Imaging brain function with optical topography
For more than two decades, the single channel measurement technique of NIR spectroscopy (NIRS) has been successfully used to measure the haemodynamic response to brain activity in both adults and neonates (Hoshi 2003, Ferrari 2004). It has been used to record functional activity for research into brain cognition, and to examine brain development in infants (Obrig and Villringer 2003). An obvious limitation of NIRS is the lack of any spatial information. It is natural to consider combining multiple NIRS measurements in order to localise the origin of signals in the brain; indeed, the first optical topography studies were carried out in this way (Gratton et al. 1995).
NIRS and optical topography measure the haemodynamic response following brain activation and therefore rely on the spatial and temporal variability in the concentration and oxygenation of haemoglobin in the blood. The penetration depth of optical topographic systems is generally assumed to be approximately half the optode spacing. For typical spacings around 25-35 mm, this gives a penetration depth of about 15 mm, which is sufficient for sampling the adult cortex. Recent reviews of optical topography have been provided by Strangman et al. (2002a), Hebden (2003) and Koizumi et al. (2003).
The most commonly used method for imaging brain function is functional MRI, which is sensitive to the Blood Oxygen Level Dependent (BOLD) signal. BOLD MRI detects signal changes caused by the magnetic susceptibility of HHb being greater than that of HbO and other brain tissues, and generates images of these changes in a few seconds. It is currently viewed as the “Gold Standard” for functional brain imaging due partly to the ability to register fMRI images directly onto a structural MR image taken during the same examination. However, optical topography does have some advantages over fMRI which have led to it developing an increasing role in functional brain imaging. For example, images can be acquired very rapidly (a 50 Hz acquisition rate has been demonstrated by Franceschini et al. (2003)), and it is inexpensive and portable, enabling images to be obtained at the bedside or in the laboratory. Moreover, mechanisms which generate the NIRS signal are closely correlated with the BOLD signal of fMRI but add information by distinguishing quantitatively between changes in [HHb] and [HbO] (Strangman et al. 2002b). The primary drawback of optical imaging is the inherently low spatial resolution, which is about 10 mm at best. However, although the spatial resolution of fMRI is typically 3 mm and can be as good as 1 mm, the image processing techniques which are commonly used model the image as a smoothed random field and so impose a smoothing filter with a 8-10 mm Gaussian kernel before analysis (Turner and Jones 2003). In certain circumstances, therefore, there may be little difference between the effective spatial resolutions of the two techniques.
Optical topography of the adult brain has been largely pioneered by researchers from the Hitachi Medical Corporation (Tokyo, Japan), who have used the ETG-100 Optical Topography System (Figure 3), described in section 2.2, to examine a range of cognitive tasks (Takahashi et al. 2000, Koizumi et al. 2003). Optical methods are well suited for imaging tasks related to language, as the investigation is silent, unlike fMRI (Watanabe et al. 1998). Optical topography is beginning to be routinely applied to pathological conditions: Watanabe et al. (1998, 2002) showed that optical topography could detect changes in [HbO] and [HHb] during epileptic seizures, and Matsuo et al. (2003) showed a difference in the response to viewing traumatic images between normal controls and patients who suffered post traumatic stress disorder from the 1995 sarin nerve gas attack on the Tokyo subway. Meanwhile, Obata et al. (2003) showed that visual evoked responses were not affected by alcohol.
Franceschini et al. (2003) have used a related approach to image the sensorimotor cortex using a system with 8 laser diode sources at each of two wavelengths and 16 detectors. Each channel was frequency-encoded, allowing all the sources and detectors to be active simultaneously and decoded in software to maximise the image acquisition rate. Using this method, images could be obtained at 50 Hz. They found an increase in [HbO] and a decrease in [HHb] contralateral to the stimulated side with a reduced ipsilateral response, consistent with equivalent fMRI and PET studies.
The Hitachi approach generates images by assuming that a change in intensity originates midway between the source and detector which measured that change. The image is produced by mapping the changes according to the positions of the appropriate source-detector pair. This technique has some drawbacks: the spatial resolution cannot be better than the optode spacing, it is difficult to apply it to irregular arrangements of connectors, and it does not naturally allow the inclusion of prior information (Yamamoto et al. 2002). These limitations have been addressed by Boas et al. (2001c) who showed that a linear reconstruction approach improves the quantitative accuracy compared to single channel NIRS measurements (and, by implication, images generated by mapping, which are equivalent to multiple NIRS measurements). Furthermore, such an approach allows multiple topographic measurements to contribute to each pixel, which can yield a two-fold improvement in spatial resolution and localisation accuracy (Boas et al. 2004a, b). Culver et al. (2003c) have used a similar approach based on linear reconstruction to image the rat cortex.
Most optical topographic images have been obtained using systems which measure intensity alone, meaning that the effects of µa and µ's cannot be uniquely separated (Arridge and Lionheart 1998). This has been addressed by Franceschini et al. (2000) who used a frequency-domain system to acquire images of activity in the adult cortex with a time resolution of 160 ms. The system uses 8 sources at each of two wavelengths (758 and 830 nm) and two detectors, and is produced commercially by ISS Inc, USA. Toronov et al. (2001a) used a similar system to record optical data during motor activity simultaneously with fMRI images. They found close spatial and temporal agreement between the optical and BOLD signals. In a related study, Toronov et al. (2001b) showed that both the amplitude and the phase of the NIRS signal correlated with the BOLD response.
Optical imaging techniques are particularly well suited to imaging infants, being portable and somewhat less sensitive to motion artefact than fMRI. The first report of optical topography on premature babies was by Chance et al. (1998a) who used a system with 9 sources and 4 detectors on adults and on a 4-week old neonate. The system measured intensity only but used a phase cancellation technique to improve sensitivity. Sources were arranged on a grid and a signal applied to alternate sources at phase shifts of either 0? or 180? such that, in a homogeneous medium, the phase shift is 90? and the amplitude is zero directly between the sources, where the detectors are placed. The Hitachi systems have also been used to investigate brain activity in infants. Taga et al. (2000) observed spontaneous fluctuations in [HHb] and [HbO] in eight sleeping neonates with periods of 8-11 s which were attributed to vasomotion. Oscillations in [HHb] led those in [HbO] by 3?/4, which was explained by a localised increase in brain activity leading to an increase in [HHb], followed by an increase in cerebral blood flow which supplies additional oxygen to increase [HbO]. A similar effect has been observed in BOLD fMRI. More recently, the same group examined 20 infants a few months old and were able to record images during visual activation from eight infants (Taga et al. 2003). The other 12 were rejected due to movement, crying, lack of attention and poor contact due to hair. They concluded that the convenience of optical topography represented a significant advantage over other techniques. Kusaka et al. (2004) carried out a similar study using a CW system supplied by Shimadzu Corporation and showed different responses between adults and infants. Higher cognitive functions have also been imaged using the Hitachi system - Peña et al. (2003) showed that the temporal cortex in neonates is activated more strongly by normal speech than by speech played backwards, and Tsujimoto et al. (2004) showed that the same area of the brain, the lateral prefrontal cortex, appears to be responsible for working memory in both adults and pre-school children. Both of the latter two articles comment on the convenience with which optical topography can be used to image babies and children. Vaithianathan et al. (2004) has built and tested a CW topography system with an interface consisting of a flexible pad, designed to be conveniently applied to an infant’s head.
Hintz et al. (2001) used a CW topography system to image passive motor activity in premature infants. Their system acquires a total of 144 independent measurements in approximately 3 s, and images were reconstructed using a non-linear approach. Bluestone et al. (2001) used an even more sophisticated image reconstruction technique to generate 3D images of a region of interest beneath the adult forehead during a Valsalva manoeuvre, in which the intrathoracic pressure is increased (and therefore cardiac output decreased) by expiration against a closed airway. They used a CW system (Schmitz et al. 2002) to acquire data from 15 optodes on the forehead with a temporal resolution of 3 Hz.
Optical topography has been extensively used to image functional brain activity in both adults and neonates. During the past five years it has evolved beyond a laboratory technique to be used to address hypothesis-led questions about neurophysiology and brain development. It has unique advantages over other brain imaging techniques, including its excellent temporal resolution and its ability to distinguish between [HHb] and [HbO]. It can be used in a natural, relaxed environment, making it well suited for psychological studies, and it can be used to image awake infants.
Imaging the neonatal brain with optical tomography
Optical topography is able to measure and display haemodynamic changes occurring in the cortex, but is far less sensitive to deeper tissues. Optical tomography, however, uses widely spaced sources and detectors to measure light which has passed through central regions of the brain, which can be affected by disruption to the supply of blood and oxygen around birth. Light travelling through these regions is heavily attenuated and therefore more intense light sources and more sensitive detectors are required. Even so, the acquisition time is typically a few minutes compared to less than a second in the case of optical topography.
Brain injury in preterm and term infants is a major cause of permanent disability and death. Several groups have developed optical tomography of the infant brain with the aim of providing diagnostic and therapeutic information about the more important types of brain injury. Intraventricular haemorrhage may occur in premature infants whose cerebral vasculature is too weak to withstand the fluctuations in blood pressure which occur during birth (Whitelaw 2001), while periventricular leucomalacia is damage to the white matter which is common in premature infants (Volpe 2001). Hypoxic-ischaemic brain injury around the time of birth is a major cause of brain injury in the term infant (Wyatt 2002). These three mechanisms of brain injury all manifest themselves as disruption to the supply of blood and oxygen to vulnerable areas of the brain. Currently, they are either diagnosed clinically, by ultrasound (which gives only anatomical information) or MRI (which clinicians may be reluctant to request if it means moving a very ill infant out of an intensive care environment). Optical tomography may provide a bedside system to identify infants at risk, to diagnose injury and to monitor treatment (Thoresen 2000). For a detailed review of optical imaging of the neonatal brain, see Hebden (2003).
The first tomographic images of the neonatal were recorded by the group of Benaron, by reconstructing measurements of mean photon flight time made across the neonatal head. Thirty-four pairs of sources and detectors were attached to the head using a headband (Hintz et al. 1998) and were used to obtain 2D tomographic slice images of anatomical disorders (Hintz et al. 1999, Benaron et al. 2000) and functional activation (Benaron et al. 2000). Eight optical images were generated from six infants in the first month of life, four of whom had IVH. Six of the eight images compared favourably with CT, ultrasound and MRI images. Later, Benaron et al. (2000) generated images from two neonatal subjects, one a control and the other with a stroke. They detected a significant difference between the two subjects, and the location of the stroke agreed with a CT image. Images were also presented, using the same approach, of functional motor activity. The images obtained in these studies were remarkable, given the relatively simple instrumentation (Benaron et al. 1994b) and crude reconstruction techniques (Benaron et al. 1994a), and encouraged the suggestion that optical tomography could play a major diagnostic role in neonatology.
More recently, the group at University College London has used a purpose-built 32 channel time-resolved imaging system (Schmidt et al. 2000) and a non-linear image reconstruction approach based on finite element modelling (Arridge et al. 2000b) to generate 3D images of the neonatal head. Sources and detectors are coupled to the head using a foam-lined helmet which is custom built for each infant. The positions of the connectors within the helmet are used to define the outer surface of a finite element mesh (Gibson et al. 2003) which is used for image reconstruction. The first published results were images of the brain of an infant with IVH, which revealed an increase in blood volume in the region of the haemorrhage as shown on an ultrasound examination (Hebden et al. 2002). Another of the babies studied so far was mechanically ventilated, allowing images to be reconstructed from data acquired at two wavelengths during changes in inspired oxygen and carbon dioxide (Hebden et al. 2004b). The reconstructed images of [HHb] and [HbO] agreed qualitatively with physiological predictions. Recently, images have been obtained from a pair of twins, recorded consecutively on the same day. One twin was anatomically normal while the other had an IVH which had caused additional bleeding into the brain tissue. Images of blood volume and oxygenation in the baby with the normal brain (Figure 4a) are symmetrical about the midline and appear to show a decrease in blood volume in the white matter. The haemorrhage (Figure 4b) can be seen as an increase in blood volume and, a decrease in oxygenation (from about 65% to about 10%). The location of the increase in blood volume correlated well with the site of the IVH determined from an ultrasound image, but interestingly, the location of the increase in oxygenation appeared to correlate more closely with the infarct of the haemorrhage into the brain tissue.
Encouraging results have been obtained by neonatal optical tomography, although so far the application is not as widely used as optical topography. This is partly due to the additional complexity of the instrumentation and the practical difficulties of acquiring data from premature babies who may be very ill. A further difficulty is obtaining an adequate reference measurement for data calibration. Theoretical advances such as solving for the coupling coefficients and the incorporation of prior anatomical information from MRI and improvements to instrumentation are expected to lead to significant improvements in image quality and clinical acceptance.
In 2000, more than 1 million women were diagnosed with breast cancer worldwide (Ferlay et al. 2001). Early detection decreases mortality (Tabar et al. 2003), so many countries routinely screen for breast cancer. X-ray mammography is the screening method of choice (Fletcher and Elmore 2003) but its effectiveness may depend on issues such as the age of the woman, family history of cancer, body mass index, the use of hormone replacement therapy, the use of computer aided detection, and the availability of additional clinical or imaging information (Blarney et al. 2000, Warren 2001, Houssami et al. 2004, Banks et al. 2004). It requires ionizing radiation and its benefits for younger women are unclear (Lucassen et al. 2004). Other imaging modalities such as MRI and ultrasound may be useful in certain circumstances but neither is suitable for screening of asymptomatic women (Morris et al. 2003, Warner et al. 2004).
It is well known that tumours are associated with increased vascularisation (Rice and Quinn 2002) so optical methods provide a natural method for both interrogating tissue to identify disease and for spectroscopically determining the blood volume and oxgenation of a suspicious lesion seen on x-ray mammography to improve specificity. A major disadvantage of optical mammography is the inherently poor spatial resolution. A successful screening tool must be able to identify tumours smaller than 1 cm, as mortality increases rapidly for tumours which exceed this size (Webb et al. 2004). Because of this, optimizing spatial resolution is an important concern in optical mammography. Researchers are also exploring other uses of optical mammography which are less dependent on resolution, such as staging previously identified suspicious lesions, and monitoring the response to new and existing forms of therapy. One of the attractions of the technique is its suitability for repeated investigations on the same subject.
Despite many previous efforts to determine them, the in vivo optical properties of healthy breast tissues and common lesions are not fully known. Most early work (e.g. Peters et al. (1990)) concentrated on measuring the properties of breast tissue in vitro and, as blood and water are the primary chromophores, it is not clear how in vitro values relate to the intact breast. Even optical properties measured in vivo will depend on the geometry of the measurement (particularly if the breast is compressed) and the model used to derive the optical properties from the measurements. Measurement at a single location can give information only about the local average properties. Even so, such measurements have been shown to correlate with age, body mass index and menstrual state (Shah et al. 2001, Cerussi et al. 2001). Perhaps the most reliable measurements of breast optical properties to date are those of Durduran et al. (2002), who obtained frequency-domain measurements at 750, 786 and 830 nm in 52 women using a parallel plate geometry with gentle compression. The average optical properties (± standard deviation) were µa = 0.0041 ± 0.0025 mm-1, µ's = 0.85 ±0.21 mm-1 (at 780 nm), blood volume = 34 ± 9 µM and oxygen saturation = 68 ± 8%. These values agree with other published in vivo values (e.g. Suzuki et al. (1996)), but the µa is approximately double that of corresponding in vitro measurements, possibly because of the lower blood volume in excised breast tissue. Grosenick et al. (2004b) measured the optical properties of tissue and tumour in 50 women at 680 and 785 nm by assuming a spherical tumour in an infinite, homogeneous slab. They found that µa in the tumour was between two and four times that of surrounding tissue due to the increased blood volume, and µ's was slightly elevated. These results agree with similar measurements made by Fantini et al. (1998) and Chernomordik et al. (2002b). Holboke et al. (2000) report a similar increase in µa but a 50% reduction in µ's.
A number of alternative approaches to optical mammography have been evaluated. Most clinical studies have been performed using instruments which compress the breast, though generally more gently than in x-ray mammography. This reduces the attenuation of the transmitted light and ensures that the geometry of the problem is well-known. The breast is compressed between either two parallel arrays of sources and detectors, or between plates over which individual sources and detectors are scanned in a rectilinear manner. However, the latter method is only suitable for generating projection images (analogous to x-ray mammograms) since it does not yield sufficient depth information for a 3D reconstruction.
Groups based in Berlin and Milan have between them performed more than 300 clinical studies as part of Optimamm, a consortium funded by the European Union (see www.optimamm.de and a forthcoming special issue of Physics in Medicine and Biology). Both systems acquire time-domain measurements at multiple wavelengths by scanning a single source and detector in tandem across a gently compressed breast. The Milan group have explored the effect of using a wide range of wavelengths (Pifferi et al. 2003, Taroni et al. 2004a), and the Berlin group have investigated the benefits of acquiring additional “off axis” measurements (Grosenick et al. 1999). Figure 5 shows an optical mammogram recorded by Grosenick et al. (2003) with two views of a tumour in the left breast of a patient, and the corresponding images of the healthy right breast. Both groups report that they can successfully identify around 80- 85% of radiologically identified tumours (Grosenick et al. 2004a, Taroni et al. 2004b), and it is likely that second generation systems which incorporate improved spatial and spectral discrimination will yield improved detection rates.
Compressed breast geometries have also been evaluated for imaging by Pera et al. (2003) using a frequency-domain instrument built by Siemens Medical Engineering, and by Culver et al. (2003a) using a hybrid system which determines the bulk optical properties using frequency-domain measurements and spatial information from up to 105 continuous wave measurements (see section 2.4).
One of the potential disadvantages of breast compression is the corresponding reduction in the blood volume. Although this improves overall transmission, blood represents the principal source of contrast between tissues, and the most likely means by which tumours may be identified and characterised. Several groups avoid compression by surrounding the breast with rings of sources and detectors. This arrangement is also ideal for generating 3D images. The main disadvantage of this approach is that a much greater volume of tissue is sampled, and therefore the detected light intensities are much lower and have a greater dynamic range than for the compressed breast geometry.
The 3D breast imaging approach has been pioneered by the group at Dartmouth College. They are exploring optical mammography as part of a larger study involving four different breast imaging modalities (optical, impedance, microwave and MR elastography (Poplack et al. 2004)). They use a frequency-domain optical tomography system (McBride et al. 2001) with 6 laser diodes operating at wavelengths from 660- 836 nm multiplexed to 16 source positions, and 16 detectors, located around the circumference of a 2D ring whose diameter can be varied to fit the breast. 3D images are reconstructed from three 2D datasets, each consisting of amplitude and phase measurements (Dehghani et al. 2003b). Recently, the group has successfully extended their technique to reconstruct directly for haemoglobin and water concentrations, blood oxygenation, and scattering parameters (Pogue et al. 2004). A similar approach has been investigated by Jiang et al. (2002) using a single-wavelength CW system.
The 32-channel time-resolved system at UCL has also been used to generate 3D images of the uncompressed breast, either by using a 2D ring of connectors or a 3D hemispherical tank filled with a tissue matching fluid (Yates et al. 2004), an approach originally employed for the prototype developed by Phillips Medical Systems (Colak et al. 1999).
A sophisticated CW system for imaging both breasts simultaneously has recently been presented by Barbour et al. (2004). It is designed to acquire images which reveal haemodynamic phenomena within the breast.
Because the prognosis for breast cancer depends largely on the size of the tumour, the relatively poor spatial resolution of optical mammography may limit its use in routine screening. However, the technique may provide a powerful method for the examination of suspicious lesions previously identified by other means. With this in mind, attempts have been made to use the prior information from other medical imaging techniques to condition the optical image reconstruction. In principle, this will allow images with the excellent physiological information content of optical imaging to be reconstructed with the high spatial resolution provided by MRI, ultrasound or x-ray mammography. Ntziachristos et al. (1998) built and tested a combined MR/optical imaging system and successfully used it to record optical and MR images from volunteers with a range of benign and malignant lesions (Ntziachristos et al. 2000, 2002c). Li et al. (2003) built an optical imaging probe which was interchangeable with a standard film cassette in an x-ray mammography tomosynthesis instrument. They used this combined system to generate 3D x-ray images of the compressed breast which they segmented into “suspicious” and “background” regions, which were then used as anatomical prior information in the optical tomography image reconstruction. The spatial resolution of the optical image was enhanced such that the reconstructed absorber was confined to a volume even smaller than the suspicious region identified in the x-ray images (see Figure 6).
Exogenous contrast media can also be used to improve the sensitivity to small lesions. The most widely used contrast agent is indocyanine green (ICG), which has been approved for use on human subjects by the US Food and Drug Administration as an NIR absorbing and fluorescing dye, and has been used for breast imaging by Ntziachristos et al. (2000) and Intes et al. (2003). When injected intravenously, ICG binds immediately and totally to blood proteins, primarily albumin. This ensures that ICG is confined almost entirely to the vascular compartment except for incidences of abnormal blood capillaries with high permeability, as in the case of tumours with high vascularity. Thus ICG is primarily an indicator of blood volume, although it provides a certain degree of specificity for some types of tumour. Other, more selective contrast agents and fluorescent dyes for breast cancer detection are currently being evaluated in animal models (Ntziachristos and Chance 2001) (see section 4.5 below).
One of the attractive benefits of using contrast agents for optical tomography is that the imaging process involves reconstructing a change in optical properties, which is generally more tolerant of experimental and modelling errors. Therefore the use of contrast agents could lead to significant improvements in image quality, albeit at the expense of a more invasive investigation.
The application of optical imaging to other areas of the body is restricted by the limited penetration of light across large thicknesses of tissue. Muscle tissue has been widely investigated using NIRS, although relatively few optical imaging studies have been reported. The forearm muscle has probably been the most frequently studied, the first time by Maris et al. (1994) using a frequency-domain optical topography system to map differences in the hemoglobin oxygenation in finger extensor muscles during exercise. Graber et al. (2000) have used a CW optical tomography system for a series of real-time dynamic arm imaging experiments, and Araraki et al. (2000) and Hillman et al. (2001) have used time-domain systems to measure changes in tissue absorption in response to finger flexion exercises at two wavelengths from which cross-sectional images of haemoglobin concentration were reconstructed.
Another relatively advanced application of diffuse optical imaging is the examination of the finger for rheumatoid arthritis. Hielscher et al. (2004) have produced images by scanning a single source and detector along the finger, in which contrast is provided by an increase in blood volume due to inflammation and changes in the optical properties of the synovial fluid. Because the synovial fluid is low scattering and the diameter of the finger is small, the diffusion approximation may not be appropriate for image reconstruction (Hielscher et al. 1999). Similar work is being carried out by Xu et al. (2001, 2002b) who have generated 3D images using the diffusion approximation from continuous wave data.
A related application, which could prove to be clinically significant but is not yet an imaging technique, is the optical examination of the heel bone, which is routinely examined by x-ray in middle-aged women to detect osteoporosis. Pifferi et al. (2004) have obtained preliminary transmission spectra through the heel and found that the bone mineral density (determined from the absorption spectrum of bone) decreased with age in seven volunteers. Ugryumova et al. (2004) have proposed a similar application based on optical coherence tomography. 4.5 Molecular imaging Molecular imaging is a rapidly growing research field in which contrast agents are bound to specific proteins or genes and used to image the distribution of targeted molecules in small animal models (Weisslader and Ntziachristos 2003, Cherry 2004). The recent growth in molecular imaging has largely been triggered by the success of the Human Genome Project and improvements in biochemical techniques. Molecular imaging has traditionally involved tracking radioactive tracers (Goertzen et al. 2002, Comtat et al. 2002) but optical techniques, particularly fluorescence imaging, are increasingly finding a role (Cubeddu et al. 2002, Ntziachristos et al. 2002b). Combining the new contrast agents with diffuse optical imaging in animal models is likely to become a very powerful tool in the development of new drugs and the study of haemodynamics in animal models (Culver et al. 2003b).
Several interesting theoretical issues must be addressed in molecular imaging. The animal typically used in drug development is small compared to human organs and photons may travel only a few transport scattering lengths before detection. This means that the diffusion approximation may not be valid, and therefore higher order approximations are required (Klose and Hielscher 2003). Because of the difficulty of attaching optical fibres to such a small experimental model, Schulz et al. (2003, 2004) have developed non-contact imaging techniques, based on first determining the surface of the animal and then solving a forward model where a free-space region is coupled to a diffusive region. Hillman et al. (2004) have developed an approach, designed for imaging the rat cortex, in which a microscope acts as a non-contact source and detector, providing data which is modelled using a Monte Carlo simulation and reconstructed using diffuse optical tomography techniques. The use of prior information in optical molecular imaging is also valuable, both to assist the optical image reconstruction (Xu et al. 2003) and to independently validate the optical results (Siegel et al. 2003).
Fluorescence imaging is ideally suited to molecular imaging as the small penetration depths in small animals ensures that a strong signal is obtained. However, Ntziachristos et al. (2002a) are optimistic that, given improvements in detector technology, it will be possible to measure a fluorescence signal across the breast, and Godovarty et al. (2004) have already measured fluorescence signals across a breast- like phantom. Progress in this area could lead to development of a range of valuable clinical techniques (Ntziachristos and Chance 2001). Fluorescence imaging of the breast has been reviewed in detail by Hawrysz and Sevick-Muraca (2000).