lication of Micrographia by Robert Hooke in 1665.2 Hooke was an accomplished maker of ... The discovery of cells by Hooke and bacteria and protozoa by.
Introductory Lecture From biomolecules to biodiagnostics: Spectroscopy does it all Michael Jackson Institute for Biodiagnostics, National Research Council Canada, 435 Ellice Ave., Winnipeg, Manitoba, Canada R3B 1Y6 Received 28th October 2003, Accepted 28th October 2003 First published as an Advance Article on the web 2nd December 2003
Optical spectroscopy and imaging have been used in medicine since medicine was first practised. However, the more sophisticated instrumental methods now under development have made little impact on clinical medicine. In this paper a brief overview of the development of optical diagnostics is presented, highlighted by some successful pre-clinical applications. The reasons for the slow penetration of optical diagnostics into clinical practice are discussed.
The past Spectroscopy and imaging are, and always have been, at the core of medicine and biology. The earliest, and still the most common, medical diagnostic instrument is actually a highly advanced spectroscopic imaging system. This spectroscopic imaging system is capable of interrogating skin and determining the extent of blunt traumatic injury and alterations in blood flow, metabolism and tissue hydration (oedema). The imaging system has a self-adjusting aperture to control light levels and a self-focussing lens. The light-sensing device is an intricate arrangement of four types of exquisitely sensitive photosensors in a two-dimensional array of hundreds of millions of pixels covering an impressive spectral range. Spectroscopic images are acquired and transmitted to a control unit in real time, where the data is processed using a complex neural network. This spectroscopic imaging system can be viewed as the prototype for all modern optical spectroscopic imaging systems. Of course this spectroscopic imaging device is the human visual system. The human visual system allows us to evaluate the degree of redness of skin, which is directly related to local blood flow. Changes in blood flow may be related to emotional factors (rage, embarrassment), hormonal changes (for example in menopause) or illness/injury (inflammation). The same system allows us to evaluate oedema (accumulation of tissue water) and the extent of blunt trauma injury (the severity and extent of bruising). Using this spectroscopic imaging system, humans are able to determine the emotional state of a potential enemy, the hormonal state of a potential mate and to assess the degree and extent of injury. Impressive though the human visual system is, it has four fundamental limitations: spectral range, spatial resolution, sensitivity and specificity. As humans evolve and want to ‘‘ see ’’ and understand more of the world around us, these limitations mean that the human visual system can no longer provide the answers we are looking for. Nowhere is this more obvious than in the fields of biology and medicine. To overcome these fundamental limitations and allow more advanced diagnosis, new, enabling technologies are required. DOI: 10.1039/b313625f
Faraday Discuss., 2004, 126, 1–18 This journal is # The Royal Society of Chemistry 2003
1
The first of these enabling technologies to be developed was thermal imaging. The ancient Egyptians practised a crude form of ‘‘ thermal medicine ’’, in which areas of disease were identified based upon temperature differences sensed by running the fingertips over the body. The relationship between body heat and disease was also known to the ancient Greeks, including the Greek physician Hippocrates who wrote in 400 BC ‘‘ in whatever part of the body excess of heat or cold is felt, the disease is there to be discovered. ’’ The ancient Greeks improved upon the Egyptian diagnostic technique with the development of a crude form of thermal imaging. The Greeks immersed the body in wet mud and areas that dried more quickly than others, indicating a warmer region, were considered to be diseased.1 Primitive though this technique is, it should be considered a considerable advance, as it one of the earliest documented uses of contrast enhancement in medicine. Medical imaging remained limited to simple visual inspection for two millennia, until the publication of Micrographia by Robert Hooke in 1665.2 Hooke was an accomplished maker of compound microscopes, invented in 1595, and Micrographia was a detailed illustrated record of Hooke’s microscopic observations of insects and plants. The most famous of Hooke’s observations is his studies on thin slices of cork. In ‘‘ Observation XVIII ’’ of the Micrographia, he wrote: ‘‘ I could exceedingly plainly perceive it to be all perforated and porous, much like a Honey-comb, but that the pores of it were not regular.. . .these pores, or cells,. . .were indeed the first microscopical pores I ever saw. . . ’’. Hooke’s observations intrigued a Dutch tradesman Antony van Leeuwenhoek, who began building his own simple microscopes. Leeuwenhoek’s skill at grinding lenses enabled him to build microscopes that magnified over 200 times, allowing him to see clearer and brighter images than any of his contemporaries, including Hooke, could achieve. With these incredibly simple instruments (essentially a single lens and a spike as a sample holder) Leeuwenhoek discovered protozoa (in 1674) and bacteria (in 1683).3 The microscope represents the first truly enabling (and transformational) technology in medical imaging and spectroscopy. The discovery of cells by Hooke and bacteria and protozoa by Leeuwenhoek were key discoveries that lead to the foundation of pathology as a modern discipline and the postulation of the Schleiden–Schwann cell theory,4 a fundamental tenet of modern biology. The microscope clearly alleviated one of the four fundamental limitations to visual diagnosis: spatial resolution. However, the problems of spectral range, specificity and sensitivity still remained. Specificity and sensitivity to disease states were addressed to some extent by the development of histological staining techniques. Essentially, staining improves sensitivity and specificity based upon reactions between exogenous agents and tissue components to produce differential staining of cell and tissue components. The most widely used staining technique is that of haematoxylin and eosin, (H&E), in which basophilic structures such as the nuclei of cells are stained blue, black, purple or grey by the haematoxylin whilst the acidoiphilic structures are coloured pink or purple by the eosin. Haemotoxylin staining was first demonstrated over a hundred years ago by Waldeyer using an extract of logwood5 (Haematoxylon campechianum). The reliance on natural dyes diminished considerably with the synthesis of the first aniline dye (aniline purple or mauve) and staining techniques for tissue began to proliferate with advances in synthetic capabilities. For example Gram produced his stain for microorganisms in 18846 followed by Daddi’s ‘ Sudan ’ stain for lipids in 1896.7 The development of microscopy and staining techniques heralded a period of rapid advances in tissue and cellular characterisation. However, medicine still relied exclusively upon the visible region of the electromagnetic spectrum. This reliance precluded the direct acquisition of information concerning the biochemistry of disease states and severely limited the nature of samples that could be analysed. Simply put, visible microscopy suffered from the need to have very thin sections of tissue (allowing only ex vivo studies) and a lack of visible chromophores (relying upon the addition of stains to be able to deduce and biochemical information). Two key developments in the 19th century changed this. The first was the discovery of infrared light in 1800 by the British astronomer Sir William Herschel. Herschel discovered the existence of infrared radiation when he tried to measure the heat produced by separate colors of a rainbow spectrum.8,9 He noted that the highest temperature fell beyond the red end of the spectrum, implying the existence of invisible light beyond the red. Herschel termed this light ‘‘ calorific rays ’’, a term soon abandoned in favour of infrared (literally meaning beyond the red). Experiments continued with infrared light, leading to the construction of the first infrared spectrometer in 1835.10 In 1840 Herschel’s son, Sir John 2
Faraday Discuss., 2004, 126, 1–18
Herschel, acquired the first thermal image of the sun on a piece of paper (using carbon particles suspended in alcohol).11 This image, which he termed a thermogram, formed the basis of modern thermal imaging. Infrared spectroscopy then expanded rapidly, thanks to renowned spectroscopists ˚ ngstrom, whose pioneering work between 1850–1900 established IR such as Coblentz and A spectroscopy as an important tool for chemical characterisation and astronomy. The second major event of importance to medical image in the 19th century was the discovery of X-rays by Wilhelm Ro¨ntgen in 1895.12 Ro¨ntgen was investigating the range of cathode rays in air when he noticed that a screen coated in the fluorescent material barium platinocyanide would glow while the rays were being produced. He assumed unknown rays, which he called X-rays, were being emitted from the walls of the tube. To his surprise, Ro¨ntgen found that the rays could pass through his hand and cast shadows of his bones on the fluorescent screen. In this way X-ray imaging was born (with help from Thomas Edison, who developed the X-ray imaging screen within a year of Ro¨ntgen’s discovery of X-rays). The next major development in the field of spectroscopy and imaging that has relevance to biology and medicine is the discovery of Raman scattering in 1928 by Sir Chandrasekhara Venkata Raman.13 who was awarded the Noble prize for the discovery in 1930. Raman had a strong interest in scattering, having previously demonstrated that the colour of the sea was dictated by scattering. In further experiments Raman demonstrated that when sunlight was focussed through the optics of a telescope and through a series of filters (to produce monochromatic light) onto a sample, most of the scattered light is observed at the same frequency as the incident light. However, a small fraction of the scattered light is shifted in frequency, a process now known as Raman scattering. This frequency shift is now known to be the result of absorption of photons to promote molecular vibrations and re-emission of photons at a frequency corresponding to the difference in energy between the photon absorbed and the energy required to promote the molecular vibration. Although the Raman effect was demonstrated in 1928, it took many years for Raman spectroscopy to become a routine analytical tool. The main reason for this lack of utility is the weak nature of the Raman effect, only a very small fraction of light (about 1 photon in 1 million) impinging upon a sample is scattered in this way. Thus, intense monochromatic light is required to ensure that sufficient scattered light is produced to allow detection. Routine use only became feasible with significant instrumental developments such as development of the laser in 1958 by Schawlow and Townes.14 The use of laser illumination allowed stimulation of the Raman effect at detectable levels routinely, resulting in the first laser Raman spectrum of a protein in 1970.15 Development of a method for microscopic sampling was crucial to the emergence of Raman spectroscopy in the biological arena16 in 1975. However the technique of Raman microscopic imaging really made major strides with the integration of the charge coupled device (invented in the 1970s) into Raman imaging systems, allowing detection of the extremely weak Raman effect with high sensitivity. While Raman techniques were advancing, infrared techniques were also evolving. Studies by Elliot and Ambrose17 in 1950 demonstrated that protein conformation could be studied by IR spectroscopy. Around this time Blout and Mellors18 and Woernley19 were beginning to analyse human and animal tissues to determine if discrete spectroscopic signatures could be identified from various tissues. These studies met with limited success due to a combination of the relatively unsophisticated instrumentation available, little knowledge of spectroscopic properties of biological molecules and the complexity of the samples under investigation. However, at about the same time as Ambrose, Elliot, Blout and Woernley were conducting their studies, a new piece of instrumentation that would revolutionise the application of IR spectroscopy in biology was demonstrated. Barer20 and colleagues demonstrated that it was possible to couple a reflecting microscope to an infrared spectrometer and obtain infrared spectra of small particles (crystals and fibres). The applications of this new technique in biology are obvious: studies of tissues with very high spatial resolution. However, despite this instrumental advance, studies on tissues were essentially abandoned in favour of studies of the isolated building blocks of tissues. Of course it was not just optical spectroscopic and imaging techniques that were undergoing dramatic advances. New techniques based upon nuclear magnetic resonance were also developed during this time period. In 1946 Felix Bloch et al.21 and Edward Purcell et al.22 independently demonstrated that when certain nuclei were placed in a magnetic field they absorbed energy in the radiofrequency range of the electromagnetic spectrum, and re-emitted this energy when the nuclei Faraday Discuss., 2004, 126, 1–18
3
returned to their original state. This discovery laid the foundations for the development of the technique of nuclear magnetic resonance (NMR) spectroscopy, which rapidly became an important analytical method in biology, making enormous contributions in the field of protein structure. Twenty-five years later Raymond Damadian demonstrated that the nuclear relaxation times of tissues and tumours differed, suggesting that NMR techniques could be used for tissue characterisation. In 1973 Paul Lauterbur presented images of test tubes of water obtained using magnetic resonance effects, results that would pave the way for the development of magnetic resonance imaging (MRI) as a medical diagnostic modality and revolutionise medicine.23 Two years later Richard Ernst proposed magnetic resonance imaging using phase and frequency encoding and the Fourier transform,24 an approach that forms the basis of current MRI techniques. The field developed rapidly, and by 1987 echo-planar imaging was being used to perform real-time imaging of the cardiac cycle. The above brief (and by no means comprehensive) walk through the evolution of biological/ medical imaging and spectroscopy highlights a number of important issues. Firstly, the development of any imaging or spectroscopic technique requires input from a large number of disciplines and an even larger number of laboratories. For example to develop the infrared microscope in it’s current form required at a minimum contributions from Herschel (discovery of IR light), Fourier (the mathematics required to process interferograms), Melloni (constructed the first IR spectrometer), Michelson (developed interferometry), Barer (the first IR microscope) and Schawlow and Townes (inventors of the laser). Equally importantly, this example show the long lead times that are required to develop and establish new techniques, in this case 200 years. Of course occasionally so-called transformational technologies, of which MRI is a clear example, have a much-compressed time frame of development and implementation. In the case of MRI, a new technique required less than 50 years from discovery of the underlying phenomena to full and widespread clinical acceptance. The rapid development and acceptance of MRI as a tool of profound clinical and biological importance raises a provocative question: why have optical spectroscopy and imaging not been able to match this success despite a much longer (and some may say richer) history? The remainder of this article will outline some of the successes of optical spectroscopy and imaging in medicine and suggest some reasons why these powerful tools have yet to make a clinical impact
The present Before outlining the success of optical spectroscopy and imaging it is perhaps useful to summarise the advantages of spectroscopic and imaging approaches, and to highlight the strengths and weakness of the major spectroscopic and imaging tools that may be considered for biological and clinical use. Spectroscopy vs imaging Spectroscopy. For the purposes of this article spectroscopy may be broadly defined as the study of the absorption characteristics of materials following interrogation with a broadband source of electromagnetic radiation. Spectroscopic techniques that have found a place in biology and medicine include NMR, infrared, Raman, UV/Vis circular dichroism and fluorescence spectroscopy. With respect to biological and clinical situations, spectroscopy may be performed in three ways: 1 Ex vivo spectroscopy. Ex vivo spectroscopy can provide information on individual materials or assemblies. The major advantages to this approach are that systems may be easily manipulated (temperature, pH etc) and that materials may be studied in isolation, without spectroscopic interferences. However, the physiological relevance of information obtained in such experiments is always questionable. Ex vivo spectroscopy is mainly used in biology to study the structure and dynamics of proteins. NMR spectroscopy is undoubtedly the technique of choice for soluble protein analysis, providing 3-D structure at atomic resolution. However, at present NMR techniques are limited to relatively small proteins (although the mass that can be studied is constantly increasing) and cannot be applied to membrane proteins due to the low mobilities of these proteins. Furthermore, NMR 4
Faraday Discuss., 2004, 126, 1–18
spectrometers may be prohibitively expensive for many labs ($400 K–1 M). In contrast IR and Raman spectroscopy are excellent tools for studying all types of proteins, including membrane proteins, although the information obtained is limited to secondary structure. In addition to studying static structures, both infrared and Raman can be used to study protein dynamics, through reaction induced difference spectroscopy and resonance Raman spectroscopy respectively.25,26 Furthermore, IR and Raman spectroscopies can be used to study lipids, nucleic acids, carbohydrates etc., and have been used extensively to study complex mixtures of these materials.27 UV/Vis circular dichroism (CD) spectroscopy has also been used to study proteins. As with IR and Raman techniques, CD spectroscopy only provides information on 2 structure of proteins, however it has the advantage that it requires concentrations approximately two orders of magnitude lower than IR and Raman techniques. Light scattering effects limit the usefulness of CD spectroscopy to the study of membrane proteins, unless they have been solubilised with the use of detergents. The range of biological materials that can be studied with CD spectroscopy is limited, as this technique is only useful for chiral molecules. Fluorescence spectroscopy is generally the most limited of the optical spectroscopic techniques used for ex vivo analysis, and can be technically challenging. In general, fluorescence techniques require the addition of exogenous probes to allow information to be extracted. Typically this information relates to properties (hydrophobicity etc) of specific regions or domains within proteins or provides information on protein dynamics (for example through techniques such as fluorescence resonance energy transfer). 2 In situ spectroscopy. In situ spectroscopy can provide information on complex systems (cells and tissues), and allows studies of materials in a more natural environment. However, the spectra obtained tend to be very complex with significant spectral overlap from many materials, making assignment to individual species difficult. Spectra may in fact be so complex that little if any direct chemical information can be obtained. In addition, spatial information (a key factor in clinical spectroscopy) is often not obtained. Many of the tools used to study biological materials in isolation can be applied to studies in situ. For example 1H NMR spectroscopy provides information on low molecular weight metabolites such as lactate, choline, amino acids and simple sugars such as fucose, and is useful for studying tissue biopsies and biological fluids.28 Unfortunately, the information obtained is a spatial average, and no information on larger macromolecules (membrane lipids, structural proteins, nucleic acids) is obtained. In contrast, IR and Raman techniques can be used to provide either a spatial average or to assess microscopic regions of tissues. These techniques are excellent tools for analysis of structural proteins, membrane lipids and nucleic acids in situ, with a spatial resolutions of 25 mm easily obtained for infrared techniques (6 mm with a focal plane array, 3 mm with a synchrotron) and 1 mm possible with Raman techniques.29 3 In vivo spectroscopy. In vivo spectroscopy allows information to be obtained from intact animals, and provides information on all spectroscopically active materials present at the measurement site, weighted to concentration. It provides true physiological information, but again spectra are complex and assignment of absorption features may be difficult. In addition problems with light scattering can be significant. Again, most of the techniques described above can be used in vivo. Fluorescence spectroscopy/endoscopy has been used to analyse the endogenous fluorescence of tissues (for example NADH, tyrosine and tryptophan fluorescence) of a variety of tissues in vivo, with limited success to date.30 This limited success is largely due to the low intensity exogenous fluorescence from most tissues, coupled to the low specificity of changes in fluorescence for particular disease states. Near infrared and Raman spectroscopy provide more specificity and more chemical information, and are thus potentially more useful as in vivo diagnostic tools. Potential diagnostic utility is further enhanced by the favourable light transport characteristics of tissue in this spectral range. Imaging. In contrast to spectroscopic techniques, imaging techniques are based upon an assessment of the interaction of light with a particular chromophore or type of chromophore (rather than all chromophores that may be spectroscopically active) and assessment of the distribution of the chromophore in 2 or 3 dimensions. As discussed above, the most valuable of the current imaging modalities is undoubtedly MRI. Diagnostic MRI relies very heavily upon proton imaging (T1 , T2 , DW (diffusion weighted), contrast enhanced imaging etc), and is based upon Faraday Discuss., 2004, 126, 1–18
5
differences in relaxation rates of protons in tissue components, tissue water and bulk water. Such techniques provide medium image contrast and biochemical specificity is low, but the over-riding advantage of MRI is production of medium resolution 3 dimensional images of tissues. Optical imaging techniques that have been demonstrated include near infrared and fluorescence imaging. Clinical near infrared imaging techniques are based upon the differential absorption of near infrared light by oxy and deoxyhaemoglobin and water. Such techniques have slightly higher biochemical specificity than MRI and medium contrast. Images are restricted to medium spatial resolution images in 2 dimensions. Fluorescence imaging can be used with exquisite biochemical specificity with the appropriate choice of fluorophore (for example a labelled antibody), but again images are restricted to medium spatial resolution images in 2 dimensions Spectroscopic imaging. Clearly both spectroscopy and imaging have clinical potential. However, while spectroscopic techniques are generally high information content techniques, they provide little spatial information. In contrast imaging techniques provide spatial information but convey little biochemical information. A combination of the two approaches, spectroscopic imaging, should in principle combine the main advantages of the two techniques and provide both biochemical and spatial information. Conceptually this may be thought of as simply simultaneously interrogating samples with multiple wavelengths and simultaneously assessing the distribution of multiple materials. In magnetic resonance spectroscopic imaging the 3 dimensional distribution of multiple proton-containing species (lactate, amino acids, fucose) or multiple phosphorouscontaining species (ATP, ADP, AMP, Pi, PCr, etc) can be imaged, providing images with very high metabolic specificity, medium contrast and medium spatial resolution. Optical spectroscopic imaging may be used to simultaneously assess the distribution of multiple endogenous chromophores such as NADH, oxy and deoxyhaemoglobin and water etc.,31 or to monitor exogenous chromophores such as labelled antibodies, green fluorescent protein and luciferase. Optical techniques can thus provide haemodynamic and metabolic specificity with medium contrast. Some successes The development of optical spectroscopy and imaging is a broad field, and it is not the intention of this article to describe in depth the advances that have been made in these areas. Rather the main aim is to highlight the nature of the information that may be obtained and some of the pitfalls that must be overcome. This article will therefore concentrate on one specific example; work conducted in our laboratories aimed at developing a rapid diagnostic test for skin cancer will be outlined. These studies began with ex vivo analysis of a range of tissues to allow an understanding of the spectroscopic properties of tissues to be developed, and advanced to in vivo spectroscopic studies on volunteers and preliminary imaging studies on rodents. This example therefore provides an overview of many of the difficulties and potential solutions encountered using a variety of techniques. Why skin cancer?. Skin cancer is the most common cancer in most countries. In Canada in 1992 alone 50 300 new cases of skin cancer were reported, compared to 19 300 cases of lung cancer, 16 200 cases of colorectal cancer and 15 700 cases of breast cancer. Skin cancer is thus as prevalent as the next three leading types of cancer combined. The incidence continues to rise, with 64 200 new cases in 1997, an increase of 14 000 cases annually in five years. Fortunately, the malignancy is readily accessible and can be removed in a minor surgical procedure. In fact, if caught early enough skin cancer has a cure rate of 100%. Unfortunately, early diagnosis of skin cancer remains difficult. Skin cancer is not one disease but a class of related conditions, many of which appear similar when inspected visually. Diagnosis based upon visual inspection is therefore subjective. To understand this subjectivity, consider an abnormal skin growth. The growth may be pigmented or non-pigmented. If it is non-pigmented and malignant, then it is probably basal cell carcinoma or squamous cell carcinoma. However, the clinical course of these two forms of skin cancer is very different. Basal cell carcinoma spreads laterally across the surface of the skin, without penetrating the deeper layers of the skin. Thus, although disfiguring, basal cell carcinoma rarely metastasises and is rarely fatal. However, squamous cell carcinoma metastasises readily, and is often fatal. It is therefore important to be able to distinguish between the two types of skin cancer. Visually, discrimination between these two types of cancer is difficult. 6
Faraday Discuss., 2004, 126, 1–18
If the growth on the skin is pigmented it may be one of the many types of pigmented lesion commonly referred to as moles and technically referred to as naevi. Types of naevi include congenital naevi, acquired naevi, blue naevi and dysplastic naevi. Alternatively, malignant melanoma may be suspected. Initially, malignant melanoma is suspected based upon the unusual appearance or behaviour of naevi. The so-called ‘‘ ABCD ’’ warning signs guide assessment of the likelihood of malignant melanoma: A: Asymmetry in shape; Appearance of a new mole. B: Borders that are notched and/or, irregular; Bleeding,.C: Colour is variable or contains blue, grey, white, pink or red; Change in shape, size or colour; Concern of patient. D: Diameter exceeding 6 mm. Diagnosis of malignant melanoma based upon such an approach is obviously subjective. Complicating matters further, basal cell carcinomas may be pigmented, and appear similar to malignant melanoma. Even if a diagnosis of malignant melanoma or pigmented basal cell carcinoma is ruled out, abnormal growth of naevi is clinically relevant as some types of naevi carry a definite risk of progression to malignant melanoma. Upon initial presentation the clinical problem is therefore difficult, requiring classification of an abnormal skin growth as squamous cell carcinoma, pigmented or non-pigmented basal cell carcinoma, malignant melanoma, or an acquired, congenital of dysplastic naevus. Ultimately some additional information concerning grade (i.e. degree of aggressiveness, depth of penetration etc.) is required, which is even more difficult to ascertain visually. Once an initial diagnosis has been made a more definite classification requires excision of tissue and histological analysis. It is only at this point that the diagnosis becomes definitive. A non-invasive technique that could remove the subjectivity in the characterisation of skin lesions and could be performed in the physicians/ dermatologists office would be of practical benefit. We have taken a multi-step approach to skin cancer diagnosis. The first step was to determine the optical technique and spectral region of choice. Based upon skin transmission characteristics we chose the near infrared spectral region as that most likely to allow in vivo diagnosis. The second step was to gain a broad understanding of the fundamental spectroscopic properties of tissues, and to determine whether or not information concerning skin cancer could be extracted from biopsies. We chose to study biopsies using mid infrared spectroscopy rather than near infrared spectroscopy, as the information content of mid infrared spectra is higher than (although related to) that of near infrared spectra. The rationale for this was simple: if we could not understand the optical properties of biopsies and obtain clinically useful information from mid infrared spectra, it would be highly unlikely that we could obtain such information in vivo from near infrared spectra. The third step was to transfer our knowledge of ex vivo specimens to in vivo lesions, and develop an in vivo spectroscopic technique for classifying skin lesions. The final stage was to develop an imaging modality for skin cancer diagnosis. Step 1. Ex vivo characterisation of biopsies. Our studies began with attempts to understand the infrared spectroscopic properties of tissues. The wavelengths of infrared light absorbed by a sample (called its infrared spectrum) provide information relating to the composition and structure of the sample. In other words, the infrared spectrum is a molecular fingerprint. As the molecular fingerprint of tissue changes during a disease process, differences should be seen in the infrared spectra of normal and diseased tissues. We therefore undertook a major study investigating the spectroscopic properties of a wide range of tissues. Fig. 1 shows spectra of breast and liver tissue obtained using macroscopic sampling techniques. Spectra of malignant breast tissue were acquired by placing 10 mm sections of breast tissue between CaF2 windows without allowing the tissue to air dry. Spectra of liver tissue were obtained from 10 mm section of liver tissue that were placed onto CaF2 windows and allowed to air dry. Obvious differences are apparent between spectra of breast and liver, and between the different breast and liver samples. It may be surmised that the spectral differences between the three sections of breast tissue are in some way related progression of the disease. However, each section was characterised by a pathologist as arising from invasive ductal carcinoma of the breast at the same stage of development. Spectral differences are unlikely therefore to reflect differences in disease progression. In fact, these differences are simply reflective of the normal variations that are to be expected in animal tissues. The upper trace is dominated by absorptions from lipid-like species, as indicated by prominent absorptions from CH2 , C=O and C–O–C groups and lacks an amide I absorption. As the amide I absorption arises from cellular and structural proteins, this section does not contain Faraday Discuss., 2004, 126, 1–18
7
Fig. 1 Mid infrared spectra of macroscopic sections of breast (a–c) tumours and liver tissue (d–e)
connective tissue of epithelial/tumour cells. In fact, this tissue section is dominated by adipose tissue. In contrast, the section giving rise to spectrum (b) is dominated by absorptions from protein, specifically collagen. This section is therefore mainly composed of connective tissue. The section giving rise to spectrum (c) is also dominated by protein, but does not exhibit absorptions from collagen. This section is therefore dominated by absorptions from epithelial/tumour cells. This brief discussion highlights an important drawback to the analysis of tissue using spectroscopic techniques: the lack of spatial information that may be obtained. Each of the tissues giving rise to spectra in Fig. 1 may have contained epithelial cells, but information on the cells can only be obtained from spectrum (c). This has important implications for those spectroscopists trying to develop spectroscopic techniques for diagnosis based upon macroscopic measurements. Consider the schematics of tissue sections presented in Fig. 2. In Fig. 2a we have two sections of tissue with the same number of adipose cells, but different numbers of other cells. One section has four epithelial cells, the other has two malignant cells. Now if we assume for the sake of argument that the tumour cells have twice the DNA content of normal epithelial cells, then spectroscopically we will
Fig. 2 Schematic representation of tissue heterogeneity. See text for description
8
Faraday Discuss., 2004, 126, 1–18
detect the same amount of DNA in both sections. If we try and develop diagnostic tools based simply upon quantification of DNA in tissue sections (or from in vivo spectra), the tumour in section 2 will not be detected. Fig. 2b further illustrates the importance of spatial information. This figure shows four sections which have epithelial cells randomly distributed, arranged as a focal group surrounded by fat, in organised striations or as a discrete structure within the tissue. A macroscopic assessment of these sections will not allow discrimination between the four tissue types, even though they are histologically distinct. Spectra alone do not provide sufficient information for characterisation; spatial information is required. Step 2. Ex vivo characterisation of skin biopsies by infrared microscopy. IR microscopy is an essential tool for tissue studies as it allows both spectral and spatial information to be obtained.29 An IR microscope focuses IR light onto a small area of a sample, allowing high quality spectra to be obtained from small samples. Coupling an IR microscope to a high precision computer controlled stage allows the automated collection of IR spectra at each position in a two-dimensional grid defined over a tissue section. Information collected at each point (pixel) is combined to produce an IR spectroscopic map of the tissue section. Large spectral maps are routinely collected containing thousands of spectra per map and enormous amounts of both spatial and spectral information. We have used this approach to characterise skin tissue. A large number of skin biopsies were analysed by acquiring spectra from small regions (25 mm) of histologically identifiable structures within the sections. In this way we established a large database of spectra from the major histological structures in normal skin and the major types of skin cancer. Representative normalised spectra of epidermis, dermis, hair follicles and a basal cell carcinoma are shown in Fig. 3. Distinct differences are noted between major histological structures, even to the naked eye. It is clearly possible to distinguish the dermis, based upon the presence of unique absorptions from collagen at 1204,1240, 1280 and 1635 cm1. Hair follicle and sebaceous glands can be easily distinguished by an elevated lipid content (1740 cm1). Spectra of basal cell carcinoma and epithelium are more difficult to distinguish. Nevertheless, if normalised spectra of the two tissue type are superimposed (Fig. 4a), it becomes apparent that there is a significant difference in the apparent intensity of the absorptions arising from nucleic acids. This may be taken to be indicative of an increase in the nucleic acid content of the tumour cells in the basal cell carcinoma. However, if the spectra of the tumour and normal epithelium are overlaid without normalisation (i.e. raw data is presented,
Fig. 3 Mid infrared microscopic spectra of epidermis (a), basal cell carcinoma (b), hair follicle (c) and dermis (d). Inset: regions of tissue from which spectra were acquired.
Faraday Discuss., 2004, 126, 1–18
9
Fig. 4 Normalised (a) and raw spectra (b) of skin epithelium and basal cell carcinoma.
Fig. 4b), then the data suggest a different conclusion, in this case that there is a significant differences in protein content between the tumour and epithelial cells. Clearly the mode of presentation of the data can affect the conclusions drawn. Which mode of presentation, and which interpretation, if either, is correct? Data is often normalised (for example to the amide I intensity) to ensure that differences in tissue thickness are not mistaken for differences in tissue composition. However, normalisation precludes the possibility of quantitatively assessing absolute variations in tissue composition. In reality, this means that both raw and normalised data should be analysed, and that data from the entire spectral map should be assessed rather than just a few spectra. In this example, examination of a large number of spectra suggest that the major biochemical differences between basal cell carcinoma cells and epithelial cells is a reduced protein content in tumour cells, probably as a result of nuclear enlargement. Problems with spectral interpretation and data presentation illustrate the need for less subjective approaches to tissue characterisation, which led us to apply a number of pattern recognition techniques to infrared spectra of skin components. The simplest approach to pattern recognition analysis of spectroscopic data is to use techniques such as cluster analysis. Cluster analysis techniques compare spectra using some metric of similarity, and partition spectra into groups based upon this measure of similarity. Membership of groups is varied in an iterative process, for example to minimise the variance within groups while maximising the separation between groups, thus optimising the number of clusters formed. For spectroscopic maps, pixels belonging to the same cluster are displayed in the same colour, producing a coloured image showing cluster memberships. The results of a cluster analysis applied to a spectroscopic map from a section of basal cell carcinoma (Fig. 5a) are shown in Fig. 5b. Clusters corresponding to the epidermis, dermis, hair follicles and tumour are easily observed.32 Cluster analysis obviously shows groups of pixels that may allow some correlation with sample histology. However, this introduces an element of subjectivity. Other pattern recognition techniques make use of the fact that we often are in possession of information regarding clinical properties of a sample. For example in the case of skin lesions we have a clinical diagnosis (obtained by examination of a biopsy by a dermatopathologist). This information may be used to train techniques such as linear discriminant analysis (LDA) to recognise the particular combinations of peak frequencies, absorption band widths, relative intensities etc. in spectra that are characteristic of spectra from a particular clinical grouping. The trained LDA algorithm can then be applied to spectra of unknown 10
Faraday Discuss., 2004, 126, 1–18
Fig. 5 Photomicrograph of a section from a basal cell carcinoma (a) and the results of fuzzy C-mean cluster analysis of spectra acquired by mapping the section using infrared microscopy (b). See text for more details.
origin, and the unknown spectra partitioned into one of these clinical groupings based upon the spectral pattern found.33 The results of such an analysis are shown in Table 1. The results in Table 1 demonstrate that mid infrared spectroscopic analysis of skin lesions coupled to a supervised pattern recognition analysis methodology can be used to characterise Table 1 Non-subjective analysis (IR predictions are in rows, pathological classification in columns, correct classifications are in bold) (a) Classification of tissue types BCC Basal cell carcinoma Epidermis Follicle
157 0 0
(b) Classification by tumour types SCC Squamous cell carcinoma Basal Cell carcinoma Melanocytic
0 68 2
BCC
Follicle
Accuracy
Specificity
Sensitivity
0 0 51
100 100 96.2
100 98.1 100
100 96.4 100
Melanocytic
Accuracy
Specificity
Sensitivity
48
2
0
96
95.5
91.4
5
151
1
96.2
95.1
90.7
2
2
30
88.2
99.7
99.9
(c) Subdivision of BCC lesions Nodular Nodular Infiltrating Superficial
Epidermis
53 8 10
Infiltrating 6 9 0
Superficial 1 1 27
Accuracy 88.3 50 73
Specificity 64.3 95.0 96.4
Faraday Discuss., 2004, 126, 1–18
Sensitivity 55.3 83.3 91
11
pathological lesions. Not only can the technique discriminate between basal cell carcinoma and normal tissue components, but it can also be used to distinguish between various types of skin lesions (basal cell carcinoma, squamous cell carcinoma and malignant melanoma, Table 1b). Furthermore, pattern recognition analysis of IR spectra may also be used to characterise various types of the same class of lesion, for example nodular and infiltrative basal cell carcinoma (albeit with reduced accuracy) as shown in Table 1c. Importantly for clinical utility, pattern recognition of infrared microscopic data can be presented in a ‘‘ physician friendly ’’ manner. In essence, methods have been developed to present the data in a way familiar to physicians that resembles the way data is presented from standard histological analysis. In brief, techniques such as LDA return a coefficient (between 0 and 1) that indicates the probability that the spectrum at any pixel arises from a particular tissue type. If these values are plotted as a function of position within the spectroscopic grid, then a map showing the probability of tissue belonging to a particular class may be produced. This can be converted into an eight bit grey scale image with values ranging from 0 to 255, and so-called ‘‘ likelihood maps ’’ produced for each class of tissue (Fig. 6). In fact, three different grey scale maps may be converted to red, green and blue scale maps and combined to produce a 24 bit RGB (red, green, blue) map showing the likelihood of tissue at any pixel belonging to three classes of tissue.34 These new non-subjective but visually appealing processing tools coupled to the potential for automation of infrared microscope techniques, point to a potential role for infrared microscopic analysis of tissues in the pathology laboratory. Step 3. Non-invasive diagnosis of skin cancer using near infrared/visible spectroscopy. We have shown that biopsies from basal cell carcinoma (BCC), squamous cell carcinoma (SCC) and melanocytic tumors have distinct mid-IR signatures when compared to normal skin. However, the diagnostic potential of mid-IR spectroscopy in-vivo is limited, since complete absorption of mid-IR light results with samples greater than 10–15 mm in thickness. In contrast, near-IR light is scattered to a much greater extent than it is absorbed, making tissues relatively transparent to near-IR light, thus allowing the examination of much larger volumes of tissue and the potential for in-vivo studies.
Fig. 6 Visual representation of the results of a linear discriminant analysis of spectra from the section shown in Fig. 5a. The LDA coefficient for each class (value between 0 and 1 representing the likelihood that the pixel corresponds to that tissue class) is converted to an 8 bit number. The 8 bit coefficient for tumour, dermis and epidermis for each pixel are then plotted as a function of position within the tissue as a greyscale image (black ¼ 0, white ¼ 255). The three greyscale images may be converted to red, green and blue 8 bit images and combined to produce a 24 bit RGB (red, green, blue) image showing the likelihood of spectra belonging to three classes simultaneously.
12
Faraday Discuss., 2004, 126, 1–18
The near-IR region is often sub-divided into the short (600–1000 nm) and long (1000–2500 nm) near-IR wavelengths, based upon the technology required to analyze light in these wavelength regions. At shorter near-IR wavelengths, the heme proteins (oxy- and deoxyhemoglobin and myoglobin) and cytochromes dominate the spectra, and their absorptions are indicative of regional blood flow and oxygen consumption. Long wavelength near-IR absorptions arise from overtones and combination bands of the molecular vibrations of C–H, N–H and O–H groups. The absorption of near-IR light therefore provides information concerning tissue composition (i.e. lipids, proteins) and oxygen delivery and utilization, and we believe can be used to diagnose skin cancer. In-vivo visible and near-infrared spectra (400–2500 nm) of skin neoplasms (actinic keratoses, basal cell carcinoma, banal common acquired melanocytic nevi, dysplastic melanocytic nevi, actinic lentigines and seborrheic keratoses) were collected using a dispersive near infrared spectrometer equipped with a bifurcated quartz fibre optic. Near infrared light was brought from the instrument sources through one branch of the fibre to the tip of the fibre probe that was placed lightly against the skin. Diffusely reflected light was collected at the fibre tip by a randomly arranged array of collection fibres and transmitted to the detectors for computation of spectra. The mean of all spectra collected for the skin lesions analysed is shown in Fig. 7a. The shaded area indicates the spectral variance. Spectra of the different lesions could not be readily distinguished by visual analysis. However, if control spectra (acquired from an adjacent area of normal skin of the same patient) were subtracted from each lesion spectrum, residual features were produced which indicated differences in the spectra, and so chemistry, of the various skin lesions and healthy skin tissue (Fig. 7b). The fact that differences exist between spectra of healthy and abnormal skin tissue strongly suggests that near infrared/visible spectroscopy may be used to diagnose skin cancer.35 Of course to be clinically useful the technique should not require examination of a spectrum (or difference spectrum) by a dermatologist, which would introduce an element of subjectivity. Ideally, the diagnosis should be made using some type of statistically-based, non-subjective procedure. We have applied the same LDA strategy described above to difference spectra from benign and premalignant/malignant lesion groups. The results are shown in Table 2. Some of the more difficult visual diagnoses were successfully distinguished using this methodology. In fact, our results demonstrate that near infrared/visible spectroscopy could form the basis of a clinical method to diagnose skin lesions. It is rapid (i.e. acquisition time of minutes), simple to perform and noninvasive. Measurements are accurate and reproducible; the only outlier spectra in this study occurred from large patient movements, a problem that can be removed, or at least reduced, by shorter data acquisition times. Collection of spectra causes little or no patient discomfort, does not alter the basic physiology of the skin, poses no hazard to the patient and does not interfere with any other standard clinical diagnostic practices. The test could be performed by a non-specialist and, therefore, might be a useful tool for pre-screening skin lesions. Step 4. Imaging skin. It should be quite clear that spectroscopic techniques can provide information on tissue chemistry and, when coupled to the appropriate data analysis strategy, can be used diagnostically. However, spectroscopic techniques suffer from two important drawbacks: Sampling problems. Ex vivo, in situ and in vivo spectroscopy all suffer from problems with sampling. How can an investigator be sure that the correct region of tissue (i.e. that affected by the disease being studied) is in fact being sampled? For ex vivo studies, this requires some certainty that the section being analysed actually does contain diseased tissue, and that the stage or grade of the disease in this sample matches that provided by the clinician. Simply put, are the investigator and clinician looking at the same thing? The same problem arises with in situ analysis: during mapping experiments has the investigator actually mapped a region of tissue that corresponds to the clinical diagnosis? For in vivo studies, correct sampling is dictated by positioning of fibre optic probes: is the appropriate region of the lesion being interrogated? For in situ and in vivo sampling methods, spatial resolution is also a key issue, as this has a direct impact on sensitivity and specificity. In the case of in situ experiments on excised tissue, the pixel (aperture or laser spot) size is the critical determinant of spatial resolution. For in vivo experiments, the critical determinant is the diameter of the fibre probe. If the pixel size or fibre diameter is too small, the acquired signal will be too weak to adequately process the data. On the other hand, if the fibre diameter or pixel size is too large, then the spectrum of the tissue of interest may be swamped by the spectrum of surrounding tissue. Obviously, the appropriate pixel size and fibre diameter are Faraday Discuss., 2004, 126, 1–18
13
Fig. 7 Mean near infrared spectra of a range of skin lesions (solid line) and the variance (shaded area) in spectra.
determined by the sample under investigation, and there is little possibility that ‘‘ one size fits all ’’. In practice this means that each sample studied essentially requires unique experimental geometry. Data presentation. The second major drawback to spectroscopic technologies is the nature of the data obtained. Very few physicians will be comfortable with spectra, or ‘‘ squiggles ’’ as many physicians perceive them to be. Traditionally, physicians are used to data presented as images and this is unlikely to change. In a few instances (for example spectroscopic maps) data can readily be converted to something that approaches the type of image that physicians are used to, but in general the output of a spectroscopic technique is likely to be either a ‘‘ squiggle ’’ or a number derived from this squiggle. The question then is can physicians be trained to accept this form of data? A picture is worth a thousand words. We are all aware of the old adage ‘‘ a picture is worth a thousand words ’’ and nowhere is this truer than in the field of medicine. Spectroscopic techniques may prove to be extremely powerful diagnostic tools, but if the information cannot be presented to a physician in a format that they are comfortable with the technology will not find a use in clinical 14
Faraday Discuss., 2004, 126, 1–18
Table 2 Results of linear discriminant analysis of near infrared spectra compared to the accuracy of a practising dermatologist
Dysplastic nevi vs. banal nevi Dysplastic nevi vs. actinic lentigines Actinic keratoses vs. actinic lentigines Actinic keratoses vs. seborrheic keratoses BCC vs. seborrheic keratoses BCC vs. banal nevi Dysplastic nevi vs. seborrheic keratoses
Accuracy by LDA
Clinical accuracy
97.7 92.0 89.9 84.3 81. 81.5 72.4
89.6 100 78.0 94.4 94.8 91.1 100
medicine. In reality, physicians want non-invasive imaging technologies. With this in mind we have begun the development of a number of optical imaging and spectroscopic imaging techniques for the diagnosis of a variety of conditions. The first technique that we have developed is immunofluorescence imaging as a tool for skin cancer diagnosis.36 Immunofluorescence combines the specificity of immunological technologies with the sensitivity of fluorescence measurements. Briefly, the technique consists of injecting a fluorescently labelled antibody into a tumour-bearing subject, stimulating fluorescence and localizing the antibody (and so the tumour) by fluorescence imaging. We have tested this technique in an animal model of melanoma. Human melanoma cells were injected subcutaneously on the shoulder of immune compromised mice (which would not reject the human cells). After 7 days to allow growth of the tumour cells, an anti-cancer antibody labelled with the fluorophores Cy5 was injected via the tail vein. Imaging was performed using a charge coupled device camera equipped with the appropriate filters at various times post injection. Fluorescence could be seen at the tumour site and in the kidneys 15 minutes post injection, peaking at 2 hours post injection (Fig. 8). Immunofluorescence can be used to localise malignant cells in sub-surface structures and superficial tissues such as skin with very high specificity. In essence the degree of specificity is determined by the specificity of the antibody used, while the depth that can be probed is a complex function of
Fig. 8 Immunofluroescnce image of a tumour-bearing mouse following injection of a fluorescently labelled antibody.
Faraday Discuss., 2004, 126, 1–18
15
illumination wavelength, emission wavelength, tissue optical properties, the quantum yield of the dye and rate and extent of antibody uptake by the tumour. Ideally, a small antibody fragment (to optimise diffusion properties) labelled with a dye with high quantum efficiency with absorption and emission maxima in the region 650–730 nm should be used. The main flaw with such an imaging approach is that while it can provide very high sensitivity and specificity, the specificity may in fact be too great. For example to distinguish between a basal cell carcinoma and a squamous cell carcinoma would require two antibodies. If staging is required, then antibodies would be required that are specific for each stage of tumour that is of interest. In practice this means injecting a unique contrast agent for each type of tumour and stage of tumour development. Of course the requirement for an injectable agent to provide diagnostic information is also a drawback, as this increases the likelihood of side effects of the diagnostic tool and imposes further regulatory hurdles. The ideal imaging tool would be capable of distinguishing between a wide number of conditions without the need for injectables and without the need for ‘‘ tuning ’’ for different disease states. This can be achieved by spectroscopic imaging. Spectroscopic imaging avoids the problems of using only a single chromophore or contrast mechanism by analysing the optical properties of tissues at multiple wavelengths, with each wavelength providing information on different chromophores. Spectroscopic imaging may also be thought of as multi-contrast imaging (a more descriptive term than ‘‘ hyperspectral imaging ’’, a term that seems to be gaining general acceptance). We have developed such imaging approaches using high sensitivity, near infrared enhanced CCD cameras equipped with liquid crystal tuneable filters. The main application of this system to date has been imaging of comprised skin, for example in areas such as wound healing and burn assessment.37 Computer controlled wavelength selection using a liquid crystal tuneable filter allows sequential collection of images at pre-programmed wavelengths, with no user intervention, in a matter of seconds. The data obtained are predominantly related to haemodynamics and tissue oxygenation, important parameters that allow tissue viability to be assessed. This imaging system has been used to demonstrate the feasibility of determining the viability of surgical grafts and to determine the extent and severity of burns. This latter application is of particular interest as it often takes a burn surgeon a significant period of time to be able to determine the nature of a burn. Burns are not static injuries, and evolve over time. It may take a number of days for the true extent of tissue damage to be apparent. Near infrared oxygenation imaging can rapidly distinguish areas of tissues that will become non-viable within a matter of seconds, an important clinical advantage.
The future There is no doubt that optical technologies can provide diagnostic information. A variety of optical methods (including infrared and Raman techniques) have been shown to be able to provide valuable information from biopsy samples. In vivo optical methods have been published that allow non-subjective diagnosis of diseases such as arthritis and a variety of forms of cancer. However, despite the optimism encountered in the literature and at international meetings, these recent applications of optical spectroscopy have yet to make an impact on clinical practice. In part this may be due to a lack of understanding of the potential power of the technique by clinicians, and also due to the economic barrier to entering this field of research. However, these barriers could surely be overcome if the technique could be proven to be clinically useful. Perhaps this is where the true failure lies: biological optical spectroscopists have yet to make a convincing case that optical techniques can be clinically relevant, that they are more than tools that may occasionally address an interesting point of academic interest to a pathologist. If this is indeed the case, then we must ask ourselves why we have failed to demonstrate clinical utility. The answer may be simple but unpleasant: perhaps biological optical spectroscopy is not advancing. Biological optical spectroscopy and imaging reached a point of promise in the 1990s but appear now to be stagnant fields of research. Much of the research being presented orally and in the literature may best be summarised as ‘‘ more of the same ’’ with little truly new information being presented. Perhaps even this description is overoptimistic: at times much current work seems merely derivative. The mode of thought appears to be ‘‘ if this works on tissue A then the same should work on tissue B ’’. Such an approach would be commendable if the thrust of the work undertaken was to validate previous work and then allow extension of the work. However, it often 16
Faraday Discuss., 2004, 126, 1–18
appears to be an approach that is taken because researchers are bereft of new ideas yet need to publish or present. In short, at this point a lack of imagination would seem to be a major problem. The quality of many studies may also be compromised by the subtle manner in which pressure to ‘‘ patent or perish (commercially) ’’ has replaced the pressure to ‘‘ publish or perish (academically) ’’. While the ‘‘ publish or perish ’’ attitude had the merit of encouraging peer review, dissemination of information and discussion, the ‘‘patent or perish’’ culture that has slowly invaded this branch of science has seriously impeded information transfer, discouraged open debate (who wants an open debate on new technology when potential investors may be in the audience?) and resulted in a rush to publicise early (sometimes speculative, often misleading) findings in an attempt to raise funding. When these early results prove to be overly optimistic, or anticipated publications fail to appear subsequently, granting agencies, venture capitalists and clinical collaborators very soon lose interest. In addition to questions about the quality and relevance of work that is being undertaken, questions of statistical power need to be addressed. Before clinical acceptance can even be thought about large multi-centre studies with statistical power and methods to overcome problems with inter-instrument variability and instrumental drift are needed. The inability of many institutions (or lack of venture capital) to fund such studies may ultimately be the major factor preventing clinical acceptance. Such financial considerations represent a serious impediment to development of optical spectroscopy and imaging as a clinical tool. While many groups are working towards development of tools that will gain clinical acceptance, it is perhaps better to describe much of this effort as devoted to development of a platform to aid funding request to granting agencies. The addition of ‘‘ bio ’’spectroscopy and ‘‘ bio ’’imaging to proposals are seen by many as an easy way to obtain funding, and unfortunately, many granting agencies appear to hold the same view. The lack of expertise of many newcomers to the field of biospectroscopy and bioimaging (and perhaps a lack of serious intent) results in poor progress, and ultimately will lead to a lack of interest on the part of granting agencies in the future. Financial considerations will ultimately determine the extent to which any new optical technique achieves clinical acceptance. Crass as it may seem, unfortunately in the real world the prime factor motivating the development of new diagnostic technologies, clinical acceptance, requires commercialisation. In other words, someone has to sell a user-friendly product. This requires at least two things: regulatory approval and profit. Obtaining regulatory approval is an extremely expensive and time-consuming process. To make a commitment of this magnitude requires a high degree of certainty that the technology will eventually make money for investors, not that it will save lives or save health care system money. Issues such as cost savings and improved quality of life are of interest to investors only to the extent that they can be used as marketing aids. In reality the issue is ‘‘ can anyone make money by selling a diagnostic tool based upon optical spectroscopy or imaging? ’’. The answer to that question will determine whether or not these techniques ultimately make a clinical impact. Of course this discussion assumes that the new diagnostic technology provides a benefit to the healthcare system. In other words, that the new technique is at least as good (preferably better) and more costs effective than existing techniques. Unfortunately, to date there is no indication that any of the new optical techniques under development can reliably and consistently provide the diagnostic information that is required. There are no published long-term, multi-site, multi-instrument studies that validate optical technology for clinical use. In other words, proof of clinical utility has yet to be convincingly presented for any of the new optical techniques. If we accept that the problems described above are serious limiting factors in the progression of optical diagnostics, what can be done to promote development of optical techniques as useful clinical tools? I suggest that four things are needed: (i) A core of optical spectroscopists committed to development and expansion of the optical diagnostics needs to emerge. (ii) Open debate and large, multi-centre studies need to be encouraged. (iii) Investigators must tackle clinically important questions that cannot be addressed easily with other techniques. (iv) Finally, and most importantly, investigators must realise that many optical techniques, at least as currently practised, are unlikely to become clinical tools for reasons of calibration transfer, complexity, cost and automation. Investigators must also realise that the fact that a technique can Faraday Discuss., 2004, 126, 1–18
17
provide diagnostic information does not mean that the technique will gain clinical acceptance. Only a small fraction of the many proposed applications of optical spectroscopy and imaging are likely to become clinical realities, and the factors that determine which techniques succeed are likely to be economic rather than scientific. This summary paints a rather bleak, and some may say cynical, picture of the future of optical diagnostics. It should perhaps be viewed more as a realistic assessment of where we as a community currently stand. The big question is: where are we going?
References 1 S. Hodge Jr., Thermography & Personal Injury Litigation, Wiley Law Publications, New York, 1987, p. 71. 2 R. Hooke, in Micrographia, or, Some physiological descriptions of minute bodies made by magnifying glasses with observations and inquiries thereupon, Science Heritage Ltd., Lincolnwood, Ill, 1987 (facsimile). 3 B. J. Ford, The van Leeuwenhoek Specimens, Notes & Records of the Royal Society, 1981, vol. 36, p. 37. 4 T. Schwann, Mikroskopische Untersuchungen u¨ber die Ubereinstimmung in der Struktur und dem Wachstum der Tiere und Pflanzen, G. Reimer, Berlin, 1839. 5 H. W. Waldeyer, Zeitschrift fu¨r rationelle Medicin, 1863, 20, 193. 6 C. Gram, Fortschritte der Medicin, 1884, 2, 185. 7 L. Daddi, Arch. Ital. Biol., 1896, 26, 143. 8 W. Herschel, Philos. Trans. R. Soc., 1800, 90, 255. 9 W. Herschel, Philos. Trans. R. Soc., 1800, 90, 284. 10 M. Melloni, Ann. Chim., 1836, 61, 375. 11 A. J. Collins, E. F. J. Ring, P. A. Bacon and J. D. Brookshaw, Clin.Radiol., 1976, 27, 237. 12 W. C. Ro¨ntgen, Nature, 1896, 53, 274. 13 C. V. Raman and K. S. Krishnan, Nature, 1928, 121, 501. 14 A. L. Schawlow and C. H. Townes, Phys. Rev., 1958, 112, 1940. 15 R. C. Lord and N. T. Yu, J. Mol. Biol., 1970, 50, 509. 16 M. Delhaye and P. Dhamelincourt, J. Raman Spectrosc, 1975, 3, 33. 17 A. Elliot and E. Ambrose, Nature, 1950, 165, 921. 18 E. K. Blout and R. C. Mellors, Science, 1949, 110, 137. 19 D. L. Woernley, Cancer Res., 1952, 12, 516. 20 R. Barer, A. R. H. Cole and H. W. Thompson, Nature, 1949, 163, 198. 21 F. Bloch, W. W. Hanson and M. Packard, Phys. Rev., 1946, 69, 127. 22 E. M. Purcell, H. C. Torrey and R. V. Pound, Phys. Rev., 1946, 69, 37. 23 P. G. Lauterbur, Nature, 1973, 242, 190. 24 A. Kumar, D. Welti and R. R. Ernst, J. Magn. Reson., 1975, 18, 69. 25 M. Jackson and H. H. Mantsch, in Protein–Ligand Interactions: Structure and Spectroscopy, ed. S. E. Harding and B. Z. Chowdhry, A Practical Approach, Oxford University Press, Oxford, 2001, vol. 2, ch. 6, p. 239. 26 I. K. Lendev, A. S. Karnoup, M. C. Sparrow and S. A. Asher, J. Am. Chem. Soc., 1999, 121, 8074. 27 M. Jackson and H. H. Mantsch, in Infrared Spectroscopy of Biomolecules, ed. H. H. Mantsch and D. Chapman, Wiley-Liss, Inc., Toronto, 1996, p. 311. 28 T. Bezabeh, R. L. Somorjai, I. C. P. Smith, A. E. Nikulin, B. Dolenko and C. N. Bernstein, Am. J. Gastroenterol., 2001, 96, 442. 29 M. Jackson and H. H. Mantsch, in Handbook of Vibrational Spectroscopy, ed. J. M. Chalmers and P. R. Griffiths, John Wiley & Sons, Chichester, 2002. 30 B. Ramjiawan, M. Jackson and H. H. Mantsch, in Encyclopedia of Analytical Chemistry, ed. R. A. Meyers, John Wiley & Sons Ltd., Chichester, 2000, vol. 1, pp. 5–20. 31 M. G. Sowa, L. Leonardi, A. Matas, B. Schattka, M. D. Hewko, J. R. Payette and H. H. Mantsch, in Encyclopedia of Analytical Chemistry, ed. R. A. Meyers, John Wiley & Sons Ltd., Chichester, 2000, vol. 1, pp. 251–281. 32 L. M. McIntosh, J. R. Mansfield, A. N. Crowson, H. H. Mantsch and M. Jackson, Biospectroscopy, 1999, 5, 265. 33 L. M. McIntosh, M. Jackson, H. H. Mantsch, M. F. Stranc, D. Pilavdzic and A. N. Crowson, J. Invest. Dermatol., 1999, 112, 951. 34 J. R. Mansfield, L. M. McIntosh, A. N. Crowson, H. H. Mantsch and M. Jackson, Appl. Spectrosc., 1999, 53, 1323. 35 L. M. McIntosh, R. Summers, M. Jackson, H. H. Mantsch, J. R. Mansfield, M. Howlett, A. N. Crowson and J. W. P. Toole, J. Invest. Dermatol., 2001, 116, 175. 36 B. Ramjiawan, P. Maiti, A. Aftanas, H. Kaplan, D. Fast, H. H. Mantsch and M. Jackson, Cancer, 2000, 89, 1134. 37 M. G. Sowa, L. Leonardi, J. R. Payette, J. S. Fish and H. H. Mantsch, Burns, 2001, 27, 241.
18
Faraday Discuss., 2004, 126, 1–18
