A Review of Errors in Multi-frequency EIT Instrumentation A McEwan, G Cusick and D S Holder Department of Medical Physics and Bioengineering, UCL, London, UK E-mail:
[email protected] Abstract MFEIT (Multi-Frequency Electrical Impedance Tomography) was proposed over 10 years ago as a potential spectroscopic impedance imaging method. At least seven systems have been developed for imaging the lung, heart, breast and brain, yet none has yet achieved clinical acceptance. While the absolute impedance varies considerably between different tissues, the changes in the spectrum due to physiological changes are expected to be quite small, especially when measured through a volume. This places substantial requirements on the MFEIT instrumentation to maintain a flat system frequency response over a broad frequency range (dcMHz). In this work the EIT measurement problem is described from a multi-frequency perspective. Solutions to the common problems are considered from recent MFEIT systems and the debate over four-terminal or two-terminal (multiple source) architecture is revisited. An analysis of the sources of MFEIT errors identifies the major sources of error as stray capacitance and common mode voltages which lead to a load dependence in the frequency response of MFEIT systems. A system that employs active electrodes appears to be the most able to cope with these errors (Li et al. 1996). A distributed system with digitisation at the electrode is suggested as a next step in MFEIT system development.
Keywords— Multi Frequency Electrical Impedance Tomography Errors Review Instrumentation 1 Introduction Our group at UCL has been interested in using EIT of the human head to investigate conditions such as epilepsy and stroke. Multi-Frequency EIT has the potential to be used for imaging in acute stroke, as it could be employed in casualty wards for urgent neuroimaging where X-ray CT scanning is not practical. As timedifference imaging is not possible, either absolute or frequency-difference mode must be employed. We have elected for frequency-difference, as it still offers the possibility of reduction of some instrumentation error across frequencies. But bio-impedance changes slowly with frequency, so most applications require comparison across a frequency range of 6-7 orders of magnitude. Over these ranges large impedance changes are expected, for example ischaemic brain may change by 100-130%. A confounding factor is that the background impedance will also change significantly with frequency, leading to small differential changes of only 15-75%; these are local changes, and will be attenuated by partial-volume effects and the impedance presented by the CSF, skull and scalp (Horesh et al. 2006). Several MFEIT systems have been developed and described in the literature but none has yet been shown to be clinically successful. A recent modelling study of acute stroke suggested an accuracy requirement of better than 0.1% over a broad range of frequencies (10Hz-1MHz) and loads (1-100Ω) (Horesh et al. 2005). While this level of performance has been achieved in the lab with test phantoms (McEwan et al. 2006), many research groups have found it challenging in vivo (Romsauerova et al. 2005). There have been several thorough reviews of EIT instrumentation in general (Boone and Holder 1996a), (Saulnier 2005),(Bayford 2006). In contrast, the purpose of this review is to focus more specifically on the more demanding case of multi-frequency, as opposed to time difference, EIT measurements, to review the error and noise sources encountered and discuss implemented and proposed methods of mitigating them. Some points are illustrated by measurements from our systems. For example, in a recent review of ‘Bioimpedance Tomography’ (Bayford 2006), the evolution and development of EIT hardware and algorithms are summarised. Four of the most recent EIT systems are
described and their diversity is attributed to differences in reconstruction algorithms. Common-mode effects and stray capacitance are suggested as the principal sources of instrumentation error. A chapter in the latest book on EIT describes instrumentation from a requirements perspective (Saulnier 2005). A current source with high output impedance, and high precision sources and sensors are suggested. Also included is a detailed summary of some of the latest EIT systems. A previous review on EIT instrumentation (Boone and Holder 1996a) considered the differential measurement problem and the principal sources of error reported by previous authors. These were considered to arise mostly from common-mode signals and stray capacitance, with imbalance between electrodes creating the greatest source of error. Existing EIT systems were introduced along with the circuits used to overcome sources of error. In addition to these reviews, some authors have quantitatively assessed the sources of error. A complete electrode model simulation showed that errors due to electrode position, boundary shape and contact impedance caused significant problems in static EIT image reconstruction. Adjacent current injection patterns were found to be superior to trigonometric (Kolehmainen et al. 1997). Errors in another MFEIT system (10-100kHz) were found to increase with frequency (Schlappa et al. 2000). Errors due to load variation were 3% while errors due to cable movement and contact impedance were a few percent.
1.1 The measurement problem EIT is performed using an array of contacting electrodes connected to a number of bio-impedance measurement channels. Errors between the channels are important as they can critically affect image quality. The equivalent circuit (Figure 1) neatly introduces the EIT measurement problem for the case of application of constant current. This is applied through the complex impedance of two electrodes to a load modelled as four resistors in a loop. Recording of the resulting voltage is again made through electrodes with complex impedance. The output impedance of the current source (Ro) and the input impedance of the instrumentation amplifier (Ri) are in parallel with stray capacitances (Co, Ci). The three major sources of error are the consequences of the common mode voltage (Vcm), that is the mutual voltage appearing at the two measurement electrodes; the stray capacitance which is unwanted capacitance to ground caused by use of cables and switches etc.; and the contact impedances (Red,Ced,Rer,Cer) between the system and the impedance of interest. Ri Ced Ro
Cer
Co Red
Is
Ci
load
Rer
+ -
Vcm
Figure 1: Equivalent circuit of the EIT measurement problem (Boone and Holder 1996a).
1.1.1 The signal source Single or multiple constant current or voltage sources are used to inject current. The most important characteristics are high output impedance for a current source (Ro,Co), and the matching between sources as any imbalanced current that flows to ground through the input impedance of the amplifier will cause an unwanted common mode voltage. An often used current source is a Howland current pump, which relies on matching between resistors to achieve high output impedance (Saulnier 2005). An ideal current source would have infinite output impedance; thus, all the current delivered must flow through the load. This is equivalent to
stating that the current is load independent. If the source impedance is reduced, the current divides between the load impedance and the output impedance, which renders the current passing through the load dependent on its impedance. Whilst building a high output-impedance current source is fairly straightforward at low frequencies, stray capacitance, which shunts the current source output, reduces the source impedance as frequency rises. Current source compliance, that is, the range of load impedances for which the source delivers constant current, is also important to maintain linearity in the measurement. As the load increases, the voltage swing at the output of the current source has to increase. The compliance is limited by the output range of the amplifiers used in the current source, which is often less than the supply voltage and may not be symmetric. Some systems use voltage sources, since these are easier to design (Halter et al. 2005; Saulnier et al. 2006). Calculation of the impedance requires knowledge of the current, normally achieved by measurement of the voltage drop over a small sense resistor, a technique which is also sensitive to the output impedance of the voltage source and is degraded by stray capacitance. 1.1.2 The electrodes EIT requires skin-electrode contact and typically uses ECG or EEG-type electrodes which are placed on abraded skin with a conducting gel to improve contact. This preparation is required to reduce the contact impedance which has two major components (Kolehmainen et al. 1997): 1) an electrochemical component, arising from the conversion of electron current to ionic current at the interface between the metal electrode and ionic solution (such as a saline tank or battery cell); and 2) the higher impedance of the layers of the stratum corneum which causes skin impedance. Abrasion is used to remove the outermost (keratin) layers which have a higher impedance (Yamamoto and Yamamoto 1976). Variation in abrasion and the skin cause variations in contact impedance of up to 20% on the same subject (see Section 3.2.3). Heating effects of the gel and changes in sweat ducts cause changes in contact impedance over time (Boone and Holder 1996b; Rosell et al. 1988a). This is a serious concern, particularly for EIT at lower frequencies (
