3


Resistor Phantom - System Characterisation

The average signal amplitude across all measurements and repeats was 2.66 mV \(\pm\) .24 mV standard deviation. The noise in the first 100 frames was 0.356 \(\mu\)V \(\pm\) .058 \(\mu\)V, equivalent to 0.013 % \(\pm\) .002 % or 77.5 dB \(\pm\) 1.3 dB SNR. The noise increased over the longer term, with 0.637 \(\mu\)V \(\pm\) .064 \(\mu\)V or 0.024 % \(\pm\) .002 % and 72.4 dB \(\pm\) 0.66 dB for one hour, and 1.522 \(\mu\)V \(\pm\) .205 \(\mu\)V or 0.17 % \(\pm\) .074 % and 64.9 dB \(\pm\) 1.24 dB and four hour recordings. As has been found with other EIT systems (Oh 2007), this decrease in SNR was not observed when considering each consecutive block of 100 frames figure \ref{SNR}. Over the course of the four hour recording, the mean change in voltage was 5.61 \(\mu\)V \(\pm\) 1.04 \(\mu\)V or 0.21 % \(\pm\) 0.04 %. The reciprocity error (RE) was 0.42 %.

\label{SNR} SNR (mean and st. dev) per 100 frames (n=363 measurements) over entire 4 hour recording on Cardiff phantom

Resistor Phantom - Frequency response

Both EEG systems both contain sinc filters used for anti-aliasing, with cut off frequencies quoted at 1.7 kHz and 7.5 kHz for the BioSemi and ActiCHamp respectively. In both systems, frequency response correlated with the gain of the respective filter, figure \ref{freqresp}, with a mean error of 0.07 % and 0.09 % for frequencies below the cut off for the BioSemi and actiCHamp respectively. The amplitude decrease was 1.86 mV V or 21.0% from 20 Hz to 2 kHz for the BioSemi and 0.57 mV or 6.26% for the actiCHamp across the same range. Across the full 20 kHz range the total decrease was 9.15 mV or 99.6 %. Below these cut-off frequencies, there was no significant change in mean SNR across frequency, with 68.9 dB \(\pm\) 9.42 dB for the BioSemi and 71.4 dB \(\pm\) 12.6 dB for the actiCHamp.

\label{freqresp} Amplitude response of the ScouseTom system as recorded with the BioSemi and ActiChamp systems (n=64 measurements). In both cases the response matches the gain of the Anti-Aliasing filter within 0.1 %.

Time difference - imaging haemorrhage

The noise in the baseline recordings (n=115 in N=15) before intervention was 0.64 \(\mu\)V \(\pm\) 2.12 \(\mu\)V equivalent to 1.90 % \(\pm\) 3.6 % or 42.4 \(\pm\) 11.9 dB. The mean SNR per experiment (N=15) ranged from 28.4 to 62.1 dB. Over the 10 minute injection period, changes of 20.6mV \(\pm\) 5.4mV occurred (n=7) figure \ref{FigureStroke}, with a further increase of 2.2mV \(\pm\) 1.4mV over the remaining measurement time. The time course of the impedance change was consistent with the injection period of the blood into the brain. Physiologically representative localised impedance changes could successfully be reconstructed in 5 out of the 7 experiments, figure \ref{FigureStroke} b.

\label{FigureStroke} Single EIT dataset and image reconstruction for stroke experiment. Ten minutes of baseline data was recorded before haemorrhage onset, indicated by the dotted vertical line.

Replace this text with your caption

Time difference - scalp recordings

The signal amplitude across all subjects was 5.4 mV \(\pm\) 1.97 mV, and the drift during all recordings was 15.1 \(\mu\)V or 0.26 %. The noise overall across all subjects was 37.2 \(\mu\)V \(\pm\) 39.0 \(\mu\)V, or 0.71 % \(\pm\) 0.65 %. The noise in each subject ranged from 8.55 \(\mu\)V to 53.9 \(\mu\)V, or 0.18 % to 0.94 %. The equivalent SNR overall was 46.5 dB \(\pm\) 4.14 dB, ranging from 41.6 to 56.0 dB between subjects.

Triggered averaging - rat somatosensory cortex

The baseline noise was 0.37 \(\mu\)V \(\pm\) 0.048 \(\mu\)V or 0.009 % \(\pm\) 0.005 of the baseline. Significant impedance changes (\(\delta\)V \(>3\sigma\)), a single injection pair example in figure \ref{EPDZ}, ranged from 1.20 \(\mu\)V to 18.7 \(\mu\)V (mean 3.97 \(\mu\)V), equivalent to 3.13 to 33.0 (mean 8.86) SNR. EIT reconstructions with this dataset show the onset at 7ms approximately 1 mm below the surface within the primary somatosensory cortex (S1), before expanding to a larger volume reaching a maximum between 11-12 ms, figure \ref{EPRecon}, then spreading to adjacent areas and finally disappearing at approximately 18 ms. Using the reduced bandwidth of 250 Hz, the noise was 0.18 \(\mu\)V \(\pm\) 0.06 \(\mu\)V or 0.004 % \(\pm\) 0.002 of the baseline.

\label{EPDZ}Impedance changes in rat somatosensory cortex during evoked potentials using two 57 channel epicortical arrays, for a single pair of injecting electrodes. Channels with significant changes (\(dV>3\sigma\)) are highlighted.

\label{EPRecon} Example EIT image during electrical forepaw stimulation at T=11 ms after stimulation, corresponding to the peak in figure \ref{EPDZ}. Activity is centred in the primary somatosensory cortex (S1), matching expectations from established literature.


Multifrequency - scalp recordings

The variation in SNR across frequency, figure \ref{MFSNR}, was not significant (\(P<0.05\)), ranging from 44.1 to 45.5 dB. However, the reduction of current amplitude at frequencies below 200 Hz, reduced the SNR by 1dB compared to subsequent frequencies. The longer term recordings also did not demonstrate any significant frequency dependence (\(P<0.05\)) in SNR with 41.6 \(\pm\) 7.9 dB, 41.4 \(\pm\) 8.25 dB and 40.7 \(\pm\) 8.2 dB for 0.2 1.2 and 2 kHz respectively. Similarly the drifts 0.74 %, 0.61 % 0.68 % of were also not significantly variable across frequency.

\label{MFSNR} SNR of multi frequency scalp recordings on stroke patients (n=540 N=23)

Impedance Spectrum Measurement

Healthy brain tissue showed a non linear decrease of 15% impedance over 0-250Hz, with ischaemic brain showing a decrease of 7 % over the same range, with a more linear slope figure \ref{FigureSweep}. Above 250Hz, the impedance of both tissue types decreased at the same rate.

\label{FigureSweep} Impedance spectrum of healthy (n=112, 4 rats) and ischamic brain (n=56, 2 rats) as measured with ScouseTom system. Error bars correspond to the standard error.


System characterisation

Results from the resistor phantom with realistic loads demonstrated a short term SNR of 77.5 dB \(\pm\) 1.3 dB, with a reduction of 5 dB over an hour, and a total of 12.6 dB over the whole four hour period. As has been observed in existing EIT systems (Oh 2007), this apparent decrease is as a result of low frequency drifts, likely from temperature changes in the current source, rather than degradation of the performance of the system figure \ref{SNR}. Overall, the SNR was lower than figures reported for other systems, which are capable of measuring with SNR above 90 dB (Khan 2015, Wi 2014). The amplitude of the voltages recorded was 2.66 mV, orders of magnitude less than the input range of the EEG amplifier used (\(\approx\) 250mV), and did not take full advantage of the high dynamic range offered by these amplifiers. The other EIT systems in the literature incorporate a programmable gain amplifier before digitisation to overcome this problem, often necessitating detailed calibration before data collection. It is therefore possible that certain measurements with larger injected currents and voltage amplitudes could far exceed the SNR reported in this study. One of the few systems in the literature which targeted the same frequency range, and thus with the same constraints on current and voltage amplitude, demonstrated a noise of 0.1%, close to an order of magnitude greater than those reported here (McEwan 2006). The reciprocity error (RE) of 0.42 % is comparable to other EIT systems, despite the lack of calibration (Oh 2007, Wi 2014, Khan 2015). This is likely as a result of the comparatively low frequency used in this study, where the effect of stray capacitance in minimal.

The frequency response of the system as a whole matched that of the EEG systems used, and the SNR was consistent across frequency. This suggests that beyond correcting for the EEG amplifier gain, frequency specific calibration will not offer significant benefits to the accuracy of the system. If higher frequencies (\(>20\) kHz) were recorded using a different EEG system, then stray capacitance could no longer be ignored, and calibration would likely be required.

Time difference

The decrease in SNR in animal and human experiments is in agreement with previous studies \cite{fabrizi2007analysis} which demonstrated system noise accounts for approximately 5 %, with physiological noise dominating. As with the long term resistor phantom recordings, the drifts across the recordings reduced the equivalent SNR. This physiological variability is evident in the range of SNR between subjects in both rat and human subjects. The system noise is thus masked by electrochemical or physiological changes over the course of the experiment. Despite this, conductivity changes resulting from the haemorrhage model could be imaged and correctly localised in the majority of cases. The SNR in the scalp recordings was decreased further in some recordings by artefacts resulting from subject or electrode movement. Between patients the SNR ranged from 41.6 to 56.0 dB, equivalent to 0.18 % to 0.94 %, which was less than the 2 % observed by \citet{Romsauerova2006} using a similar current amplitude and frequency, and within the 37.9 to 62.8 dB range observed by \citet{Xu2011} using an order of magnitude greater amplitude at 50 kHz. Thus the ScouseTom system offers comparable or greater performance to existing systems, but with increased versatility in experimental settings.

Triggered averaging

The baseline noise in these recordings agreed with that observed in previous fast-neural activity experiments, which reported noise of 0.18 \(\mu\)V \(\pm\) 0.04 \(\mu\)V, using a bandwidth of 250 Hz (Oh 2011, Packham 2016) and averaging over 120 repeated stimuli. However, the results presented in these studies were obtained after averaging of only 60 trials, suggesting a decrease in background noise when using the ScouseTom system. This not only offers benefits in terms of signal quality, but shorter data collection times reduce the effects of physiological drifts and allow for longer injection protocols, potentially improving EIT reconstructions.

Previous experiments with this methodology with mechanical stimulation of the rat whisker by Aristovich et al. (2016) were found to be in good agreement with established literature, and cross validated with other imaging techniques both in initial onset (Armstrong-James 1991), and spread of activity (Petersen 2007). Whilst cross validation was not performed in this study, the literature is equally well established regarding forepaw stimulation, and the location and depth of the area of onset in figure \ref{EPRecon} matches expectations (Peeters 2001) (Masamoto 2007) (Lowe 2007). Due to the high density of electrodes required, existing 32 channel systems limit measurements to a single cortical area one one side of the brain. The increased electrode count of the ScouseTom enables greater coverage of the cortex, as well as the possibility of simultaneous recording with additional depth electrodes.


The stability of the system across frequency previously demonstrated in the resistor phantom experiments was also present in recordings in stroke patients, as SNR did not demonstrate a significant frequency dependence. However, overall the SNR did decrease compared to recordings in subjects at rest in section \ref{scalp}, which were largely a result of a greater prevalence of motion artefacts from patient movement. The stability over frequency and increased electrode movement were further demonstrated in the drifts over the 25 minute recordings, which were similar for all three frequencies. Whilst the imaging in these applications is still the subject of research (Malone 2015) (Jang 2015), the datasets collected with this system constitute the most comprehensive recorded, and form a basis for future research.


The frequency sweep mode of operation is effectively that of a dedicated impedance analyser, with the flexibility of addressing any pair of electrodes for current injection, and recording voltages on all electrodes in parallel. These results were consistent with previous studies in the area (Ranck 1963), (Logothetis 2007), and also allowed comparisons between two and four terminal measurements simultaneously.

Design criteria and technical limitations

The design criteria set out initially were necessarily broad in order for the system to adapt to the four modes of operation successfully. The configurability of the system meets the requirements for range of injected currents, electrode count, synchronisation and recording of simultaneous EEG signals. In doing so, the system has enabled experiments not previously possible, particularly when used in triggered averaging mode. The noise performance of the system was comparable to existing EIT systems, and was consistent across frequency. Whilst the \(<0.1\) % noise criteria was met in epicortical recordings, the noise in scalp recordings was closer to \(0.65\) %, which may preclude imaging epileptic seizures using the system in its current form. However, use of a higher frequencies would allow for higher currents to be used, as well as reducing in band physiological noise. Therefore, future experiments should be performed with higher bandwidth systems such as the actiCHamp or g.tec HIamp, with a higher carrier frequency.

The main limitation of the ScouseTom is the maximum usable frequency of 20kHz imposed by the bandwidth of the EEG systems. This range is sufficient for EIT of fast-neural and stroke, where the impedance contrast is limited to frequencies \(<\) 5 10 kHz (Malone 2014, Aristovich 2016, Vongerichten 2016), but other brain monitoring applications may benefit from the higher current injections afforded at frequencies \(>\) 50 kHz (Fabrizi 2006, Fu 2014, Manwaring 2013). Currently, the majority of the data processing is not performed in ”real time”, as with some commercially available systems. However, as most EEG recorders provide open source software, or allow streaming of data via TCP/IP, suitable software can be developed to allow for real time applications where needed.

Reproducibility and recommendations for use

All of the hardware designs, PCB layouts and associated software have been made available (see Appendix) to allow interested parties to replicate and modify the system. The ScouseTom system is quick to reproduce, as all but two components are commercially available, and the bespoke PCBs are simple compared to those used in other research EIT systems. Within the UCL group, assembly and testing of a new system once all the parts have been acquired can be performed with a week. The major expense is the appropriate EEG system, with the current source and custom PCBs costing around a quarter of the two systems used in this study. Therefore, the additional cost to those already performing electrophysiological experiments is minimal, as the system is compatible with most state-of-the-art EEG systems. Thus this system greatly reduces the cost and complexity for those interested in including impedance measurements into their experiments.


The ScouseTom, a new highly versatile EIT system has been utilised successfully in a range of experiments, particularly those involving EIT of the head or nerve. A key advantage of the system is the high level of control it offers over all aspects of the experimental process. This versatility is a result of the stable performance of the hardware across frequency and load, and adaptable modes of operation. The new system facilitated EIT experiments into time difference imaging of stroke, multi-frequency stroke type classification, stimulation-triggered evoked potentials. In all four modes of operation used, the performance of the system was comparable to, or exceeded that of existing systems in the literature. By incorporating commercial and open-source hardware and software where possible, the complexity in reproducing the system is minimised. All relevant schematics and software are available on an open source license, with reproduction and contribution encouraged.

Appendix - Hardware & Software Resources

All of the hardware and software is released under a GNU General Public License v3.0, with contribution and distribution welcomed.

Controller hardware schematics, PCBs and firmware, along with 3D printable case models and documentation available at https://github.com/EIT-team/ScouseTom

Software for processing data available at https://github.com/EIT-team/Load_data


  1. P Metherall, DC Barber, RH Smallwood, BH Brown. Three dimensional electrical impedance tomography. Nature 380, 509–512 Sheffield, 1996.

  2. Inéz Frerichs. Electrical impedance tomography (EIT) in applications related to lung and ventilation: a review of experimental and clinical activities. Physiol. Meas. 21, R1–R21 IOP Publishing, 2000. Link

  3. Fusheng YOU, Wanjun Shuai, Xuetao Shi, Feng Fu, Ruigang Liu, Xiuzhen Dong. In vivo Monitoring by EIT for the Pig’s Bleeding after Liver Injury. 110–112 In IFMBE Proceedings. Springer Science \(\mathplus\) Business Media, 2009. Link

  4. Ryan J. Halter. Development of a multi-high frequency electrical impedance tomography system for breast imaging.. Dartmouth College Library Press Link

  5. Anna N. Vongerichten, Gustavo Sato dos Santos, Kirill Aristovich, James Avery, Andrew McEvoy, Matthew Walker, David S. Holder. Characterisation and imaging of cortical impedance changes during interictal and ictal activity in the anaesthetised rat. NeuroImage 124, 813–823 Elsevier BV, 2016. Link

  6. L Fabrizi, M Sparkes, L Horesh, J F Perez-Juste Abascal, A McEwan, R H Bayford, R Elwes, C D Binnie, D S Holder. Factors limiting the application of electrical impedance tomography for identification of regional conductivity changes using scalp electrodes during epileptic seizures in humans. Physiol. Meas. 27, S163–S174 IOP Publishing, 2006. Link

  7. T Dowrick, C Blochet, D Holder. In vivobioimpedance changes during haemorrhagic and ischaemic stroke in rats: towards 3D stroke imaging using electrical impedance tomography. Physiol. Meas. 37, 765–784 IOP Publishing, 2016. Link

  8. Preston K Manwaring, Karen L Moodie, Alexander Hartov, Kim H Manwaring, Ryan J Halter. Intracranial electrical impedance tomography: a method of continuous monitoring in an animal model of head trauma. Anesthesia and analgesia 117, 866 NIH Public Access, 2013.

  9. Feng Fu, Bing Li, Meng Dai, Shi-Jie Hu, Xia Li, Can-Hua Xu, Bing Wang, Bin Yang, Meng-Xing Tang, Xiu-Zhen Dong, others. Use of electrical impedance tomography to monitor regional cerebral edema during clinical dehydration treatment. PloS one 9, e113202 Public Library of Science, 2014.

  10. Kirill Y. Aristovich, Brett C. Packham, Hwan Koo, Gustavo Sato dos Santos, Andy McEvoy, David S. Holder. Imaging fast electrical activity in the brain with electrical impedance tomography. NeuroImage 124, 204–213 Elsevier BV, 2016. Link

  11. Hun Wi, Harsh Sohal, Alistair Lee McEwan, Eung Je Woo, Tong In Oh. Multi-Frequency Electrical Impedance Tomography System With Automatic Self-Calibration for Long-Term Monitoring. IEEE Trans. Biomed. Circuits Syst. 8, 119–128 Institute of Electrical & Electronics Engineers (IEEE), 2014. Link

  12. H. McCann, S. T. Ahsan, J. L. Davidson, R. L. Robinson, P. Wright, C. J. D. Pomfrett. A portable instrument for high-speed brain function imaging: FEITER. In 2011 Annual International Conference of the IEEE Engineering in Medicine and Biology Society. Institute of Electrical & Electronics Engineers (IEEE), 2011. Link

  13. Shadab Khan, Preston Manwaring, Andrea Borsic, Ryan Halter. FPGA-based voltage and current dual drive system for high frame rate electrical impedance tomography. IEEE transactions on medical imaging 34, 888–901 IEEE, 2015.

  14. Shi Xuetao, You Fusheng, Fu Feng, Liu Ruigang, Dong Xiuzhen. High precision Multifrequency Electrical Impedance Tomography System and Preliminary imaging results on saline tank. In 2005 IEEE Engineering in Medicine and Biology 27th Annual Conference. Institute of Electrical and Electronics Engineers (IEEE), 2005. Link

  15. T Oh, O Gilad, A Ghosh, Martin Schuettler, David S Holder. A novel method for recording neuronal depolarization with recording at 125–825 Hz: implications for imaging fast neural activity in the brain with electrical impedance tomography. Medical & biological engineering & computing 49, 593–604 Springer, 2011.

  16. 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, S199–S210 IOP Publishing, 2006. Link

  17. Tom Tidswell, Adam Gibson, Richard H Bayford, David S Holder. Three-dimensional electrical impedance tomography of human brain activity. NeuroImage 13, 283–294 Elsevier, 2001.

  18. Kirill Y Aristovich, Gustavo S Dos Santos, David S Holder. Investigation of potential artefactual changes in measurements of impedance changes during evoked activity: implications to electrical impedance tomography of brain function. Physiological measurement 36, 1245 IOP Publishing, 2015.

  19. Richard Bayford, Andrew Tizzard. Bioimpedance imaging: an overview of potential clinical applications. Analyst 137, 4635–4643 Royal Society of Chemistry, 2012.

  20. A Romsauerova, A McEwan, L Horesh, R Yerworth, R H Bayford, D S Holder. Multi-frequency electrical impedance tomography (EIT) of the adult human head: initial findings in brain tumours, arteriovenous malformations and chronic stroke, development of an analysis method and calibration. Physiological Measurement 27, S147 (2006). Link

  21. Sujin Ahn, Tong In Oh, Sung Chan Jun, Jin Keun Seo, Eung Je Woo. Validation of weighted frequency-difference EIT using a three-dimensional hemisphere model and phantom. Physiological measurement 32, 1663 IOP Publishing, 2011.

  22. Emma Malone, Gustavo Sato dos Santos, David Holder, Simon Arridge. Multifrequency electrical impedance tomography using spectral constraints. IEEE transactions on medical imaging 33, 340–350 IEEE, 2014.

  23. C Gabriel, A Peyman, E H Grant. Electrical conductivity of tissue at frequencies below 1 MHz. Physics in Medicine and Biology 54, 4863–4878 IOP Publishing, 2009. Link

  24. A. Vongerichten, G. Santos, K. Aristovich, D. Holder. Impedance changes during evoked responses in the rat cortex in the 225–1575 Hz frequency range.. In XVth International Conference of Electrical Bioimpedance and XIVth conference on Electrical Impedance Tomography. (2013).

  25. International Electrotechnical Commission. IEC 60601-1. Medical Electrical Equipment: Part 1: General Requirements for Basic Safety and Essential Performance 2 (2002).

  26. B Packham, H Koo, A Romsauerova, S Ahn, A McEwan, SC Jun, DS Holder. Comparison of frequency difference reconstruction algorithms for the detection of acute stroke using EIT in a realistic head-shaped tank. Physiological measurement 33, 767 IOP Publishing, 2012.

  27. L Fabrizi, Alistair McEwan, E Woo, David S Holder. Analysis of resting noise characteristics of three EIT systems in order to compare suitability for time difference imaging with scalp electrodes during epileptic seizures. Physiological measurement 28, S217 IOP Publishing, 2007.

  28. Emma Malone, Markus Jehl, Simon Arridge, Timo Betcke, David Holder. Stroke type differentiation using spectrally constrained multifrequency EIT: evaluation of feasibility in a realistic head model.. Physiological measurement 35, 1051–66 (2014). Link

  29. T Dowrick, C Blochet, D Holder. In vivo bioimpedance measurement of healthy and ischaemic rat brain: implications for stroke imaging using electrical impedance tomography. Physiol. Meas. 36, 1273–1282 IOP Publishing, 2015. Link

  30. H Griffiths. A Cole phantom for EIT. Physiological measurement 16, A29 IOP Publishing, 1995.

  31. L Fabrizi, a McEwan, T Oh, E J Woo, D S Holder. An electrode addressing protocol for imaging brain function with electrical impedance tomography using a 16-channel semi-parallel system.. Physiological measurement 30, S85–101 (2009). Link

  32. Tong In Oh, Eung Je Woo, David Holder. Multi-frequency EIT system with radially symmetric architecture: KHU Mark1. Physiological measurement 28, S183 IOP Publishing, 2007.

  33. Andy Adler, Marcelo B Amato, John H Arnold, Richard Bayford, Marc Bodenstein, Stephan H Böhm, Brian H Brown, Inéz Frerichs, Ola Stenqvist, Norbert Weiler, others. Whither lung EIT: Where are we, where do we want to go and what do we need to get there?. Physiological measurement 33, 679 IOP Publishing, 2012.

  34. H Jasper. Report of the committee on methods of clinical examination in electroencephalography. Electroencephalogr Clin Neurophysiol 10, 370–375 (1958).

  35. Robert Oostenveld, Peter Praamstra. The five percent electrode system for high-resolution EEG and ERP measurements. Clinical neurophysiology 112, 713–719 Elsevier, 2001.

  36. Kirill Y Aristovich, Gustavo Sato dos Santos, Brett C Packham, David S Holder. A method for reconstructing tomographic images of evoked neural activity with electrical impedance tomography using intracranial planar arrays. Physiological measurement 35, 1095 IOP Publishing, 2014.

  37. Markus Jehl, Andreas Dedner, Timo Betcke, Kirill Aristovich, Robert Klofkorn, David Holder. A Fast Parallel Solver for the Forward Problem in Electrical Impedance Tomography.. IEEE transactions on bio-medical engineering 9294, 1–13 (2014). Link

  38. Shiwei Xu, Meng Dai, Canhua Xu, Chaoshuang Chen, Mengxing Tang, Xuetao Shi, Xiuzhen Dong. Performance evaluation of five types of Ag/AgCl bio-electrodes for cerebral electrical impedance tomography. Annals of Biomedical Engineering 39, 2059–2067 (2011). Link

  39. B Packham, G Barnes, G Sato, Dos Santos, K Aristovich, O Gilad, A Ghosh, T Oh, D Holder. Empirical validation of statistical parametric mapping for group imaging of fast neural activity using electrical impedance tomography. Physiological Measurement 37, 951 (2016). Link

  40. Michael Armstrong-James, Christopher A Callahan, Michael A Friedman. Thalamo-cortical processing of vibrissal information in the rat. I. Intracortical origins of surround but not centre-receptive fields of layer IV neurones in the rat S1 barrel field cortex. Journal of comparative neurology 303, 193–210 Wiley Online Library, 1991.

  41. Carl CH Petersen. The functional organization of the barrel cortex. Neuron 56, 339–355 Elsevier, 2007.

  42. RR Peeters, I Tindemans, E De Schutter, A Van der Linden. Comparing BOLD fMRI signal changes in the awake and anesthetized rat during electrical forepaw stimulation. Magnetic resonance imaging 19, 821–826 Elsevier, 2001.

  43. Kazuto Masamoto, Tae Kim, Mitsuhiro Fukuda, Ping Wang, Seong-Gi Kim. Relationship between neural, vascular, and BOLD signals in isoflurane-anesthetized rat somatosensory cortex. Cerebral cortex 17, 942–950 Oxford Univ Press, 2007.

  44. Andrew S Lowe, John S Beech, Steve CR Williams. Small animal, whole brain fMRI: innocuous and nociceptive forepaw stimulation. Neuroimage 35, 719–728 Elsevier, 2007.

  45. Emma Malone, Gustavo Sato dos Santos, David Holder, Simon Arridge. A reconstruction-classification method for multifrequency electrical impedance tomography. IEEE transactions on medical imaging 34, 1486–1497 IEEE, 2015.

  46. J Jang, JK Seo. Detection of admittivity anomaly on high-contrast heterogeneous backgrounds using frequency difference EIT. Physiological measurement 36, 1179 IOP Publishing, 2015.

  47. James B. Ranck. Analysis of specific impedance of rabbit cerebral cortex. Experimental Neurology 7, 153–174 Elsevier BV, 1963. Link

  48. Nikos K. Logothetis, Christoph Kayser, Axel Oeltermann. In Vivo Measurement of Cortical Impedance Spectrum in Monkeys: Implications for Signal Propagation. Neuron 55, 809–823 Elsevier BV, 2007. Link

[Someone else is editing this]

You are editing this file