Parallel, multi frequency EIT measurement, suitable for recording impedance changes during epilepsy

A typical EIT system consists of an electrode array, a current source, and a voltage measurement unit. The electrode array is connected to the object to be imaged and a sinusoidal current, of a defined amplitude and frequency, is injected through a pair of electrodes, while voltages are measured at each of the remaining electrodes. This process is repeated for a number of different pairs of injecting electrodes, yielding an array of voltage measurements which can be used to reconstruct the conductivity profile of the target object. EIT systems may inject through a single electrode pair, or multiple electrode pairs simultaneously, at frequencies from 10 Hz – 1 MHz and typically have 16–32 electrodes. Voltage acquisition at each electrode is performed either serially or in parallel, and frame acquisition times are in the range 10 ms – 5 minutes. Existing systems include fEITER [20] (32 electrodes, parallel record, single pair injection at 10 kHz, 100 frames/s), KHU [21] (16 – 64 electrodes, parallel voltage record, multiple pair injection, 11 Hz – 500 kHz, 100 frames/s) and Swisstom Pioneer Set (32 electrodes, single pair injection at 50 kHz, 50 frames/s). Previous attempts to image 3D epileptic activity employed the UCLH Mark 1b system [22] a 31 electrode system, with serial data record and an injection current of 38.4 kHz. However, it was not possible to capture reproducible impedance changes during seizures using this approach, due to the limitations of the hardware.

In epilepsy, there are two types of abnormal activity. Seizures usually comprise continuous epileptic spiking which lasts for up to a minute or so. Each spike is caused by a depolarization lasting up to 70 ms due to the opening of ion channels and causes a surface negative potential. Within a few seconds, there are movements of water from the extracellular to the intracellular space, as brain cells run out of energy to control their osmotic pressure. These are termed ‘ictal’ events. Between seizures, there may be frequent individual spike discharges, perhaps every minute or so, which are termed ‘interictal’ spikes and are not usually noticed by the subject. The impedance change associated with the seizure has two components. For each individual spike, the cell depolarisation causes an impedance decrease of approximately 1% at DC. At measurement frequencies below ~10 kHz, the applied current will predominately flow through the extracellular space, but as ion channels open at the seizure onset, some current will flow through the intracellular space, reducing the magnitude of the impedance change to about 0.1% [23]. Following this initial decrease in impedance, there is an increase of impedance due to cell swelling of up to 10%, lasting several seconds [24].

In order to correlate impedance changes to epileptic activity, it is necessary to record EIT data simultaneously with EEG. Systems, such as the UCLH Mark 1b, which employ serial data record, are not ideally suited to capturing these types of events, as it is necessary to switch between the measurement electrodes. This reduces the temporal resolution of the measured impedance signal, while also introducing a switching artefact into the EEG recording. Switching of the injecting electrodes introduces an artefact which further compounds these problems. The contamination of the EEG signal occurs in all of the EIT systems previously mentioned. While it is possible to remove some of the switching artefact [25], this requires the use of separate electrodes for recording EIT and EEG data, and does not entirely eliminate the problem.

Switching of the recoding electrodes can be eliminated by employing parallel voltage recordings, and several systems have been reported with this capability, including the SwissTom and KHU systems. Overcoming the need to switch the injecting electrodes is more difficult, requiring some form of multiplexing of the input signal. Various implementations have been reported, including frequency division multiplexing (FDM) [26,27] where current is injected simultaneously through multiple electrode pairs, at different frequencies; and code division multiplexing (CDM) [28], where each injected current is modulated by a unique binary code, allowing for multiple simultaneous injections at the same frequency.

Two approaches to imaging seizures can be undertaken. The first is to measure the slow impedance change over several seconds for a single seizure. The second, more difficult approach is to measure the fast impedance change of each individual spike. This approach is more demanding, as a finer time resolution, of several milliseconds, is required. Additionally, the impedance change, of less than 1%, is significantly smaller than for the slow component, making it more difficult to detect. Other applications of neural EIT, such as the imaging of evoked potentials (EPs), can take advantage of averaging multiple identical events together to increase the signal strength. The use of averaging is more difficult in the case of epilepsy, as spiking events occur unpredictably and can vary in strength and number, reducing the uniformity required for averaging. Additionally, some patients display more than one seizure type and it would be advantageous to image each separately.

The choice of operating frequency, bandwidth and number of electrodes associated with the EIT system are also important factors. For neural events, the optimal measurement frequency has been reported to be ~2 kHz [11]. If only the slow component of impedance change is considered, a bandwidth of several Hz would be sufficient. For the fast changes, occurring over several milliseconds, the signal bandwidth needs to be >200 Hz to capture all necessary information. No conclusive data is available on the optimal number of electrodes required, but for existing studies involving 3D neural imaging, 60 is the minimum number of electrodes used [14].

In principle, several of the EIT system described above are able to record the slow impedance changes, but they lack the capability to effectively record the EEG signal accurately. With regard to recording the fast changes, none of the systems have a small enough time resolution, without resorting to averaging, due to the need to switch between injection/recording electrodes. For imaging with EIT in epilepsy, modifications to EIT hardware and data collection are required to achieve the joint goals of improved temporal resolution and avoidance of EEG artefact. We propose that this can be achieved through a number of modifications: frequency division multiplexed injections, in the kHz range, parallel voltage recording, support for at least 60 electrodes and simultaneous recording of EEG data.

The purpose of this work is to describe the construction and testing of such a system. Experiments in a saline filled tank verify the operation of the system for imaging short duration events of several milliseconds, and preliminary data from rat experiments verifies that the system meets the requirements in a practical experimental setting.

To evaluate the parallel, multi frequency approach to EIT imaging, time difference experiments were produced using a saline filled tank (0.9 % saline). Four independent current sources were assembled. The saline tank had 32 electrodes uniformly spaced in a circle around the center. The current sources were connected to inject between opposite electrodes (1-17, 5-21, 9-25, and 13-29). The contact impedance of each electrode was less than 2 kΩ at 1 kHz. All 32 electrodes were connected to the BioSemi EEG recording system.

Three illustrative frequency groupings were chosen, with spacing between frequencies of 100 Hz, 10 Hz and 1 Hz (Table 1).

For each frequency configuration, a baseline recording of the tank was taken. Following this, a Perspex rod, 2 cm in diameter, was introduced into the saline tank as a perturbation, at three separate locations – the centre of the tank, halfway between the centre and electrode 20 and next to electrode 31.

Data processing and image reconstruction were carried out in MATLAB. For each of the four frequencies used, the data was bandpass filtered and demodulated, producing a set of voltages at each of the 32 electrodes. Two filter designs were considered. The first used a narrow bandwidth (1/10^th of the frequency spacing). The second used a bandwidth equal to the frequency spacing (200 Hz) with notch filters used at each of the unwanted injected frequencies, which gives the maximum possible temporal resolution. The difference between the two sets of voltages was far less than the average level of noise seen in the system and can be said to be insignificant. As such, either approach to filtering can be implemented.

Difference data was produced by subtracting the baseline voltages from the voltages obtained in the presence of the Perspex rod. Image reconstruction was performed using a 0^th order Tikhonov regularisation [30], on a 60,000 finite element mesh of the cylindrical tank.

Frequency groups A&B (100 Hz and 10 Hz spacing) resulted in successful reconstruction of the location of the Perspex rod, accurate to within 4 mm. Group C, with 1 Hz spacing between frequencies, did not result in acceptable images (Figure 4). In practice, such narrow spacing will not be required. As previously discussed, the required bandwidth will be dictated by the time resolution needed to image the fast impedance change, with a minimum of 200 Hz required.

Fig. 4

These results verify that the hardware operates correctly, but represent a simple test condition, as only static objects are considered. The results are comparable to those produced by existing EIT systems. In order to test the operation of the system under more realistic conditions, comparable to those associated with seizure localisation, an additional experiment was performed. In this second experiment, coins were dropped into the saline filled tank. A coin was chosen for convenience, but for more statistically robust results, other objects, such as a ball bearing could also have been chosen. Each coin disturbed the saline as it fell to the bottom of the tank, leaving behind an air pocket which persisted for somewhere in the region of 10-100 ms. Imaging the effects of the coins poses a greater challenge, as the durations of the events are relatively short and the exact nature of the air pocket produced is highly variable. In particular, it is dependent on the speed and angle at which the coin breaks the surface of the saline.

The same experimental setup and imaging method was used, with four current sources injecting at 1500 1800, 2100, and 2400 Hz. A total of five coins were dropped into the tank at different locations, one every 30 seconds. By splitting the recorded data into 1ms frames, and imaging each frame with reference to the first frame, it was possible to reconstruct the locations at which the coins were dropped and the subsequent disturbance of the saline.

There were noticeable variations in magnitude and duration of the disturbance of the saline, corresponding to the varying trajectories of the coins. There is a clear relationship between the size of the perturbation created and its duration. The effect of the first coin is both the shortest and smallest (10 ms), while the fourth coin causes the largest and longest lasting disturbance (>100 ms) (Figure 5). This is as expected; larger air pockets will naturally take longer to form and disperse.

Fig 5

Having verified the functionality of the hardware through tank experiments, preliminary work was undertaken to test the system in an animal model, to measure the slow impedance change during epilepsy in rats [11, 31] (Figure 6A). A female Sprague-Dawley rat was anaesthetised with isoflourane, and maintained with α-chloralose and fentanyl.

Fig. 6

The skin on top of the skull was incised in the mid-line and two craniotomies were performed. Electrode arrays [32] were placed on either side of the rat brain. Further details of the anaesthesia and surgical procedure are available in [31]. Thirty electrodes were used for recording on the left hemisphere and sixty on the right. Epilepsy was triggered in the hippocampus, by the injection of 4-AP into the right hemisphere. Two parallel current sources were used, operating at 2.2 kHz and 2.6 kHz. The peak injecting current was 50 μA and the currents were sourced/sinked through the left/right electrode arrays respectively, with diametrically opposite injection pairs chosen. The BioSemi system was used for simultaneous voltage recording.

The raw recorded data, containing both EEG and EIT data, was split into the two separate components (Figure 6B). Examples of extracted EEG and impedance data (dZ) are shown in Figure 7, over 15 seconds. Seizure onset occurred at 9 seconds. For the seizures shown, the range of impedance changes were from -0.75 % to 1.6 % at 2.2 kHz and -0.7 % to 0.6 % at 2.6 kHz. Data was recorded over 4 hours and a total of 209 seizures were observed. The maximum impedance changes seen at 2.2 kHz and 2.6 kHz were 6.28 % and 3.91 % respectively. Impedance data gathered from all electrodes during a single seizure is shown in Figure 8. The magnitudes of impedance changes measured were in agreement with existing literature on the subject [24].

Fig. 7

Fig. 8