3.4 Error del m´ etodo
5.1.1 Estimaciones de ´ ordenes de magnitud
There are four main applications of brain imaging with EIT: stroke[7],[48],[80], epilepsy [28],[30], evoked responses[81], and fast neural activity[46]. Each of these techniques aim to exploit impedance changes in the brain caused by different underlying physiological effects. EIT of brain function has been the focus of the UCL group for several decades, with particular emphasis on stroke and fast neural applications in recent years.
1.3.1
Bioimpedance of stroke
Although there have been TD recordings made in stroke models, the difficulty in collecting accuratein-vivo measurements has precluded any single exhaustive study examining the impedance changes during acute stroke. However, a review of the literature by Horesh[82] enables predictions of the changes that will occur. A comparison of the conductivity spectra of the relevant tissues is given in fig. 1.11 from the results of this review[82]. The conductivity of normal brain is approximately 0.1 S/m at frequencies below 100 Hz, where current passes predominantly through the extracellular space. This increases to 0.18 S/m at 1 MHz as current flows through the cell membranes into the intracellular space. In contrast to healthy brain, the ischaemic tissue has a conductivity 0.07 S/m less at 10 Hz, and this difference reduces to around 0.02 S/m by 100 Hz with this difference maintained up to 1 MHz. Blood has a substantially higher conductivity of 0.7 S/m below 100 kHz, and this increases up to 0.82 S/m at 1 MHz.
The change in conductivity spectra between normal and ischaemic brain tissue is caused by a reduction in extracellular space due to cell swelling as a result of hypoxia caused by the
100 101 102 103 104 105 106 107 108 10−2 10−1 100 Frequency (Hz) C o n d u ct iv it y S m − 1 Normal Blood Ischemic
Figure 1.11: Change in conductivity over frequency of normal brain tissue, ischaemic brain tissue and
blood, adapted from[82]
ischaemic event[83]. At low frequencies, the current cannot pass through the cell membrane, and must pass through the extracellular space. So this reduction caused by ischaemia results in a conductivity decrease of approximately 10-20 % at low frequencies (< 100 Hz). In contrast, a haemorrhage produces an area of high conductivity which is approximately flat across frequencies below 1 MHz[39]. It is these spectral differences which EIT of stroke attempts to exploit. Whilst there are differences in all three spectra up to approximately 1 MHz, the focus of stroke EIT is on frequencies below 2 kHz, where the difference is greatest. Therefore the possibility of distinguishing an ischaemic stroke is maximised.
1.3.2
Bioimpedance of intracranial bleeding
Tissue Conductivity (S/m) Skull 0.018 Scalp 0.44 CSF 1.79 Dura matter 0.44 Grey matter 0.3 White matter 0.15 Air 0.001 Blood 0.7
The conductivity changes arising from intracranial bleeding are similar to those expected from haemorrhagic stroke as they both involve blood entering the skull. However, depending upon which type of injury, the blood may replace different tissues within the skull, and thus present a different contrast. For example, in the case of a subdural haematoma (SDH), blood would replace both the CSF between the brain and the dura, and also compress the brain. Whereas with a interventricular haemorrhage (IVH), the blood enters the ventricles which are primarily filled with CSF. As the summary by Horesh[82]table 1.1 demonstrates, blood has a substantially different conductivity to most head tissues, representing an increase of 60 to 130 % increase compared to brain tissues, or a 150 % decrease compared to CSF. It is difficult to estimate the contrast between tissue from in vivo measurements, as the point spread function of the EIT reconstructions themselves, distrubuted the real conductivity changes over a wider region. Manwaring, Moodie, Hartov,et al.[14]found a mean conductivity difference of 19.5±11.5 mS/m during injection of 1 ml of blood in a piglet brain, an order of magnitude
less than would be expected from the values in table 1.1. It is possible the injected blood diffused over a larger area than intended, or cell ischaemia secondary to the introduction of blood into the white matter may have caused an impedance change in the opposite direction, cancelling out some of the change as a result of the blood. Therefore, care must be taken when conducting experiments into EIT of intracranial bleeding, not to overestimate the contrast blood represents in the brain.
1.3.3
Bioimpedance of epilepsy
It is well-established that the impedance of brain tissue increases abruptly during seizure activity. Animal experiments in the 1960s demonstrated that the impedance of cerebral tissue increases by 1-12 % during induced and spontaneous epileptic activity[84],[85]. Rao[86] measured local cortical impedance changes arising from electrically induced seizures of 9.5 % at 47 kHz. The increase in impedance has been attributed to cell swelling[87],[88], as intense neuronal activity creates an osmotic gradient which leads to movement of water inside the cells, causing the extracellular space to shrink. As with ischaemic stroke, at low frequencies (<50 kHz), the injected current flows mostly around the extracellular fluid and consequently a reduction in the extracellular space causes an increase in the measured tissue resistivity and impedance[89]. Cell swelling develops soon after the onset of ictal and inter-ictal events and, according to some sources, may even precede electrographic changes[87],[90].
Besides its secondary effect through cell swelling, seizures also cause a direct, transient change in impedance. Highly synchronous neural activity causes a fast drop in impedance lasting a few ms due to the opening of ion channels, which allow current to flow through the intracellular space. These fast impedance shifts have been described in anaesthetised animals during evoked responses[46],[91]. Recent developments in the UCL group have demonstrated this fast neural signal is present during seizures[92], with a transient change
of approximately 0.1 % lasting 2 ms.
1.3.4
Data collection and electrode localisation
The current procedure for collecting data in stroke patients in clinical studies within the UCL group is based on that described in[7]. 32 EEG electrodes are placed on the scalp of the patient by hand after preparation of the electrode sites with abrasive paste to reduce the contact impedance. The hardware used in these trials - UCL Mk2.5 - measures the contact impedance at 10 kHz and a threshold of 2.5 kΩper electrode is set. The electrodes are held in place during recording by the paste and by elasticated material. The electrodes are localised by a procedure called photogrammetry. Targets are placed at the positions of the electrodes and multiple photographs are taken of the patient’s head. The three dimensional locations of these electrodes can be reconstructed from the two dimensional images based on their scale, rotation and relative positions. The accuracy of this localisation has been found to be approximately 0.5 - 2 mm. An example of the data collection from an unpublished clinical study is shown in figure 1.12. The shape of the finite element model can give rise to significant errors in EIT images if it is not representative of the geometry of the volume under examination [73]. To minimise these errors in stroke EIT, FEMs specific to the patient generated from CT or MRI scans, are used in the image reconstruction[93]. The electrode coordinates obtained via photogrammetry are then rotated and scaled manually in the reconstruction software to bring them into alignment with the mesh geometry. In the stroke application of EIT in the ambulance, it will not be possible to have these accurate meshes before the patient arrives in the stroke unit as they would not have had a CT or MRI. Therefore, it is important that the electrode localisation is as accurate as possible to infer the geometry of the patients scalp from these measurements. Using these measurements it would be possible to warp a generic head mesh to match the measured geometry, thereby better representing the actual measurement conditions and improving reconstructed images. The measurement accuracy required is currently unknown, thus the aim of this project is to at least match the existing technique of photogrammetry.
Figure 1.12: Example of data collection for stroke EIT, photogrammetry markers are visible above elastic cap
1.3.5
Summary of previous studies
Prompted by the encouraging results from Bagshaw, Liston, Bayford,et al.[64]which showed a measurable change on the scalp during seizures, an attempt to image epilepsy in human subjects was undertaken by Fabrizi, Sparkes, Horesh,et al.[30]. The UCLH Mark 1b system was used which utilized a single impedance four-terminal measuring circuit multiplexed to up to 31 electrodes[94]on the scalp of patients undergoing pre-surgical EEG-telemetry. Unfortunately no reproducible impedance changes were found during seizures, as the comparatively small boundary voltage changes were masked by large baseline drifts and movement artefacts from the electrodes. Another limitation highlighted by the authors was the lack of accurate patient FEM and electrode localisation.
A pilot clinical MFEIT pilot study was performed by Romsauerova et al. [7]which consid- ered 7 patients with pathologies which mimic the impedance characteristics of a haemorrhage. Data were collected at 16 and 64 kHz and images reconstructed in a 30,000 element FEM generated from patient MRI data. No reproducible changes between patients were found in the raw data or the images. Since the publication of this study, work has been undertaken within the UCL group to improve the instrumentation, the measurement protocol[76], the FEM[93], the reconstruction algorithm and data rejection[68]. Although the results are unpublished, these improvements have not resulted in reproducible changes across patients in subsequent studies. Two of the issues that have yet to be tackled successfully are the drift in standing potentials over time caused by changes in contact impedance over time (section 1.4), and correct electrode localisation with respect to patient geometry.