# Abstract

Electrical Impedance Tomography (EIT) could be used as a rapid and non-invasive technique to image/diagnose ischaemic or haemorrhagic stroke. However, there are currently no suitable imaging/classification methods which which can be applied to human stroke data. In part this is due to the complexity of the problem itself, but it is also affected by a lack of available data on which to evaluate different techniques. Multi-frequency EIT data (alongside MRI/CT) has been collected on 23 stroke patients, and 10 healthy volunteers, as part of a clinical trial in collaboration with the Hyper Acute Stroke Unit (HASU) at University College London Hospital (UCLH). Data was collected at 17 frequencies between 5Hz and 2kHz, with 31 current injections, yielding 930 measurements at each frequency. The raw data, collected simultaneous on all channel using an EEG amplifier sampling at 16kHz, is also made available. Bit more maybe

# TO DO

• The BioSemi file recording starts before and ends after the injection starts - can we see any EEG at these points?
• Add data processing example with figure
• Add example plot of final data

# Background & Summary (Max 700 words)

## Introduction to EIT

Electrical Impedance Tomography (EIT) is a safe, portable and inexpensive imaging method, which can produce images of the internal conductivity of an object, by injecting insensible electrical currents into an object at a frequency between several kHz and several MHz. EIT has the potential to image brain function and pathology, with current applications including localization of epileptic foci (Bagshaw 2003) (Fabrizi 2006), imaging normal haemodynamic brain function (Tidswell 2001) and fast neural activity (Aristovich 2016).

## EIT of Stroke

The ability to rapidly differentiate haemorrhagic and ischaemic stroke, to enable appropriate treatment, is one of the most important challenges in stroke management and care (Donnan 2009). Thrombolytic “clot-busting” drugs are a treatment for acute ischaemic stroke which must be administered within four and a half hours after the onset of symptoms (Hacke 2008). To enable effective treatment within this restricted time window, new methods are required to distinguish haemorrhagic from ischaemic stroke, without waiting for hospital admission and a CT or MRI scan. It is estimated that in most countries, only 4% to 10% of potentially eligible patients receive the treatment (Bambauer 2006), primarily because of delayed admission to stroke centres.

The impedance contrast between 'healthy' brain tissue, blood and ischaemic brain tissue ($$\Omega_{blood}$$ < $$\Omega_{brain}$$ < $$\Omega_{ischaemia}$$) (Dowrick 2015) means that the situation is well suited for exploration using EIT.

EIT has the potential to provide an inexpensive portable unit for use in ambulances or GP surgeries which would revolutionise thrombolytic management of stroke by providing imaging at the point of contact.

## EIT Imaging

A number of reconstruction algorithms have been proposed for the imaging of stroke using impedance data, including time, frequency (Malone 2014) and absolute imaging methods (Ma 2014). Of these, frequency difference is the most promising for a clinical setting, as it allows image construction to take place using data collected at a single point in time, whereas time difference algorithms require a 'baseline' data set, recorded before the stroke onset. Frequency difference imaging presents a substantially greater technical challenge, in terms of the algorithmic and computational complexity, and effective algorithms have not yet been developed for use with human data.

## EIT Hardware

An EIT system is typically comprised of an electrode array, current source, voltage measurement unit and switching circuitry to direct an injection current to a particular subset of electrodes. The EIT system used in this study was developed at UCL (Avery 2017) and uses a Keithley 6221 Current Source (Keithley Instruments, Cleveland, Ohio, USA), BioSemi EEG Recorder (Biosemi, Amsterdam, The Netherlands), EasyCap EEG electrodes (EasyCap, Germany) and a custom PCB for current routing and system control.

A sinosoudial current, with a given amplitude and frequency (typically in the 100 µA and kHz ranges), is injected between a pair of electrodes, referred to as an injection, and voltages are measured at all electrodes in parallel \ref{EITexample}. A single EIT measurement is the demodulate voltage amplitude at a single electrode, averaged over a certain number of periods of the waveform. The electrodes used for injection are excluded from the measurement set. A complete set of measurements, referred to as a protocol, is collected by repeating the current injection at a number of different injection pairs. The entire set of n voltage measurements, equal to the number of injection pairs multiplied by the number of measurement electrodes, is referred to as a frame of data. TODO: More on digital triggers? What is recorded etc Digital triggers, used to indicate the beginning/end of each frequency range, frame and individual injection, are also recorded by the BioSemi.

## Frequency Difference EIT Dataset

A multi-frequency EIT dataset has been collected in 23 stroke patients and 10 healthy volunteers, in collaboration with the Hyper Acute Stroke Unit (HASU) at University College London Hospital (UCLH). For each individual, three frames were collected, each at 17 frequencies between 5Hz and 2kHz with 930 individual measurements collected at each frequency. Additionally, a CT and/or MRI was performed on each of the 23 stroke patients and a clinical assesement carried out by medical personel.

## Time Difference EIT Dataset

In each of the healthy volunteers, and four of the stroke patients, an additional data set was collected, where the same measurement process was repeated after 24 hours, to obtain data suitable for time difference EIT analysis.

\label{EITexample}Example of EIT data collection process. In this simplified example, three current injection pairs are used, out of nine total electrodes. For each injection pair, voltage recordings are made on the remaining seven electrodes.

# Methods

## Ethical approval, patient selection and consent

All experiments on human subjects were approved by the UCL REC, the local Reserach Ethics Committee, and NHS/HSC R&D (IRAS ID: 168765). The following patient inclusion/exclusion criteria was established:

Inclusion criteria

• Adult patient (≥18year, no upper age limit).
• Clinical diagnosis of cortical ischaemic stroke or lobar intracerebral haemorrhage.
• The stroke has a greatest axial diameter > 1.5 cm on CT/MRI OR the patient has NIHSS > 5.
• Able to undergo the EIT assessment within 7 days of onset.

Exclusion criteria

• Expected to require critical transfer or intervention, ITU admission or to be transferred out of the HASU for any other reasons.
• Unlikely to tolerate the procedure (e.g. agitation on admission), or a very high likelihood of death within 48 hours of stroke onset.
• Additional medical illness (such as Epilepsy, skin problem, head or face metal implants) or technical aspect (such as no bed available on HASU) that interferes with EIT assessments.

Potential participants were first identified by the clinical care team, who also initiated first discussions with the patient. It they were willing to participate, written consent was obtained. Where the patient lacked the capacity to give consent, assent was given by a member of the family.

## Electrode arrangement and application

Recordings were performed using 32 EEG electrodes (EasyCap, Germany), using the configuration described in (Tidswell 2001), which includes 21 locations from the EEG 10-20 standard 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).

All electrodes sites were cleaned with surgical spirit, abraded using Nuprep gel (Weaver and Co., USA) and the electrodes affixed using Elefix conductive paste (Nihon Kohden, Japan).

## EIT Data Collection

EIT data was collected at seventeen frequencies (Hz), with the current amplitude at each frequency set to the maximum allowed under the guidelines set out in IEC 60601 (Table \ref{EIT-Setup}). In order to reduce the total data collection time, the number of current injection periods applied was reduced at lower frequencies. The experimental protocol was performed three times, for a total duration of 20 minutes.

## MRI and CT data

As part of the clinical diagnosis process in the HASU, each patient underwent a MRI and/or CT scan (DICOM format), which is included in this repository. All MRI/CT data was anonymised by removing any identifying data from the DICOM file, using DICOM Clener.

## Data processing

A MATLAB script InsertScriptName.m, detailing all of the steps taken to generate the final data set from the EEG data files, is included with the data repository.

The EEG data from the bdf source file is converted to a double precision array, with dimensions n_electrodes x n_samples. n_electrodes is 32 in all cases, n_samples is equal to the recording time in seconds multipled by the sampling frequency, Fs.
The digital triggers, also extracted from the bdf file, are used to partition the data according to frequency, into 17 segments, each comprising three frames of data.

Each segment of data is filtered (TODO: add filter details to table 1?) around the injection frequency, and demodulated to give the voltage amplitude \ref{DataProcessing}. Further segmentation is carried out by splitting the data into individual injections. The voltage amplitude is averaged over all cycles recorded for each injection. To exclude filter/demodulation artefacts, the first and last two sinwave cycles are removed before averaging.

The gain of the BioSemi has a known frequency dependence, with the -3dB point at 3kHz. In order to account for this, voltages at each frequency are divided by the apporpriate gain value to normalise all voltages. The three frames of data collected at each frequency are averaged together, to produce the final data set.

Data cleaning was carried out by TODO: what were the criteria?. Both the cleaned and uncleaned data is included in the published dataset, allowing alternative cleaning criteria to be used.

\label{DataProcessing}Example of data processing pipeline for a single frequency, using four injections. (a) Raw data is imported from BioSemi bdf file. (b) Filtering around the injection frequency and demodulation. (c) Segmentation into individual injections and averaging. The first/last cycles of the waveform are excluded from averaging.

 Frequency (Hz) Amplitude (uA) Injection Periods Collection Time per Injection (ms) 5 45 32 6400 10 45 32 3200 20 45 32 1600 100 45 32 320 200 90 64 320 300 90 64 213 400 90 64 160 500 90 64 128 600 90 64 107 700 140 64 91 800 140 64 76 900 140 64 71 1000 140 64 64 1200 160 128 107 1350 190 128 95 1700 235 128 75 2000 280 128 64

# Data Records

Data is availabile at xxxx in format yyyy

## Raw Data Files

The raw experimental data, in .bdf format, for all patients is available. Data was sampled at 16kHz on all channel simultaneous, over a dynamic range of +/-1V with 24 bit resolution.

## EIT Data

Data is organised as a 1x27 structure with the following fields:

NameTag: Experiment/Patient identifier.

Classification: Diagnosis (Healthy, Ischaemia, haemorrhage).

Location: Indicates which hemisphere the pathology is located in.

VoltagesFull: Raw voltages (mV) in 930x17 array.

VoltagesCleaned: Cleaned voltages (mV) in 930x17 array (NaN values replace removed elements).

RemovedChannels: Information on which measurements were removed during data cleaning.

Diagnosis: Detailed diagnosis information.

StudyID: Internal UCL reference, for correlating with other clinical data.

Electrode positions: Nominal X, Y and Z coordinates of each measurement electrode, relative to SOME POSITION......

## MRI/CT

MRI and CT scans are provided in DICOM format, identified as patient/studyID.

\label{my-label}EIT Data Summary. Two data sets were collected from patients 4, 6, 12, 23 and 25. IS = Ischaemia, ICH = Haemorrhage.
Tag Stroke to EIT Interval Reference Electrode Group
 Scans Available (Time from Stroke)
P1r2 24 hours Left
 SMALL IS
 CT (>12 hours) MRI (24 hours)
P3r2 27 hours Right BIG ICH
 CT (>12 hours) MRI (24 hours)
P4Ar2 18 hours Left BIG IS
 CT (2 hours) CT (26 hours)
P4Br2 42 hours Left BIG IS
P5r1 32 hours Left SMALL ICH
 CT (2 hours) MRI (>24 hours)
P6Ar2 36-48 hours Left BIG ICH
 CT (24 hours) MRI (24 hours) CT (8 days)
P6Br2 25 days Middle BIG ICH
P9r3 2-3 days Middle BIG IS
 CT (>2 days) MRI (>3 days)
P11r2 15 hours Middle BIG ICH CT (>12 hours)
P12Ar2 8 hours Middle SMALL IS
 CT (>3 hours) CT (24 hours)
P12Br2 24 hours Middle SMALL IS
P15r2 3 days Middle BIG IS
 CT (>12 hours) MRI (2 days)
P16r2 48 hours Middle SMALL IS
 CT (5 hours) MRI (24 hours)
P17r2 21 hours Middle SMALL ICH
 CT (>2 hours) MRI (4 days)
P18r2 3 days Middle SMALL IS
 CT (5 hours) MRI (48 hours)
P19Ar2 15 hours Middle BIG IS
 CT (>3 hours) CT (20 hours)
P19Br2 41 hours Middle BIG IS
P20r2 3 days Middle SMALL ICH
 MRI (>2 days) CT (8 months before stroke)
P23Ar1 >2 days Middle BIG ICH
 CT (>2 days) MRI (>3 days)
P23Br2 >6 day Middle BIG ICH
P24r2 27 hours Middle BIG ICH
 CT (2 hours) MRI (24 hours)
P25Ar2 12 hours Middle BIG IS
 CT (1 hour) CT (25 hours)
P25Br2 36 hours Middle BIG IS
P26r2 3-5 days Middle BIG IS CT (1-3 days)
S1Br2 N/A Middle HEALTHY N/A
S2Ar2 N/A Middle HEALTHY N/A
S3Br2 N/A Middle HEALTHY N/A
S4Ar2 N/A Middle HEALTHY N/A
S5Ar2 N/A Middle HEALTHY N/A
S6Ar2 N/A Middle HEALTHY N/A
S7Ar2 N/A Middle HEALTHY N/A
S8Ar2 N/A Middle HEALTHY N/A
S9Ar1 N/A Middle HEALTHY N/A
S10Ar1 N/A Middle HEALTHY N/A
Replace this text with your caption

# Technical Validaiton

Resistor phantom measurements - correct frequency spectrum etc.

# 3 Figures Allowed

• Experimental overview
• Data processing illustration
• Sample data

### References

1. Kara Z. Bambauer, S. Claiborne Johnston, Derek E. Bambauer, Justin A. Zivin. Reasons Why Few Patients With Acute Stroke Receive Tissue Plasminogen Activator. Archives of Neurology 63, 661 American Medical Association (AMA), 2006. Link

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

3. James Avery, Thomas Dowrick, Mayo Faulkner, Nir Goren, David Holder. A Versatile and Reproducible Multi-Frequency Electrical Impedance Tomography System. Sensors 17, 280 MDPI AG, 2017. Link

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