Table of contents

The papers on electrical impedance tomography (EIT) in this issue arise, as in last year's special issue, from a conference on Biomedical Applications of EIT held in London in April 2000. This was the second annual conference, organized as part of an Engineering and Physical Sciences Research Council 'engineering network' on biomedical EIT. The conference was attended by about 60 people, representing almost all of the groups in the world who actively work in this area.

There were talks in all the traditional areas of reconstruction algorithms, hardware and clinical applications, but the field is clearly changing as there were two additional sections, magnetic induction tomography and EIT of brain activity, each with seven papers from four groups. Another new trend was a contribution from a company, showing their novel system for impedance scanning of breast cancer. New advances from all these areas are represented in the twenty-one EIT papers in this issue.

The principal goals of the engineering network, and resulting conferences, are to enable transfer of ideas between groups and enhance collaboration. On both these counts, I think the conference was successful. There was clearly cross-fertilization of ideas between the conference delegates. In the final discussion, three interest groups were established in EIT of the breast, chest function and neonatal lung function. These groups have agreed to co-ordinate their activity, and so I hope this will further the need of our field as a whole to produce clinically convincing results. Another important outcome is the collaboration to produce EIDORS - electrical impedance and diffuse optical reconstruction software. This will be a library of reconstruction routines, written in the widely used MATLAB language. It will be freely available and will be of substantial advantage for anyone in the field who wishes to adapt or extend their reconstruction algorithms. One of the papers in this issue describes this software development.

Finally, I would like to thank Professor Nick Avis, Salford University, who kindly acted as editor for manuscripts in the special issue submitted by my own research group, and Bill Lionheart, UMIST, who organized a workshop on reconstruction algorithms early last year and is the driving force behind EIDORS.

In this paper we present the T-SCAN^TM technology and its use as a diagnostic tool for breast cancer detection. We show, using theoretical models with simplified geometries, that displaying planar two-dimensional maps of the currents detected at the breast's surface relate to the electric field distribution within the breast. This distribution is a manifestation of the bulk spatial inhomogeneities in the complex dielectric constant that represent the various tissue types. These differences may be used to discriminate between various pathological states. We furthermore illustrate a useful classifier, based on admittance data measured up to 2 kHz, and we argue that low frequency impedance measurements can be used successfully in breast cancer diagnosis.

A medical device which allows imaging of the distribution of conductivity in 3D in regions below the skin surface has been developed and tested. Its purpose is to enable early detection and preliminary diagnosis of breast tumours. Design of the measuring system and software are described. Results of clinical evaluation of the system are presented. EIT images of healthy and cancerous breasts are presented and discussed. The system is able to visualize various states of the breast and it may be possible to apply it to breast cancer detection.

In this paper we present the concept, the design and the test procedure for a DSP-based high-precision and high-performance wide-band (up to 10 MHz) bioimpedance analyser module for application in EIT or bioimpedance spectroscopy. The module implements a digital concept with appropriate signal conditioning hardware for voltage and current measurement, early signal digitization and subsequent digital signal processing in order to calculate the components of impedance (or admittance). At low frequencies, the module utilizes the conventional direct conversion method, whereas at high frequencies the undersampling technique is used. The advantages of the described system are the following: (a) the frequency range is extended to higher frequencies, (b) the number of data sampled per time interval is significantly reduced, and (c) the current consumption and the costs of the ADCs can be significantly reduced.

The validation procedure is performed by comparing the measured and theoretical values of the magnitude and the phase of the impedance for a commonly used tissue model. The module offers an accuracy of better than 0.012% for the magnitude of impedance and better than 0.02° for the phase.

Our previous system covered the frequency range of 0 to 1 MHz. In this new design we propose to cover the range from 0 to 10 MHz. The higher frequencies have forced us to reconsider several design decisions in view of both the physics of the problem and the performance of available electronic components. In this presentation we examine in detail the constraints faced by the designer, starting from wiring consideration to measurement techniques. We will also present the solutions we selected to overcome the limitations we discovered. The problems include phase detection, amplitude measurements, system organization and layout and finally system calibration.

Isaacson, Cheney and Seager have demonstrated that simultaneously applying trigonometric patterns of current to a circular electrode array optimizes the sensitivity of EIT to inner structure. We have found that it is less desirable to measure voltage at an electrode that also applies a current due to variable contact impedance. In order to preserve the optimum sensitivity while minimizing the effect of electrode artefacts, we have devised an approach where we sequentially apply a current between each individual electrode and a separate, fixed ground while measuring voltages at all other electrodes for each consecutive current impulse. By adding weighted sums of both the applied currents and corresponding measured voltages from individual passes, we can synthesize trigonometric patterns of any spatial frequency. Since only one of the electrodes in any given acquired data set is used as a source, this approach significantly dilutes the effect of contact impedance on the resulting voltage measurements. We present simulated data showing the equivalency between the synthesized and actual trigonometric excitation patterns. In addition, we report experimental data, both in vitro and in vivo, that show improved results using this data acquisition technique.

An electrical impedance tomography (EIT) simulation system for breast imaging has been developed using impedance data from a previous study and a finite-element model (FEM). This system has the functionality to construct various models of the breast, image the boundary voltages developed from any injection schema and provide parametric Cole-Cole modelling. Simulations indicate that breast carcinoma can be imaged and multi-frequency Cole-Cole dispersion data can be extracted. This is intended as the first stage towards providing an artificial intelligence based system capable of producing clinically relevant analytical data.

This paper describes the Sheffield Mk3.5 EIT/EIS system which measures both the real and imaginary part of impedance at 30 frequencies between 2 kHz and 1.6 MHz. The system uses eight electrodes with an adjacent drive/receive electrode data acquisition protocol. The system is modular, containing eight identical data acquisition boards, which contain DSPs to generate the drive frequencies and to perform the FFT used for demodulation. The current drive is in three sequentially applied packets, where each packet contains ten summed sine waves. The data acquisition system is interfaced to a host PC through an optically isolated high speed serial link (RS485) running at 2 Mbaud (2 Mbits s^-1). Measurements on a saline filled tank show that the average signal to noise performance of the system is 40 dB measured across all frequencies and that this figure is independent of frequency of measurement. These results suggest that the current system is 10 dB better in absolute terms than the previous Sheffield (Mk3a) system.

If electrical impedance tomography is to be used as a clinical tool, the image reconstruction algorithms must yield accurate images of impedance changes. One of the keys to producing an accurate reconstructed image is the inclusion of prior information regarding the physical geometry of the object. To achieve this, many researchers have created tools for solving the forward problem by means of finite element methods (FEMs). These tools are limited, allowing only a set number of meshes to be produced from the geometric information of the object. There is a clear need for geometrical accurate FEM models to improve the quality of the reconstructed images. We present a commercial tool called IDEAS, which can be used to create FEM meshes for these models. The application of this tool is demonstrated by using segmented data from the human head to model impedance changes inside the head.

A realistic three-dimensional finite element model (FEM) of the human head has been developed. Separate layers for the scalp, skull, cerebrospinal fluid (CSF) and brain were modelled. Hexahedral elements, a special master matrix assembly technique and an iterative successive over-relaxation (SOR) solution scheme were employed. This approach enabled rapid modelling with minimal memory requirements, which makes this method practical if used for electrical impedance tomography (EIT) or source localization inverse problems. Compared to scalp electrodes, subdural voltage sensing electrodes were three to four times more sensitive close to an oedema or source region, if it was peripheral, but this decreased to 30%-40% for central oedema or source regions. Scalp current injecting electrodes are preferable, since the maximum allowable current is 10 times larger than that of the subdural ones. The distance of voltage sensing electrodes from a region to be imaged highly affects sensitivity, so depth electrodes will be more sensitive, provided that they are close to the region of interest. Finally, the electrode size has significant effects on the input or transfer impedance.

Electrical impedance tomography is a technique that permits estimation of resistivity within a subject by reconstructing from boundary measurements. Due to the ill conditioning of the problem, images are greatly affected by boundary errors. Reconstruction algorithms account for boundary shape and electrode position by using a forward solver. In this paper is presented the realization of a 2D realistic thoracic model as a better approach compared to circular meshing. Several meshes with different discretizations are compared in terms of accuracy with a complete electrode model forward solver. By comparing the different discretizations it is possible to choose an appropriate mesh density for the 2D EIT problem.

There are a number of constraints which limit the current and voltages which can be applied on a multiple drive electrical imaging system. One obvious constraint is to limit the maximum ohmic power dissipated in the body. Current patterns optimizing distinguishability with respect to this constraint are singular functions of the difference of transconductance matrices with respect to the power norm (the optimal currents of Isaacson). If one constrains the total current (L¹ norm) the optimal patterns are pair drives. On the other hand if one constrains the maximum current on each drive electrode (an L^∞ norm), the optimal patterns have each drive channel set to the maximum source or sink current value. In this paper we consider appropriate safety constraints and discuss how to find the optimal current patterns with those constraints.

Adaptive mesh refinement techniques can be applied to increase the efficiency of electrical impedance tomography reconstruction algorithms by reducing computational and storage cost as well as providing problem-dependent solution structures. A self-adaptive refinement algorithm based on an a posteriori error estimate has been developed and its results are shown in comparison with uniform mesh refinement for a simple head model.

In this paper we demonstrate that conductivity changes caused by ventilation and perfusion in a human subject can be reconstructed from electrical impedance tomography data collected on a rectangular array of electrodes placed on a subject's chest. Currents are applied on the electrodes and the resulting voltages on the electrodes are measured. A 3D reconstruction algorithm is used to reconstruct the conductivity distribution in the region beneath the array. Time traces of the reconstructed conductivity distribution demonstrate the detected changes in conductivity due to ventilation and perfusion.

The EIDORS (electrical impedance and diffuse optical reconstruction software) project aims to produce a software system for reconstructing images from electrical or diffuse optical data. MATLAB is a software that is used in the EIDORS project for rapid prototyping, graphical user interface construction and image display. We have written a MATLAB package (http://venda.uku.fi/ mvauhkon/) which can be used for two-dimensional mesh generation, solving the forward problem and reconstructing and displaying the reconstructed images (resistivity or admittivity). In this paper we briefly describe the mathematical theory on which the codes are based on and also give some examples of the capabilities of the package.

A finite-difference model has been developed for simulating measurements in magnetic induction tomography (MIT) for biological tissues. The model has three stages: (1) computation of the distribution of current induced in a volume of dielectric due to the magnetic field from an excitation coil; both the electrical conductivity and permittivity of the dielectric are taken into account; (2) computation of the e.m.f. induced in the sensing coil directly from the excitation coil; (3) computation of the e.m.f. induced in a sensing coil due to the current distribution in the dielectric. From the results of stages (2) and (3), the change in signal in the sensing coil due to the dielectric can be obtained, in magnitude and phase, as a fraction of the signal received in the absence of the dielectric. The peak values in the modelled curves agreed to within 14% of practical measurements at 10 MHz on volumes of saline solution with conductivities in the range 0.7 to 6 S m^-1.

We evaluated analytically and experimentally the performance of a planar gradiometer as a sensing element in a system for magnetic induction tomography. A system using an excitation coil and a planar gradiometer was compared against a system with two coils. We constructed one excitation coil, two different sensing elements and a high-resolution phase detector. The first sensor was a PCB square spiral coil with seven turns. The second sensor was a PCB planar gradiometer with two opposite square spirals of seven turns, with a distance between centres of 8 cm. Theoretical sensitivity maps were derived from basic equations and compared with experimental data obtained at 150 kHz. The experimental sensitivity maps were obtained measuring the perturbation produced by a brass sphere of 12 mm in empty space. The advantage of using a gradiometer is that it can be adjusted to give a minimum signal for homogeneous objects, while increasing the sensitivity to local perturbations of the conductivity. Results show that a system using a planar gradiometer as detector has less demanding requirements for the electronic system than a system using simple coils.

Magnetic induction tomography (MIT) is a contactless method for mapping the electrical conductivity of tissue. MIT is based on the perturbation of an alternating magnetic field by a conducting object. The perturbation is detected by a voltage change in a receiver coil. At physiologically interesting frequencies (10 kHz-10 MHz) and conductivities (<2 S m^-1) the lower limit for the relative voltage change (signal/carrier ratio = SCR) to be resolved is 10^-7-10^-10. A new MIT hardware has been developed consisting of a coil system with planar gradiometers and a high-resolution phase detector (PD). The gradiometer together with the PD resolves an SCR of 2.5×10^-5 (SNR = 20 dB at 150 kHz, acquisition speed: 100 ms). The system operates between 20 and 370 kHz with the possibility of extending the range up to 1 MHz. The feasibility of measuring conductivity spectra in the β-dispersion range of biological tissues is experimentally demonstrated. An improvement of the resolution towards SCR = 10^-7 with an SNR of ⩾20 dB at frequencies >100 kHz is possible. On-line spectroscopy of tissue conductivity with low spatial resolution appears feasible, thus enabling applications such as non-invasive monitoring of brain oedema.

The aim of this study is to propose a useful method for exploring regional ventilation and perfusion in the chest. The paper describes two methods based on singular value decomposition (SVD) and Fourier transform (FT) respectively. This work shows that power spectral density (PSD) and phase images (derived from the Fourier transform) are easier to interpret and more useful tools for exploiting in vivo EIT data in healthy volunteers in order to explore the cardiovascular and respiratory systems.

A study was conducted using the Sheffield electrical impedance tomography (EIT) portable system DAS-01 P to determine the change in the cardiac image with electrode position, lung volume and body position. Sixteen electrodes were positioned in three transverse planes around the thorax at the level of the second intercostal space, at the level of the xiphisternal joint, and midway between upper and lower locations. Data were collected at each electrode level with the breath held at end expiration and after inspiring 0.5, 1 and 1.5 l of air with the subject in both the supine and sitting position. These data were analysed using a Matlab developed program that calculates the average resistivity change in the cardiac region from automatically determined borders. Results show significant individual variability with electrode position and air volume. The middle electrode most consistently shows an increase in impedance in the region of the heart during systole. In some subjects the change in the ventricular-volume-like curve showed a greater than 50% change as a function of lung volume. The pattern of variability with electrode position was not consistent among subjects. In one subject MRI images were obtained to compare actual structures with those seen in the EIT image. The results suggest that using these electrode locations reliable and consistent data, which could be used in clinical applications, cannot be obtained.

Previously, electrical impedance tomography (EIT) has been used to image impedance decreases in the exposed cortex of rabbits during brain activity. These are due to increased blood volume at the site of the stimulated cortex; as blood has a lower impedance than brain, the impedance decreases. During human brain activity similar blood flow changes have been detected using positron emission tomography (PET) and functional magnetic resonance imaging (fMRI). If blood volume also changes then the impedance of human cortex will change during brain activity; this could theoretically be imaged with EIT. EIT data were recorded from a ring of 16 scalp electrodes in 34 recordings in 19 adult volunteers before, during and after stimulation with (1) a visual stimulus produced by an 8 Hz oscillating checkerboard pattern or (2) sensory stimulation of the median nerve at the wrist by a 3 Hz electrical square wave stimulus. Reproducible impedance changes, with a similar timecourse to the stimulus, were seen in all experiments. Significant impedance changes were seen in 21 ± 5% (n = 16, mean ± SEM) and 19 ± 3% (n = 18) of the electrode measurements for visual and somatosensory paradigms respectively. The reconstructed 2D EIT images showed reproducible impedance changes in the approximate region of the stimulated cortex in 7/16 visual and 5/18 somatosensory experiments. This demonstrates that reproducible impedance changes can be measured during human brain activity. The final images contained spatial noise; the reasons for this and strategies to reduce this in future are discussed.

Previous work has demonstrated that electrical impedance tomography can be used to image human brain activity during evoked responses, but two-thirds of the reconstructed images fail to localize an impedance change to the expected stimulated cortical area. The localization failure may be caused by modelling the head as a homogenous sphere in the reconstruction algorithm. This assumption may lead to errors when used to reconstruct data obtained from the human head. In this study a 3D reconstruction algorithm, based on a model of the head as a homogenous sphere, was characterized by simulating the algorithm model, the head shape and the presence of the skull in saline-filled tanks. EIT images of a sponge, 14 cm³ volume with a resistivity contrast of 12%, were acquired in three different positions in tanks filled with 0.2% saline. In a hemispherical tank, 19 cm in diameter, the sponge was localized to within 3.4-10.7% of the tank diameter. In a head-shaped tank, the errors were between 3.1 and 13.3% without a skull and between 10.3 and 18.7% when a real human skull was present. A significant increase in localization error therefore occurs if an algorithm based on a homogeneous sphere is used on data acquired from a head-shaped tank. The increased error is due to the presence of the skull, as no significant increase in error occurred if a head-shaped tank was used without the skull present, compared to the localization error within the hemispherical tank. The error due to the skull significantly shifted the impedance change within the skull towards the centre of the image. Although the increased localization error due to the skull is not sufficient to explain the localization errors of up to 50% of the image diameter present in the images of some human subjects, the future use of a realistic head model in the reconstruction algorithm is likely to reduce the localization error in the human images due to the presence of the skull.

An attempt was made to image the thermal inertia (defined as the square root of the product of thermal conductivity, specific heat and density) of the skin to observe the distribution of blood in the skin during post-occlusive reactive hyperaemia in normal healthy volunteers. The method was based on the ability to calculate thermal inertia by successive thermographic measurements of the skin after stepwise change in ambient radiation temperature surrounding the skin area. The stepwise change was achieved within 0.1 s through a single hood. Experimentation on the undisturbed volar forearm of normal subjects at the same site showed that the measurements thus achieved were reproducible. The thermal inertia values of forearm skin in normal subjects were scattered throughout the range 1.1×10³ to 1.7×10³ W s^1/2 m^-2 K^-1. Experiments on forearm skin subjected to arterial cuff occlusion indicated that thermal inertia can be detected at a low level of blood perfusion. A linear relationship was observed between thermal inertia and blood perfusion measured by laser Doppler imager before and during blood flow occlusion. During reactive hyperaemia, the thermal inertia image exhibited a non-uniform island-shaped pattern of distribution over the forearm, suggesting that, after release from occlusion, recovery of blood flow is non-uniform.

In this paper, the influences of overlying tissue and detecting distance between the source and the detector on the measurement of a tissue oximeter were discussed. The signal-noise-ratio of the detector was also examined. A semi-infinite multi-layer Monte Carlo model was induced to simulate the migration of the photons in the skin, adipose and muscle. The thickness of the adipose layer and the separation between the source and the detector in the muscle were changed to simulate the clinical application. Partial pathlength was introduced as a characteristic parameter to evaluate the sensitivity of the oximeter. A two-wavelength (700 and 830 nm) tissue oximeter was developed to verify the results of the simulation. The Monte Carlo simulation results showed that the sensitivity of the near infrared spectroscopy (NIRS) oximeter declined greatly with increasing overlying tissue thickness. Increasing the distance between the light source and the detector improved the sensitivity. However, in order to achieve a sufficiently high signal-to-noise ratio, it was necessary to limit this distance. The results of a bicycle ergometer exercise verified the above results and showed that, for a special adipose thickness (AT), there may be a reasonable range of the emitter-detector separation.

Several late complications jeopardize the clinical performance of recipients of the Fontan operation. The underlying causes have been referred to disturbed flow dynamics in the cavopulmonary connections. Presumably, the large pressure drops occurring in the inferior and superior connections play a pivotal role in the pressure level of the entire circulation, especially in the venous. To address this issue, we retrospectively reviewed catheterization data of six patients with failing Fontan circulation and compared them with those of six patients with functioning Fontan circulation. The impact on the systemic and pulmonary pressure of the increase in the cavopulmonary connection resistances was studied through a steady-state mathematical model of the univentricular closed-loop circulation. In the patients with failing Fontan, pressure in the venae cavae was found to be significantly higher, especially at the inferior cava (19.3±2.2 versus 12.5±2.3 mmHg) with the pressure drop at the inferior cavopulmonary connection significantly increased (4.7±3.1 versus 0.33±0.82 mmHg). The proposed mathematical model permits us to clearly relate the pressure increase in the venae cavae to an increased resistance in the cavopulmonary connections. Therefore, the present analysis confirms that, to avoid possible congestion of venous circulation, the definitive palliation of univentricular heart should not cause pressure drops at the cavopulmonary connections.

Pulse oximeter probes placed peripherally may fail to give accurate values of arterial blood oxygen saturation (SpO₂) when peripheral perfusion is poor. Since central blood flow may be preferentially preserved, the oesophagus was suggested as an alternative monitoring site. A reflectance oesophageal photoplethysmographic (PPG) probe and a multiplexed data acquisition system, operating simultaneously at two wavelengths and incorporating an external three-lead electrocardiogram (ECG) reference channel, has been developed. It has been used to investigate the suitability of the oesophagus as a possible monitoring site for SpO₂ in cases of compromised peripheral perfusion. Oesophageal PPG signals and standard ECG traces were obtained from 16 anaesthetized patients and displayed on a laptop computer. Measurable PPG signals with high signal-to-noise ratios at both infrared and red wavelengths were obtained from all five oesophageal depths investigated. The maximum PPG amplitude occurred at 25 cm from the upper incisors in the mid-oesophagus. The measured pulse transit times (PTTs) to the oesophagus were consistent with previous measurements at peripheral sites and had a minimum value of 67±30 ms at a depth of 30 cm. There was broad agreement between the calculated values of oesophageal SpO₂ and those from a commercial finger pulse oximeter.

The variation of acoustic attenuation with lung density was determined in experimental studies on seven healthy human volunteers, using a change of lung volume as a means of varying lung density. White noise between 50 and 680 Hz was introduced into the mouth and the transmitted signals were recorded with four microphones on the posterior chest wall (left/right, top/base) at 24, 40, 60 and 80% of total lung capacity. The change in lung volume had a frequency-dependent effect on acoustic attenuation in all subjects. A frequency between 177 and 243 Hz was identified, where altering the lung volume between 24 and 80% of total lung capacity induced a change in attenuation of only 1.0 (±0.5) to 2.7 (±1.8) dB, while at a frequency of 364-436 Hz marked variations in attenuation 8.9 (±2.0) to 21.5 (±4.8) dB occurred with similar lung volume changes.

The theory for optimal linear combination of uncalibrated breathing movements was developed and applied in non-invasive respiratory monitoring situations for assessment. 16 patients were monitored overnight for respiratory depression during postoperative pain treatment. Intranasal/extra-oral airway pressure monitoring and pulse oximetry signals were recorded at 50 Hz. Respiratory inductive plethysmography (RIP) provided guidance to nurses regarding sensitivity settings of the pressure device during slow, shallow breathing, and vital information about breathing movements to help distinguish central from obstructive apnoeas. Subsequent analysis showed that the principal components of the standardized RIP signals would be helpful in any automated identification of pressure indicator false alarms and could provide a simple means for supplementary breath classification. The sum and difference of the scaled RIP values tracked changes in tidal volume and indicated any breathing movement asynchrony or paradox associated with obstructions. A construction was developed for emulating RIP calibration predictions of relative changes in tidal volume to within about 1%, so that invasive or demanding monitoring preparations could be by-passed altogether. The necessary signal combination and linear calibration model background is reviewed for this simple formulation, which arises from component analysis and least squares regression. The methods are illustrated for definitive non-invasive postoperative monitoring and calibration situations. Theoretical and physiological reasons for preferring the use of balanced ribcage and abdomen contributions to overall tidal volume are presented that also help clarify the greater limitations of traditional RIP monitoring practices.

ed T B Martonen WIT Press (2000) 290pp, price: £89.00, ISBN: 1 85312 613 6

The rapid growth in computer power in recent years has stimulated great interest in computational modelling of biological systems, and it is now an important growth area. This book aims to provide both reference material for researchers, and educational material for classwork. Although the title of the book suggests a general review, the material covered is quite specific, focusing on three areas: branching algorithms to generate vascular network geometries, simulated fluid flow in vascular networks and simulated interaction of laser light with the eye.

Chapter 1 gives an introduction to fluid dynamics, covering Newtonian fluids, the Navier Stokes equations, and an introduction to turbulence. Chapters 2 and 3 cover the theory and applications of algorithms to generate simulated vascular trees. Chapters 4 and 5 make up the bulk of the book and describe both simulations of blood flow in branching vessels and application of these principles in surgical reconstruction of the carotid artery bifurcation. The use of laser light in eye surgery is becoming increasingly important, and chapters 7, 8 and 9 cover the physics of the interaction of laser light with the eye, ocular fluid dynamics and thermodynamics, and the use of finite element models of the eye to simulate surgical procedures.

This book clearly demonstrates how the rigorous application of computational mathematics can illuminate problems in medicine and biology, and contains excellent examples. The chapters on fluid flow in branching vessels are particularly good, containing a thorough review of the biological and mathematical aspects of the problem as well as informative results of numerous simulations. The major limitation of the book is the absence of a unifying structure to glue the disparate elements together. Although the book is in general well written, the differences between the three subject areas are accentuated by an inconsistent style. Each chapter has an abstract, but the book has no index and some of the figures and diagrams are poorly reproduced. If you have an interest in one of the three subject areas covered in the book it may be worth buying it, although it is likely that two thirds of the book will remain unread.

ed Al Bovic San Diego: Academic Press (2000) 891pp, price: £64.95, ISBN: 0 12 119790 5

If you want to know anything about image and video processing, this book is the place to begin and, probably, to end as well. Encyclopaedic in scale, 58 chapters by 97 contributors are packed into 891 pages. The subject is divided into 10 sections: an introduction, basic image processing techniques, image and video processing, image and video analysis, image compression, video compression, image and video acquisition, image and video rendering and assessment, image and video storage retrieval and communications, and applications of image processing.

The boundaries of the coverage of the subject are quite sharply defined. First of all, an image is taken to mean a still picture that does not change with time, whereas a video evolves with time and generally contains moving and/or changing objects. There are two chapters on image and video acquisition; these are concerned with image capture, Fourier analysis, sampling rate, interpolation and reconstruction of continuous time-varying imagery. Neither image display nor perception are mentioned, even in the index. Image fusion is not included. Thus, the book is strictly limited to image and video processing and that is certainly not a criticism.

As far as the depth of coverage is concerned, the objective is ambitious. This is to provide a resource that covers introductory, intermediate and advanced topics with equal clarity. I tested the extent to which this has been achieved by trying to find the answers to some questions about image and video compression. In the 277 pages of the 10 chapters devoted to this topic, there is a wealth of information extending from lossless coding, through other coding and quantization schemes, wavelet and JPEG compression, video and object-based coding, to MPEG video standards. I learned enough to realize that the subject is so mature, and the investment in both hardware and software so significant, that there must be little to be gained from research at the margins of any new approaches.

It is the final section (134 pages) that contains the chapters that will be of most immediate interest to readers with a medical bias. This section is concerned with applications of image processing. The relevant chapters are on: computed tomography (x-ray, CT, PET and SPECT) with the emphasis on image reconstruction and three-dimensional topics; cardiac image processing, including the analysis of cardiac mechanics, perfusion and (perhaps oddly) electrocardiography; computer aided detection for screening mammography, concentrating on masses, calcifications and segmentation; and confocal microscopy. It is actually in these chapters that most information is to be found concerning the instrumentation for image acquisition.

This is not an expensive book. Indeed, it is stupendously good value-for-money. Where else, in the field of image and video processing, can the knowledge of 97 contributors, all clearly expert and many deservedly famous, be so conveniently accessed? If there is a criticism, it has to be that the 55 pages of colour illustrations (which are additional to the 891 pages of text) are gathered together in four blocks, distributed through the book. These illustrations also appear in the text, but in grey scale. So, it is sometimes necessary to turn to the colour sections to appreciate the points that are being made. But this is only a small nuisance in exchange for what is presumably a substantial economy in the price of the book.

Realistically, this is too big a book to be read from cover to cover. Consequently, it needs a good index and there is no criticism here. For example, I looked up 'ultrasound imaging' and found a reference to intravascular scanning and its combination with biplane angiography to create a map of the vessel in 3D space; this is a technique of which I was previously unaware. I also found a reference to echography, with an informative discussion of segmentation which, in this situation, is a far from trivial task.

This handbook is the first in a new Academic Press series in communications, networking and multimedia, with Jerry Gibson of Southern Methodist University as the editor-in-chief. In editing and co-authoring the handbook, Al Bovik of the University of Texas at Austin has discharged a monumental assignment with spectacular success. If you are interesting in image and video processing, you must have a copy.