Fluorescence Spectroscopy of Human Tissues
The phenomenon of fluorescence was observed first by Stokes [61
]. Much later, Stubel recognized the diagnostic potential of tissue fluorescence [62
]. Policard [63
] observed red fluorescence when examining tumors under illumination with ultraviolet and visible light. The observed fluorescence was attributed to endogenous porphyrins in the tissue. Ghadially et al.
] reported that ulcerated squamous carcinomas exhibit a red fluorescence when exposed to ultraviolet light. They concluded that this red fluorescence maybe in part due to action of bacteria on a protoporphyrin precursor. In 1965, Lycette and Leslie [64
] suggested that fluorescence spectroscopy could be used to discriminate between normal tissues and malignant tumors. Fluorescence emission spectra were recorded from excised normal tissue and malignant tumors of the esophagus, stomach, breast and thyroid, at 330 nm excitation. All tissues fluoresced in the range from 360 to 600 nm. It was found that the fluorescence intensities of malignant tumors were less than that of normal tissue from the same patient. Subsequently, the groups of Profio and Doiron [65
], Alfano et al.
], Lohmann [67
] and Yang et al.
] did pioneering studies on in vitro
and in vivo
fluorescence spectroscopy of neoplastic and non-neoplastic, animal and human tissues.
Fluorescence spectroscopy in the ultraviolet and visible spectral regions has been developed and employed to differentiate diseased from non-diseased tissues, in vivo. The altered biochemical and morphologic state that occurs as tissue progresses from a non-diseased to diseased state, is reflected in the spectral characteristics of the measured fluorescence. This spectral information can be compared to tissue histology, the current gold standard, which indicates the absence or presence and grade of disease. Mathematical algorithms can then be developed and optimized to classify tissues into their respective histologic category, based on their spectral features. These mathematical algorithms can be implemented in software, thereby potentially enabling fast, non-invasive, automated screening and diagnosis in a clinical setting.
The following section reviews the chronological evolution of fluorescence spectroscopy for various clinical applications and highlights the current status within the context of these different clinical scenarios. Specific emphasis is given to the excitation wavelength(s) used, the types of measurements that were made, the method used to dimensionally reduce the spectral variables and the corresponding classification scheme. Furthermore, the sample size for the diseased and non-diseased populations and the corresponding sensitivity and specificity are also reported. The current gold standard for the evaluation of the sensitivity and specificity of fluorescence spectroscopy is the histologic evaluation of tissue biopsies.
Currently, fluorescence spectroscopy has been employed to detect neoplastic growth in the colon, cervix, bronchus, bladder, brain, esophagus, head and neck (oral cavity, oropharynx and larynx), skin, bile duct, breast and stomach. While breast and stomach tissues have only been evaluated in vitro, in vivo studies have been performed on the other tissue sites. This may be due to the fact that the breast and stomach are not directly accessible for fluorescence spectroscopy, like the other tissues. The excitation wavelengths that have been selected to evaluate these tissues correspond to those used to excite fluorophores in the ultraviolet and visible spectral regions (see ). Measurements in the form of fluorescence spectra, transient decay profiles and images have been acquired from these tissues. However, the colon is probably the only organ in which all three measurements have been performed. While fluorescence spectra and images have been measured from the bronchus, bladder, head and neck and skin, only fluorescence spectra have been measured from the cervix, brain and esophagus, thus far.
There are generally two steps involved in the development of a mathematical algorithm, which is based on fluorescence spectroscopy. The first step is to dimensionally reduce the measured spectral variables. The second step is to develop a classification scheme for the discrimination of these useful spectral parameters into relevant histologic/histo-pathological categories. The development of current mathematical algorithms based on fluorescence spectroscopy can be classified broadly into three categories: 1) algorithms based on qualitatively selected spectral variables (fluorescence intensities at several emission wavelengths), 2) algorithms based on statistically selected spectral parameters (a more robust evaluation and use of all the measured spectral information) and 3) algorithms based on parameters that reflect the biochemical and/or morphologic features of the tissue. Classification schemes employ either a binary or probability based discrimination. Mostly, algorithms have been based on qualitatively or statistically selected spectral variables in conjunction with binary classification methods.
The sensitivity is defined as the fraction of diseased samples correctly classified and the specificity is defined as the fraction of non-diseased tissues correctly classified. Although in most of the studies, the sensitivity and specificity were evaluated retrospectively, in several studies, the sensitivity and specificity were evaluated prospectively to obtain an unbiased estimate of the performance of the technique under consideration [33,69–77
]. In some studies, the sensitivity and specificity were evaluated for a combination of fluorescence spectroscopy and conventional endoscopy [33,75
]. Finally, in the case of tissues such as the bladder, larynx, skin, breast and stomach, the sensitivity and specificity were reported for the discrimination between cancers and non-cancers, only. This is different from other studies in which the sensitivity and specificity were reported for the discrimination of cancers and pre-cancers
from normal tissues. In the majority of studies performed, the sensitivity and specificity are greater than 80% reflecting the high classification accuracy of fluorescence spectroscopy for the detection of neoplastic tissues, in vivo
. The sensitivity and specificity reported appear to be comparable to or superior to current clinical modalities that are routinely used.
Several groups have evaluated fluorescence spectroscopy for the identification of human colonic neoplasms in vitro
] and in vivo
]. Both steady-state and time-resolved, fluorescence measurements as well as fluorescence imaging have been evaluated at multiple excitation wavelengths.
Richards-Kortum et al.
] measured fluorescence emission spectra from normal colon tissues and adenomatous colonic polyps, in vitro
at a range of excitation wavelengths, spanning the ultraviolet and visible spectrum. They found that the spectral differences between normal and adenomatous colon tissues were greatest at 330, 370 and 430 nm excitation. Furthermore, at an excitation wavelength of 370 nm, fluorescence intensities at 404, 480 and 680 nm were found to be most useful for differentiating adenomas from normal colon tissues. Kapadia et al.
] obtained fluorescence emission spectra from 35 normal specimens and 35 adenomatous polyps of the colon at 325 nm excitation. They used multivariate linear regression analysis of the spectra and a binary classification scheme to develop and optimize an algorithm that differentiates adenomatous polyps from normal mucosa and hyperplastic polyps, retrospectively. The algorithm prospectively discriminated 16 adenomatous polyps from 16 hyperplastic polyps and 34 normal tissues, with a sensitivity of 100% and specificity of 98%. Yang et al.
] evaluated the ratio of the fluorescence intensities at two emission wavelengths, from fluorescence emission spectra at 325 nm excitation and at two excitation wavelengths, from fluorescence excitation spectra at 450 nm, emission of 35 colon adenocarcinomas and 39 normal colon specimens. An algorithm based on the simple ratio of fluorescence intensities and a linear decision surface separated adenocarcinomas from normal tissues with a sensitivity and specificity of greater than 90%. Finally, Chwirot et al.
] measured fluorescence images in six spectral bands at 325 nm excitation from 50 resected specimens of the colon. For the majority of neoplastic lesions, the fluorescence intensity was lower than that of corresponding normal mucosa in all of the spectral bands. The spectral bands centered at 440 nm and 475 nm seemed the most promising.
Schomacker et al.
] measured fluorescence emission spectra from 86 normal sites, 35 hyperplastic polyps and 49 adenomatous polyps of the colon, in vivo
at 337 nm excitation. Multivariate linear regression analysis and a binary classification scheme was used to differentiate spectra of hyperplastic and adenomatous polyps and resulted in a sensitivity and specificity of 86% and 80%, respectively. These values were not significantly different from the accuracy of routine clinical pathology. In a subsequent report, Schomacker et al.
] also examined the fluorescence emission spectra of 86 normal sites, 35 hyperplastic polyps, 49 adenomatous polyps and seven adenocarcinomas of the colon, obtained both in vitro
and in vivo
at 337 nm excitation. The spectra of colonic tissue measured in vivo
and in vitro
were different, primarily due to NADH fluorescence, which decays exponentially with time after resection. Again, using a similar method of analysis, adenocarcinoma and adenoma were distinguished from hyperplastic and normal tissues with a sensitivity and specificity of 80% and 92%, respectively. Adenocarcinoma and adenomatous polyps were distinguished from hyperplastic polyps with a sensitivity and specificity of 86% and 77%, respectively. Eker et al.
] also measured the fluorescence emission spectra from 32 adenomas, 114 hyperplastic polyps of the colon, in vivo
at 337 nm excitation. Stepwise multivariate linear regression and a binary classification scheme yielded a sensitivity and specificity of 100% and 84%, respectively, for discriminating adenomas from hyperplastic tissues and normal mucosa.
Cothren et al.
] obtained fluorescence emission spectra at 370 nm excitation from 31 adenomas, four hyperplastic polyps and 32 normal sites of the colon, in vivo
. They developed an algorithm for differentiating adenomas from normal and hyperplastic tissues based on 1) a drop in the maximum fluorescence intensity at 460 nm and 2) a relative increase in the red fluorescence intensity at 680 nm, which occurred as tissue progressed from normal to hyperplasia to adenoma. Using these spectral features in a Bayesian classification scheme, they could differentiate adenomas from normal mucosa and hyperplastic polyps with a sensitivity and specificity of 100% and 97%, respectively. Another interesting application of fluorescence spectroscopy was presented by Catalano et al.
]. They collected fluorescence emission spectra at 370 nm excitation from the rectal mucosa of 16 patients with a history of adenomatous polyps (within the last 18 months) and from 16, age and gender matched patients with no such history. Polyp formers had an increased fluorescence intensity at 680 nm compared with non-polyp formers. However, differences in the spectra were small and no hypothesis was stated suggesting a predictive capability of this technique for identifying patients with colorectal mucosa at risk for developing colorectal neoplasia.
In their most recent report, Cothren et al.
] developed a probability based algorithm based on fluorescence intensities at several emission wavelengths for the detection of colonic adenomas and evaluated it in a blinded manner. Fluorescence emission spectra were acquired at an excitation wavelength of 370 nm, from 91 normal tissues, 19 hyperplastic polyps and 62 adenomatous polyps, in vivo
() displays fluorescence emission spectra obtained at 370 nm excitation from normal mucosa, a hyperplastic polyp and an adenomatous polyp from a typical patient. Each exhibits strong fluorescence near 460 nm and weak fluorescence at wavelengths greater than 600 nm. Generally, the peak fluorescence intensity of normal mucosa is greater than that of adenomatous polyps, with hyperplastic polyps, intermediate in intensity. These differences were consistent within a patient. The ratio of the mean adenomatous polyp spectrum divided by the mean normal spectrum is shown in (). The flat area near 460 nm indicates a region in which the shapes of both spectra are similar, but where the fluorescence intensity of the normal mucosa is three times as intense as that of the adenoma. The rising curve beyond 600 nm shows the existence of additional structure in the red region of the adenoma spectra.
Figure 9 (a) Fluorescence emission spectra obtained in vivo at 370 nm excitation from normal mucosa, a hyperplastic polyp and an adenomatous polyp from the colon of a typical patient and (b) the ratio of the mean adenomatous polyp fluorescence spectrum divided (more ...)
A calibration data set was used to develop and optimize an algorithm to differentiate tissue types based on probability distributions of the fluorescence intensity at 460 nm and the ratio of the intensities at 680 and 600 nm. Evaluation of the algorithm on the calibration data set containing 44 normal mucosa, 12 hyperplastic polyps and 35 adenomas indicated a sensitivity of 94% and a specificity of 91% for differentiating adenoma from hyperplastic and normal tissues. The algorithm was then tested in a blinded fashion on a prediction data set. Evaluation of the algorithm on the prediction data set, which contained 50 normal mucosa, 17 hyperplastic polyps and 32 adenomas indicated a similar sensitivity and specificity of 90% and 88%, respectively, for differentiating adenomas from hyperplastic polyps and normal mucosa.
Wang et al.
] demonstrated the use of endoscopic fluorescence imaging to identify dysplasia associated with adenomatous polyps in the colon. Endoscopic white light and fluorescence images at 351 to 364 nm excitation and >400 nm emission, were collected using a video colonoscope from a polyp in an area that covered approximately 80% of that illuminated by the white light (total fluorescence image is 150x150 pixels, which corresponds to a region which is 3.5x3.5 cm2
). In total, 12 adenomas and six hyperplastic tissues were evaluated.
Single-point fluorescence intensity measurements obtained from fluorescence imaging are shown in . The fluorescence intensity of normal mucosa was approximately a factor of two greater than that of adenomatous polyps. The intensity of the hyperplastic polyps and that of normal mucosa were similar. Also, foci of dysplasia were distinguished from those of hyperplasia, suggesting that the basis of dysplasia identification was determined by factors other than gross morphology of the polyp. Blood vessels showed a 25% reduction in intensity, relative to corresponding normal mucosa. Mucosal folds were found to cast shadows that also resulted in reduced fluorescence intensity. The shadows were more prominent on fluorescence than on white light images. Twelve adenomatous polyps were identified with a sensitivity of 83%. Because additional biopsy specimens were not obtained from normal appearing colonic mucosa in this study, no comment was made regarding the specificity of this technique.
Table 2 Single-point, Fluorescence Intensity Measurements (Mean±Standard Deviation) Obtained In Vivo at 351 to 364 nm Excitation and >400 nm Emission, from 12 Adenomas, Six Hyperplastic Polyps and Six Blood Vessels and the Corresponding Normal (more ...)
In addition to the work conducted by Wang et al.
], preliminary reports [89,90
] of real-time fluorescence imaging with blue light excitation and a false-color display of the relative red and green fluorescence intensities have demonstrated that flat adenomatous lesions in the colon that were not visible by white light endoscopy (WLE) could be detected. Furthermore, hyperplastic polyps could be distinguished from adenomatous polyps. However, the sensitivity and specificity were not reported.
Unlike the previous groups, Mycek et al.
] explored the feasibility of making time-resolved, rather than steady-state, fluorescence measurements from the colon, in vivo
and assessed the sensitivity and specificity of this technique for distinguishing adenomatous polyps from non-adenomatous polyps. Time-resolved, fluorescence measurements were obtained in vivo
from the colon at 337 nm excitation and 550 nm emission (40 nm band pass), with a time resolution of 4.8 nsec. Seventeen patients with a total of 13 adenomatous and 11 non-adenomatous (six hyperplastic, three mucosal prolapse, one lymphoid aggregate, one aberrant crypt focus) polyps were studied.
) shows (a) representative fluorescence decay profiles of an adenomatous polyp and (b) the normalized profiles (to a peak intensity of 1), which reveal identical fluorescence decays between two different measurements. From these curves, the transient decay time was characterized by measuring the time interval between 0.2 of the normalized fluorescence intensity at the rising edge of the curve and at the decaying portion of the curve. For the two polyp groups, the mean value of the decay time was found to be 9.3 ± 0.4 nsec for adenomatous polyps and 10.5 ± 0.7 nsec for non-adenomatous polyps. By choosing a threshold of 9.8 nsec, the adenomatous polyps were distinguished from non-adenomatous polyps with a sensitivity of 85% and specificity of 91%. The accuracy of the time-resolved fluorescence technique is comparable to the accuracy of routine clinical pathology, which is approximately 87% as noted in two previous studies [69,83
(A) Representative fluorescence decay profiles of an adenomatous polyp and (B) the normalized profiles (to a peak intensity of one), which reveal identical fluorescence decays between two different measurements.
summarizes the results of the clinical investigations performed thus far, on fluorescence spectroscopy of the colon, in vivo
. The table indicates: (1) the excitation wavelengths, (2) the measurement type, (3) the emission wavelengths of the fluorescence intensities used and the corresponding decision scheme, (4) the sample size for the non-diseased (ND) and diseased (D) tissue categories and finally, (5) the resulting sensitivity (SE) and specificity (SP) for differentiating adenomas (A) and/or adenocarcinomas (C) from hyperplasia (H) and/or normal tissue (N). Evaluation of indicates that the two primary excitation wavelengths used were 337 and 370 nm excitation. All of the groups, except for two, measured fluorescence emission spectra. Of the remaining two, one measured fluorescence images [88
], whereas the other measured the transient fluorescence decay from colon tissues [32
], in vivo
. Either the entire fluorescence emission spectra (at 337 nm excitation) or fluorescence intensities at a few emission wavelengths (at 370 nm excitation) were used as the spectral discriminators. A binary classification scheme was used in all cases, except for one (in which a Bayesian probability based classification scheme was used) [69
]. Furthermore, in all of the studies performed, except for one, the performance of fluorescence spectroscopy was evaluated retrospectively. Only Cothren et al.
] used a calibration and prediction data set, respectively, to develop and test their technique, in an unbiased manner. The sensitivity and specificity for differentiating adenocarcinoma/adenoma from hyperplastic and normal tissues ranged from 80% to 100%, with no dramatic differences resulting from for the measurement type, excitation wavelength, emission wavelength or method of analysis used.
Table 3 A Summary of Results Obtained by Several Groups on Fluorescence Spectroscopy of the Colon, In Vivo. The Table Indicates: 1) the Excitation Wavelengths, 2) the Measurement Type, 3) the Emission Wavelengths of the Fluorescence Intensities Used and the Decision (more ...) Cervix
Several groups have explored the utility of detecting cervical neoplasia, in vitro
using fluorescence spectroscopy [91–94
]. However, Richards-Kortum and coworkers [70–72,95–102
] have done the most extensive development and evaluation of fluorescence spectroscopy for the detection of cervical neoplasia, in vivo
. All of the efforts have focused on the measurement of single-pixel, fluorescence emission spectra from cervical tissues at multiple excitation wavelengths.
Lohmann et al.
] first investigated the utility of fluorescence emission spectra at 365 nm excitation, for the recognition of cervical neoplasia, in vitro
. They found that the spectra of the cervix exhibited a single peak with a maximum at 475 nm. The fluorescence intensity increased, with the degree of increase proportional to the degree of dysplasia. In contrast, the fluorescence intensity of carcinoma was very small, with a rise in intensity at the border between the carcinoma and normal tissue. Glassman et al.
] measured fluorescence emission spectra of normal tissues and gynecological tissues with carcinoma, in vitro
at 300 and 320 nm excitation. Consistent differences were seen in the fluorescence intensity of carcinoma and normal tissues from all sites. Mahadevan et al.
] demonstrated that the greatest spectral differences in the fluorescence emission spectra of non-neoplastic and neoplastic cervical tissues, in vitro
occur at 340, 380 and 460 nm excitation. Koumantakis et al.
] studied the fluorescence emission spectra of malignancies in the female genital tract, in vivo
, including the ovaries, fallopian tube, uterus and cervix uteri, using 442 nm excitation and concluded that this could enhance selective detection of malignant tissue, and thereby reduce the risk of leaving it untreated.
Richards-Kortum and coworkers have developed multivariate algorithms, based on fluorescence emission spectra at multiple excitation wavelengths for the detection of cervical dysplasia [97,98
]. They have tested these algorithms retrospectively and prospectively on diagnostic [70
] and screening [72
] populations. Furthermore, they have evaluated the safety of using fluorescence spectroscopy relative to standard procedures, such as colposcopy [99
]. Also, they have evaluated the effect of acetic acid, cervical mucus and vaginal medications on the fluorescence emission spectra [100
]. Finally, they have constructed receiver-operator curves (ROC) for fluorescence spectroscopy, standard clinical screening and diagnostic tests and competing technologies, to evaluate the performance of each for the detection of cervical dysplasia [101
]. Finally, they have performed a cost analysis to demonstrate the potential of a diagnostic technique, such as fluorescence spectroscopy to reduce health care costs in the detection of cervical dysplasia [102
]. The following paragraphs provide a more detailed discussion of these various studies.
Initially, Ramanujam et al.
] collected fluorescence emission spectra in vivo
from 114 cervical tissue sites from a group of 28 patients at 337 nm excitation. A two-stage algorithm based on empirically selected spectral parameters demonstrated that it is feasible to discriminate between low grade squamous intraepithelial lesions (SILs) and normal epithelia as well as between high grade SILs and low grade SILs. Subsequently, Ramanujam et al.
] expanded their efforts to measure fluorescence emission spectra from the cervix at multiple excitation wavelengths. Specifically, tissue spectra from 40 patients at 337 and 380 nm excitation and from 24 patients at 337 and 460 nm excitation demonstrated that spectra at multiple excitation wavelengths provide better discrimination between dysplastic and non-dysplastic tissues.
In the most recent report, Ramanujam et al.
] obtained fluorescence emission spectra at all three excitation wavelengths (337, 380 and 460 nm) from a total of 186 normal squamous tissues, 27 normal columnar tissues, 29 tissues with inflammation, 47 low grade SILs and 70 high grade SILs in 95 patients. A multivariate statistical algorithm was developed for the differential detection of SILs and particularly, high grade SILs. There were four primary steps involved in the multivariate statistical analysis of tissue spectra. The first step was to pre-process spectra to reduce inter-patient and intra-patient variation within a tissue type. The pre-processed spectra were then dimensionally reduced to an informative set of principal components, which describe most of the variance of the original spectral data set using Principal Component Analysis (PCA). Next, the principal components, which contain diagnostically relevant information, were selected using an unpaired, one-sided student's t
-test. Finally, a classification scheme based on Bayes theorem was developed using these diagnostically relevant principal components.
Inputs into the multivariate statistical algorithm included the pre-processed fluorescence emission spectra at all three excitation wavelengths (full-parameter) and fluorescence intensities at a select, reduced number (15) of excitation-emission wavelength pairs (reduced parameter) selected from the component loadings of the principal components. The algorithm was developed on a calibration set and then tested on a prediction set with an approximately equal number of samples in each tissue category. The prospective sensitivity and specificity of the full-parameter, multivariate statistical algorithm for differentiating SILs from non-SILs was 82% and 68%, respectively. This algorithm differentiated high grade SILs from non-high grade SILs prospectively, with a sensitivity and specificity of 79% and 78%, respectively. The sensitivity and specificity of the reduced-parameter algorithm, which uses an order of magnitude fewer variables was within 5% of that reported for the full-parameter algorithm. This suggests that fluorescence intensities at a reduced number of optimal, excitation-emission wavelength pairs, rather than the entire fluorescence emission spectra can be measured and employed for the detection of cervical dysplasia.
Subsequent to this analysis, a neural-networks-based classification scheme, which replaces the steps involving PCA and Bayes theorem was developed [71
]. This algorithm used the pre-processed fluorescence intensities at the 15 excitation-emission wavelength pairs (inputs into the reduced-parameter, multivariate statistical algorithm) as inputs into a radial basis function network ensemble to classify the different tissue types. Again, the algorithm was optimized on a calibration set and tested on a prediction set of approximately equal prior probability. The prospective sensitivity and specificity of the neural-networks-based algorithm for differentiating SILs from non-SILs were 91 ± 1.5% and 67 ± 0.75%, respectively. The similarity in the performance of the linear (multivariate statistical algorithm) and non-linear (neural networks) methods of analysis provided confidence in the results obtained from this clinical investigation.
In a subsequent investigation, Brookner et al.
], measured cervical tissue fluorescence emission spectra at 337, 380 and 460 nm excitation from 54 women who had not had an abnormal Pap smear, previously (screening population). These were compared to previously reported results from 95 patients, all of whom had abnormal Pap smears (diagnostic population) [70
]. Before fluorescence spectroscopy in the screening population, a Pap smear was performed on each woman. The results indicated an abnormal finding in only four of the 54 women evaluated. In the 50 women with normal Pap smears, spectra were measured from a total of 186 sites; of these, 103 contained normal squamous tissues, 23 contained normal columnar tissues and 60 were from the transformation zone. In the four women with abnormal Pap smears, spectra were measured from 16 sites; eight contained squamous tissue, one contained columnar tissue and seven were from the transformation zone.
Classification of the normal tissue fluorescence emission spectra using the previously developed multivariate statistical algorithm [70
] indicated that 87% of normal squamous tissues and 78% of tissues from the transformation zone were correctly classified. Furthermore, 76% of normal columnar tissues were correctly classified. Hence, even though the multivariate statistical algorithm had been developed for a population in a diagnostic setting, its performance was robust in a screening setting as well.
Brookner et al.
] also examined the photochemical risks associated with the detection of human cervical dysplasia using fluorescence spectroscopy. In particular, they compared the relative risk of fluorescence spectroscopy to colposcopy (a procedure in which a low power microscope is routinely used to illuminate the cervix for evaluation). They measured the average wavelength-dependent spectral radiant exposure (J/cm2
per nanometer) during a colposcopic examination and during fluorescence spectroscopy of the cervix. To quantify the relative risk, they multiplied these illumination spectra by several action spectra (efficacy of photochemical damage as a function of wavelength) from the literature and compared the areas under the curve, corresponding to each procedure. Based on this comparison, they concluded that risks of illumination for fluorescence spectroscopy are lower than or comparable to those already encountered in routine diagnostic procedures such as colposcopy.
Agrawal et al.
] explored the fluorescence properties of several substances commonly found on the cervix: acetic acid, cervical mucus, and vaginal medications. Acetic acid is routinely used during colposcopy, to enhance visual differences between normal and abnormal areas of the cervix. Areas, which may develop into cervical cancer, undergo a transient whitening visible to the naked eye. Cervical mucus is often present on the ectocervix as well as in the fold of the endocervical canal. During colposcopy, attempts are made to remove this mucus from the cervix to improve visualization of the tissue, but complete removal can be difficult. Vaginal medications are commonly used for the treatment of yeast and other infections.
The results of this investigation showed that acetic acid introduces changes in both the line shape and intensity of fluorescence emission measured from the cervix, at 337 nm excitation. Specifically, the application of acetic acid caused a decrease in the fluorescence intensity of dysplastic and non-dysplastic tissues. However, the dysplastic tissues displayed a greater decrease in the fluorescence intensity. At emission wavelengths, below 500 nm, acetic acid caused a greater percentage of patients to exhibit differences between dysplastic and non-dysplastic tissue in their fluorescence emission spectra. At emission wavelengths above 500 nm, acetic acid provided no appreciable enhancement. The measured mucus transmission demonstrated strong protein absorption below 300 nm. In general, no significant absorption bands were observed in the visible part of the spectrum. The mucus fluorescence intensity was strongest at 280 nm excitation and 340 nm emission (tryptophan). Weaker fluorescence intensities could be observed at 350 nm excitation, 450 nm emission (NADH) and at 450 nm excitation, 535 nm emission (FAD). In examining the spectra of the 16 medications, 10 of them had clear fluorescent peaks, whereas six of the agents exhibited very broad fluorescent peaks or no recognizable peak at all. In summary, while acetic acid enhances fluorescence spectroscopy of dysplastic and non-dysplastic cervical tissues, cervical mucus and vaginal medications that produce strong fluorescence, could interfere with these measurements.
Mitchell et al.
] described ROC curves for fluorescence spectroscopy, which were generated from fluorescence spectroscopy in a diagnostic setting [70
] and compared them with ROC curves for other diagnostic methods (colposcopy, Pap smear, cervicography, speculoscopy, HPV testing), which were calculated from published reports.
Summary ROC curves were developed from independent reports of the sensitivity and specificity of each test. Furthermore, the area under the curve (AUC) and the Q point on the ROC curve (at which the sensitivity equals the specificity) were calculated. Evaluation of these features indicated that fluorescence spectroscopy outperformed the other tests, but most importantly, compared favorably with colposcopy, the current standard diagnostic technique for cervical neoplasia.
Finally, Cantor et al.
] compared five strategies for the diagnosis and treatment of cervical SILs, including those that incorporate colposcopy and fluorescence spectroscopy. On the basis of a health care perspective, they performed a cost-effective analysis using a decision-analytic model for the diagnosis and management of SILs. They compared the five strategies based on expected costs and number of cases that were treated appropriately, missed, treated inappropriately and appropriately not treated in a hypothetical cohort of 100 patients referred after an abnormal Pap smear. Data on prevalence and operating characteristics were derived from the medical literature. Costs were adjusted from hospital charge data. A see-and-treat strategy based on fluorescence spectroscopy was the least expensive, but also the least effective strategy, costing US$160,479 to detect 31.55 cases of SIL accurately in 100 patients. The most expensive strategy was colposcopically directed biopsy, at US$311,808 to find 45.78 cases; however, when both tests were used in a see-and-treat strategy modality, slightly more cases were found (46.05) ata lower cost (US$285,133).
Lam and coworkers [33,104–110
] have developed and evaluated fluorescence emission spectra and subsequently, fluorescence images at two emission wavelengths for the detection of invasive carcinoma, CIS and dysplasia of the bronchus and lung, in vivo
. This work has resulted in a commercial, laser induced fluorescence endoscopy (LIFE) device (Xillix Technologies Corporation, Richmond, BC, Canada). The LIFE device has been tested in a multi-center clinical trial to evaluate if fluorescence bronchoscopy when used as an adjunct to white light bronchoscopy (WLB) can improve the bronchoscopist's ability to locate areas suspicious of dysplasia for biopsy and histologic examination. The LIFE device has also been successfully used to evaluate neoplastic lesions in other organ sites, including the larynx [75,76,111
], gastrointestinal tract [89,90
], head and neck [73,74
] and the bile duct [77
Originally, Hung et al.
] measured the bronchial fluorescence emission spectra of 32 patients with severe dysplasia, CIS and invasive carcinoma at 405, 442 and 488 nm excitation, in vivo
. The spectra at 405 and 442 nm excitation, were similar to each other, with two peaks at 520 and 590 nm. Furthermore, the spectra of normal and neoplastic tissues showed similar line shapes at these excitation wavelengths. However, the intensity decreased as tissue progressed from normal to severe dysplasia to CIS to invasive cancer. Palcic et al.
] subsequently developed an endoscopic-based fluorescence bronchoscope that provided simultaneous white light images and pseudo-color images based on the ratio of the fluorescence intensity in the green and red wavelength bands, at 442 nm excitation (basis for the design of the LIFE device). Lam et al.
] evaluated 53 patients with known bronchogenic carcinoma and 41 volunteers, using conventional WLB and the fluorescence bronchoscope described previously [105
]. They showed that the two techniques had a similar specificity of 94%. However, the sensitivity of fluorescence bronchoscopy (73%) was 50% greater than that of WLB (48.4%) for detecting dysplasia and CIS.
Most recently, Lam et al.
] evaluated if fluorescence bronchoscopy with the LIFE device when used as an adjunct to WLB could improve the bronchoscopist's ability to locate areas suspicious of dysplasia for biopsy and histologic examination. A multi-center trial was conducted in seven institutions in the United States and Canada. WLB followed by fluorescence bronchoscopy with the LIFE device was performed in 173 patients known or suspected to have lung cancer. The histologic diagnosis for a total of 700 biopsy specimens were as follows: 321 normals; hyperplasia, metaplasia or mild dysplasia in 237; moderate/severe dysplasia in 93; CIS in nine and invasive carcinoma in 40. Thus, a total of 142 biopsy specimens were graded as moderate dysplasia or worse, a total of 102 biopsy specimens were graded as moderate/severe dysplasia or CIS and a total of 40 biopsy specimens were graded as invasive carcinoma. This was then used as the standard to determine the relative sensitivity of the bronchoscopic procedure.
A three-point classification system was used for WLB evaluation. Areas without any visual abnormality were classified as class 1. Areas with non-specific erythema, swelling or thickening of the bronchial mucosa, bronchoscopic trauma, anatomic anomalies or granulation tissue were classified as class 2. Nodular/polypoid lesions, irregularity of the bronchial mucosa, or focal thickening of the subcarina, suspicious for high grade dysplasia or carcinoma were labeled as class 3. In fluorescence bronchoscopy, green images were considered normal and class 1, areas that were slightly brown with ill-defined margins were labeled as class 2 and brownish/red images were considered as class 3. The physician's bronchoscopic classification was converted from a three-point scale to a two-point scale in which classes 1 and 2 were considered negative and class 3 was considered positive.
On a lesion-by-lesion analysis, WLB alone had a sensitivity of 24.6% for detecting moderate dysplasia or worse. With the addition of fluorescence bronchoscopy, the sensitivity increased to 66.9%. On a per patient basis, the sensitivity of WLB for detecting individuals harboring at least one lesion was 37.3%. With the addition of fluorescence bronchoscopy, the sensitivity increased to 75%. On a lesion-by-lesion basis, WLB had a sensitivity of 8.8% for detecting moderate/severe dysplasia or CIS. The addition of fluorescence bronchoscopy increased the sensitivity to 56%. Finally, WLB had a sensitivity of 65% for detecting invasive carcinoma. WLB + LIFE improved this to a sensitivity of greater than 90%. It should be noted that the addition of fluorescence bronchoscopy added an average of 13.8 minutes to a conventional WLB, which took an average of 9.4 minutes. No adverse events related to the use of this device were observed.
Several groups [27,112–115
] have measured single-pixel, fluorescence emission spectra at different excitation wavelengths from invasive carcinoma, early carcinoma and dysplasia of bladder tissues, in vivo
D'Hallewin et al.
] evaluated the fluorescence emission spectra of normal and neoplastic urothelium, including CIS at 365 and 355 nm excitation. The line shape of the fluorescence emission for CIS and TCC were identical to that of normal bladder tissue. However, the fluorescence intensity of CIS was lower than that of normal tissues (average decrease of 2.6 times), with TCC exhibiting the weakest fluorescence intensity (average decrease of 3.2 times). Using this spectral information, they demonstrated that fluorescence emission spectra at 365 nm excitation can be used to discriminate between normal bladder, CIS and TCC, in vivo
with high accuracy.
Baert et al.
] measured fluorescence emission spectra of the bladder at 337 and 405 nm excitation. Using 337 nm excitation, the ratio of the fluorescence intensities at 460 and 400 nm was used to differentiate successfully between normal urothelium and visible papillary bladder carcinoma. Using 405 nm excitation, the fluorescence intensity for some cases of dysplasia was lower than that of normal tissue. However, this was not consistent.
Koenig et al.
] explored the applicability of fluorescence spectroscopy for the detection of bladder carcinoma. Fluorescence emission spectra at 337 nm excitation were measured from a total of 35 sites with inflammation (24 chronic and 11 acute), 42 normal areas, one squamous metaplasia, seven dysplasias, 28 carcinomas and one CIS.
The fluorescence intensity of the normal urothelium was 20 times greater than that of carcinoma in the human bladder. Inflammatory tissues also showed weak fluorescence, but there were significant spectral differences between these and carcinoma. While inflammatory tissues and normal tissues had two fluorescence maxima at 385 and 455 nm, the typical spectrum of the carcinomas showed only one maximum at 455 nm. Using the ratio of fluorescence intensities at 385 and 455 nm, and a binary decision scheme, carcinoma was discriminated from non-carcinomas with a sensitivity and specificity of 97% and 98%, respectively. Differentiation of dysplastic bladder lesions from normal urothelium was not possible. Furthermore, no significant differences in spectra could be detected between the different stages (T1 or T3) or grades (1 to 3) of carcinoma. More recently, Koenig et al.
] performed an investigation in which they measured fluorescence emission spectra at 337 nm excitation from suspicious (erythematous, edematous, raised and so forth) bladder lesions and areas from which random biopsies were obtained during routine cystoscopy. They also measured epithelial thickness in all biopsy samples to determine whether it correlates with the spectra, measured in vivo
. Spectra at 337 nm excitation were measured during cystoscopy from a total of 130 bladder areas in 17 women and 58 men in whom bladder carcinoma was suspected. A total of 23 and 107 biopsies, respectively, were obtained from areas with and without carcinoma. The ratio of the fluorescence intensities at 385 and 455 nm was determined for every measured area and correlated with histologic results and epithelial thickness. Using the ratio parameter in a binary decision scheme, 95% of carcinomas were correctly classified and only 30 non-carcinomas were classified as false-positives. This suggests that the number of non-cancerous areas biopsied during routine white light cystoscopy would be reduced from 107 to 30 (72% fewer) if fluorescence spectroscopy was used to guide biopsy. Furthermore, the ratio of the fluorescence intensity at 385 and 455 nm was found to decrease exponentially with increasing epithelial thickness between 1 and 225 µ
m. Beyond 225 µ
m, there was no further ratio decay indicating that most of the measured fluorescence originates from the uppermost 225 µ
m of the epithelium.
Anidjar et al.
] investigated the ability of fluorescence spectroscopy to distinguish urothelial bladder carcinoma from normal or non-specific inflammatory mucosa. Fluorescence emission spectra were acquired from 22 normal mucosa, 13 sites with non-specific inflammation, five sites with CIS and 26 invasive cancers at 480, 337 and 308 nm excitation.
At 480 nm excitation, the fluorescence emission spectrum has a peak at 580 nm. The fluorescence intensity of inflammatory tissues was decreased compared to that of normal mucosa. The intensity was decreased further for all bladder carcinomas, including CIS. At 337 nm excitation, the fluorescence emission spectra had a peak at 450 nm. Similar changes in the fluorescence intensity were observed at 337 nm excitation as was observed at 480 nm excitation. At 308 nm excitation, the fluorescence emission spectra of normal mucosa consisted of one broad band of fluorescence with two secondary maxima at 380 and 440 nm. The spectra of tumors differ markedly from that of normal mucosa, since the two distinct fluorescence bands were shifted to 360 and 440 nm. Furthermore, the ratio of the fluorescence intensities at 360 and 440 nm increased significantly as tissue progressed from normal to inflammation to carcinoma. Using this diagnostic parameter and a binary decision scheme, a 100% sensitivity and specificity was achieved for the detection of bladder carcinoma.
summarizes the results obtained by the aforementioned groups on fluorescence spectroscopy of the bladder, in vivo. The table indicates that spectra were acquired from these tissues at a range of excitation wavelengths from 300 to 500 nm. However, only spectra at 308, 337 and 365 nm excitation were analyzed for the purpose of developing classifiers. In all the cases, the ratio of fluorescence intensities at two emission wavelengths was used in a linear decision scheme to discriminate between carcinomas and non-carcinomas. These algorithms retrospectively discriminated between these two tissue types, with a sensitivity that was close to 100% and a specificity that ranged from 70% to 100%.
Table 4 A Summary of Results Obtained by Several Groups on Fluorescence Spectroscopy of the Bladder, In Vivo. The Table Indicates: 1) the Excitation Wavelengths, 2) the Measurement Type, 3) the Emission Wavelengths of the Fluorescence Intensities Used and the (more ...) Brain
Two groups have reported on the measurement of fluorescence emission spectra from brain tissues, in vitro
and in vivo
. Montan and Strombland [116
] examined the fluorescence emission spectra at 337 nm excitation of neoplastic and non-neoplastic brain tissue, in vivo
during a craniotomy and in vitro
following tissue preservation by freezing or formaldehyde. They reported that meningiomas, but not astrocytomas differed from normal, white and gray matter. Preservation by freezing, significantly altered both the fluorescence line shape and intensity, whereas formaldehyde fixation did not affect spectral line shape, but did affect the fluorescence intensities.
Bottiroli et al.
] performed a preliminary investigation to evaluate if fluorescence spectroscopy can be used for intra-operative delineation of tumor resection margins in the brain. Fluorescence emission spectra at 366 nm excitation were measured from the brain cortex, white matter and neoplastic lesion of three patients with glioblastoma. In the cases evaluated, the fluorescence intensity of the glioblastoma was significantly lower than that of the white matter. A reduction in fluorescence intensity, although to a lesser extent, was observed in the glioblastomas, relative to the cortex tissues in two of the three patients. Compared with normal tissues, the glioblastoma spectra also showed a red-shift in the peak fluorescence emission.
Single-pixel, steady-state, fluorescence emission spectra have been measured from Barrett's esophagus, esophageal cancer and normal tissues, in vivo
. Originally, Panjehpour et al.
] and Vo-Dinh et al.
] measured fluorescence emission spectra of normal and cancerous tissue during endoscopy in patients with esophageal cancer and in cancer free volunteers at 410 nm excitation. Multiple fluorescence measurements were obtained from 32 patients including measurements from 123 normal areas and 36 esophageal cancers. Using a classification model based on linear discriminant analysis, esophageal cancers were classified with a sensitivity and specificity of 100% and 98%, respectively.
Subsequently, Panjehpour et al.
] described the use of endoscopic fluorescence spectroscopy at 410 nm excitation to detect high grade dysplasia in patients with Barrett's esophagus. In 36 patients, a total of 216 spectra were obtained from non-dysplastic Barrett's mucosa, 36 spectra were obtained from mucosa with low grade dysplasia, 10 spectra were obtained from tissues with high grade dysplasia and 46 spectra were obtained from mucosa with low grade dysplasia and focal high grade dysplasia.
The normalized fluorescence emission spectrum of a non-dysplastic Barrett's mucosa was characterized by a broad band emission with a peak at around 500 nm. A comparison between the normalized spectra of the non-dysplastic and dysplastic mucosa indicated that the latter is significantly lower in intensity and slightly red shifted. A mathematical model based on differential normalized fluorescence was developed and the fluorescence intensities at 480 and 660 nm were selected as the spectral variables. Discrimination was achieved using a binary decision scheme. The sensitivity and specificity of fluorescence spectroscopy for discriminating high grade dysplasia from low grade dysplasia and normal tissues was 40% and 96%, respectively. However, when the spectra were re-evaluated per patient, the sensitivity and specificity increased to 100% and 76%, respectively. It should be noted that erosive esophagitis contributed to an increase in the false-positive rate. Hence, further studies are required to determine if this technique can correctly classify tissues with inflammation.
Head and neck— oral cavity
Several groups have investigated the utility of fluorescence measurements for detecting invasive carcinoma, early carcinoma and dysplasia of head and neck tissues, in vitro
] and in vivo
]. Both fluorescence spectra and images have been evaluated.
Ingrams et al.
] compared the fluorescence emission spectra from a total of 12 normal (healthy mucosa or benign lesions) and 10 abnormal (dysplastic or carcinoma) tissue samples, in vitro
at multiple excitation wavelengths between 250 and 500 nm. Significant spectral differences were observed between the two groups, with the greatest differences observed at 410 nm excitation and 635 nm emission. Majumder et al.
] measured the fluorescence emission spectra of normal and cancerous oral tissues at 300, 337 and 460 nm excitation. They found that the integrated fluorescence intensity at 337 nm excitation was higher in normal tissues, relative to that in cancerous tissues. These differences were not observed at 300 and 460 nm excitation. Yang et al.
] also measured fluorescence emission spectra at multiple excitation wavelengths (280, 290, 300, 320, 330, 340 nm) for distinguishing between cancerous and normal mucosal tissues, in vitro
. They found that by using ratios of intensities at specific emission wavelengths, at the short and long excitation wavelengths in a binary decision scheme, they were able to achieve a sensitivity of 81.25% and a specificity of 93.75%.
Kolli et al.
] were the first to measure fluorescence excitation and emission spectra from normal mucosa and neoplastic lesions of the oral cavity in 31 patients, in vivo
. They found that the fluorescence properties of the neoplastic mucosa were significantly different from those of normal mucosa when excited with UV light in the 200 to 340 nm range.
Dhingra et al.
] characterized the fluorescence emission spectra of healthy and diseased oral mucosa and oropharynx, in vivo
at 370 and 410 nm excitation. Fluorescence emission spectra were obtained from normal mucosa (n
=5), hyperplasia and hyperkeratosis (n
=8), lichen planus (n
=1), dysplasia (n
=4), CIS (n
=1) and invasive SCC (n
All fluorescence emission spectra exhibited a single broad peak centered at 450 and 490 nm with excitation wavelengths of 370 and 410 nm, respectively. At 410 nm excitation, the neoplastic tissues displayed a decreased blue intensity and increased red intensity relative to non-neoplastic and normal tissue. These spectral differences were much less dramatic at 370 nm excitation. The fluorescence intensities at 490 and 640 nm emission (excitation wavelength, 410 nm), were normalized to the corresponding intensity at the contralateral normal site from the same patient. These were then incorporated into a simple binary decision algorithm to differentiate neoplastic tissues from non-neoplastic and normal tissues with a sensitivity and specificity of 100% and 87.5%.
Gillenwater et al.
] evaluated fluorescence spectroscopy for the detection of oral cavity neoplasia, in vivo
at 337, 365 and 410 nm excitation. A total of 27 sites in 10 patients were evaluated and histology indicated that 17 sites were normal and 11 sites had dysplasia or carcinoma.
In general, the peak fluorescence intensities of abnormal sites were less than those of normal sites. The fluorescence intensities of abnormal sites were increased in the red spectral region compared with those of normal sites. Although differences were noted at all three excitation wavelengths, the fluorescence intensity at 337 and 410 nm excitation provided the best discrimination between normal and abnormal tissues. An algorithm was developed to discriminate dysplasia and carcinoma from normal mucosa using two variables, which exploit these spectral differences. These were the peak intensity at 337 nm excitation and the ratio of the intensities in the red (635 nm) and blue (490 nm) regions at 410 nm excitation (red-blue peak ratios). To account for inter-patient variation, the ratio of the peak intensity at 337 nm excitation of the normal and abnormal spectra was calculated. Also, the ratio of the red-blue peak ratios at 410 nm excitation for abnormal and normal tissues was calculated. Using these two parameters in a binary classification algorithm, carcinoma and dysplasia could be differentiated from normal tissues with a sensitivity and specificity of 94.1% and 100%, respectively. Clinical impression had a lower sensitivity and similar specificity of 76.5% and 100%.
With respect to fluorescence imaging of head and neck tumors, Kulapaditharom and Boonkitticharoan [73
] aimed to: 1) test whether the LIFE device (Xillix Technologies) described previously for fluorescence bronchoscopy [33
] could correctly identify sites of cancer in the head and neck and 2) to compare the performance of LIFE and WLE for cancer detection. Thirty-two examinations with LIFE and WLE were performed in 25 patients with signs and/or symptoms suggesting cancers in the head and neck (larynx, oral cavity and tracheobronchial tree). Pre-treatment and post-treatment investigations with LIFE and WLE were conducted to see whether LIFE correctly reflects the treatment effects as defined by clinical and histologic findings. Biopsies were performed at spots in which positive findings were suggested by either or both imaging modalities. Random biopsies were obtained in cases where both techniques revealed negative findings.
In evaluating 16 cancerous lesions and 16 non-cancerous lesions, LIFE had a sensitivity and specificity of 100% and 87.5%, respectively, whereas WLE had a sensitivity of 87.5% and a specificity of only 50%. Both imaging techniques encountered certain cases of false-positives. Inflammatory masses and granuloma were the causes. A comparison of WLE and LIFE for evaluating treatment response in six patients indicated that LIFE was considered to be more reliable (accuracy 93.8%) than WLE (accuracy 69%) for differentiating diseased from non-diseased tissue.
Kulapaditharom and Boonkitticharoan [74
] also used fluorescence imaging with the LIFE device to determine the effectiveness of detecting unknown primary cancers in the head and neck. Thirteen patients with biopsy-proven cervical node metastases were evaluated prospectively with the LIFE device and panendoscopy. Among the 13 positive sites identified by the LIFE device, histopathology confirmed five as SCC, four as dysplasia, two as inflammation and two as normal. Panendoscopy with random biopsy located only two lesions. In summary, the LIFE device and panendoscopy localized unknown primary cancers at rates of 38.5% and 15.4%. The LIFE device not only aided in revealing a greater proportion of occult primary cancers, but also helped in reducing the number of unnecessary biopsies.
Betz et al.
] evaluated normal and malignant mucosa in patients with head and neck cancer through measurements of both fluorescence images and fluorescence emission spectra. Specifically, fluorescence images were acquired at 375 to 440 nm excitation and >515 nm emission on a total of 30 patients. Fluorescence emission spectra were acquired using the same excitation band from a healthy volunteer and a total of 36 patients. Evaluation of the results indicated that in 13 of the 30 patients (43.3%), tumors were distinguished from their normal surroundings through a reduction of the green fluorescence intensity. However, spectral analysis indicated that the peak fluorescence intensity at 511 nm was lower in tumors relative to that of normal mucosa in 34 of the 36 patients. The spectral line shape of the tumors did not differ markedly from that of normal mucosa.
Head and neck—larynx
There have been several reports on steady-state, fluorescence measurements and imaging of laryngeal cancers, in vivo
]. In 1995, Harries et al.
] acquired fluorescence emission spectra at 442 nm excitation, from normal and pathologically confirmed tissues of the larynx in eight patients, in vivo
. They also acquired fluorescence images using the LIFE device (Xillix Technologies), which is described previously for fluorescence bronchoscopy [33
]. The results suggested that the fluorescence properties of laryngeal tissue are similar to those of bronchial tissue and that the LIFE device has the potential to increase the accuracy of cancer staging in the larynx.
Zargi et al.
] determined the role of fluorescence imaging for the detection of laryngeal cancer. Fluorescence laryngoscopy of the red and green fluorescence intensities at 442 nm excitation was performed using the LIFE device (Xillix Technologies) [33
] in 40 patients [75
]. Histopathology of 115 biopsy specimens proved 28 sites to be positive and 87, negative for malignancy. While laryngomicroscopy alone had a sensitivity and specificity of 60% and 83%, respectively, the combination of laryngomicroscopy and fluorescence laryngoscopy had a sensitivity and specificity of 85% and 87%, respectively. In a subsequent investigation [76
], Zargi et al.
evaluated the effect of keratosis on microlaryngoscopy and fluorescence laryngoscopy in 30 patients. After histopathology, 22 sites were identified as malignant and 72 sites were identified as benign. For clinical sites that included keratosis, microlaryngoscopy had a sensitivity and specificity of 54% and 86%, respectively. The sensitivity and specificity of the LIFE device was 59% and 76%, respectively. For tissues without keratosis, microlaryngoscopy had a sensitivity and specificity of 50% and 86%, respectively, whereas LIFE had a sensitivity and specificity of 56% and 75%, respectively. Keratosis, which emits a strong fluorescence, can give a false impression of a malignant process, hence reducing the specificity of fluorescence imaging.
summarizes the results of the clinical investigations performed thus far, on fluorescence spectroscopy of the head and neck, in vivo
. Evaluation of indicates that a variety of excitation wavelengths between 337 and 440 nm have been used. Both fluorescence emission spectra and images have been measured. Note that in all cases, fluorescence intensities or ratios of intensities at several emission wavelengths were used in conjunction with a binary classification scheme. While most of the algorithms were evaluated only retrospectively, some were evaluated prospectively as well [73–76
]. The sensitivity and specificity for differentiating neoplastic from non-neoplastic oral cavity tissues ranged from 85% to 100%, except in the investigation conducted by Betz et al.
], where fluorescence imaging resulting in a significantly lower sensitivity than fluorescence spectroscopy. The sensitivity and specificity for differentiating laryngeal cancers from the normal larynx was poor for fluorescence imaging alone [75,76
], but increased substantially when it was combined with microlaryngoscopy [75
]. It should be noted that in all studies except those reported by Dhingra et al.
] and Gillenwater et al.
], the sensitivities and specificities are reported for the discrimination between cancers and non-cancers.
Table 5 A Summary of Results Obtained by Several Groups on Fluorescence Spectroscopy of the Head and Neck, In Vivo. The Table Indicates: 1) the Excitation Wavelengths, 2) the Measurement Type, 3) the Emission Wavelengths of the Fluorescence Intensities Used and (more ...) Skin
There have been mixed reports on the diagnostic potential of fluorescence spectroscopy and imaging for skin melanomas, in vivo
. The first attempt at using fluorescence spectroscopy for the detection of melanoma, in vivo
was made by Lohmann [67
]. They excited healthy skin tissues, naevi and melanomas at 365 nm excitation and recorded the corresponding fluorescence emission spectra. They found that the spectra had a peak at 475 nm. The fluorescence intensity was very low for melanoma, compared to that of normal skin tissue, and there was a local increase in the intensity in a transition zone between the melanoma and areas of healthy skin, followed by a drop in the intensity measured for the latter. Such local maxima were not found for the naevi. When the above hypothesis was tested in a larger group of 147 patients [128
], it was found that the ratio of the fluorescence intensity at 470 nm outside and within the pigmented lesions was substantially higher for the melanomas compared with naevi.
However, these reports [67,128
] were not confirmed by Sterenborg et al.
]. They measured fluorescence, EEMs of normal skin and human skin carcinomas in vivo
at excitation wavelengths ranging from 375 to 600 nm. They found no correlation between either the line shape or the intensity of the fluorescence emission and histopathology of the tissue. They concluded that fluorescence spectroscopy is not a feasible diagnostic technique for melanomas.
However, Chwirot et al.
] used fluorescence imaging for the detection of melanomas, in vivo
at 365 nm excitation and 475 nm emission, as was done by Lohmann [67
] and Lohmann et al.
], with moderate success. Fluorescence images were obtained for 90 patients with melanomas of different forms and stages, 169 patients with 205 pigmented naevi and 105 patients with 113 skin lesions of different types, hereafter referred to as others.
Healthy skin was characterized by a relatively homogeneous distribution of fluorescence intensity. The fluorescence images of melanoma showed inhomogeneous spatial distributions of intensity, both within the lesions and around them. Ratios of maximum fluorescence intensity measured for regions located up to 40 mm from the lesions, to a minimum fluorescence intensity within the lesions, were useful for the detection of melanomas. Using this ratio in a binary classification scheme, the sensitivity and specificity for differentiating melanoma from naevi and other skin lesions was 82.5% and 78.6%, indicating that fluorescence imaging is indeed a feasible diagnostic technique for melanomas.
Thus far, there has only been one report on fluorescence imaging of bile duct cancer, in vivo
]. In this pilot investigation, nine patients with bile duct cancer underwent percutaneous, transhepatic, fluorescence cholangioscopy using the LIFE-GI (Xillix Technologies) device, described previously for fluorescence bronchoscopy [33
]. Fluorescence images in the red and blue wavelength emission bands were obtained at 437 nm excitation. Histologic confirmation was performed with surgically resected specimens or biopsy specimens. Under observation of the fluorescence cholangioscope, normal mucosa was seen as light blue, whereas cancerous lesions were observed as irregular and heterogeneous dark red areas. Additionally, in seven of nine patients, a white color (strong fluorescence) was seen. It corresponded to the region of abundant fibrous stroma in the cancerous lesion. This white fluorescence (which may reflect the saturation of detector) reflects the enriched fibrous stroma, which is exposed to the internal lumen as a result of cancer filtration.
Several groups have evaluated the fluorescence excitation and emission spectra for breast cancer detection, albeit, this has been done only on tissues, in vitro
. Alfano et al.
] measured the fluorescence emission spectra of normal and malignant breast tissues in vitro
from two patients at 488 nm and at 457.9 nm excitation. Normal and malignant tissues both showed a peak at about 515 nm, while the normal tissue also exhibited additional maxima at 556 and 592 nm. Yang et al.
] obtained fluorescence excitation and emission spectra of normal, malignant and adipose breast tissues, in vitro
. They were able to distinguish malignant tissue from adipose glandular fibrous and normal tissue with a sensitivity of 91% (56 malignant tissues) and a specificity of 95% (46 benign/normal tissue specimens). Furthermore, three ratios of fluorescence intensities at different wavelengths in the excitation spectrum over the 250 to 320 nm range, for 340 nm emission, in conjunction with a binary decision scheme, yielded a sensitivity and specificity of 90% to differentiate 103 malignant tissues from 63 benign/normal tissues [133
]. Gupta et al.
] compared the fluorescence emission spectra of benign (fibroadenoma, 35 patients), cancerous (ductal carcinoma, 28 patients) and normal tissues, in vitro
at 337 nm excitation. The fluorescence intensity, which was integrated over the entire emission wavelength range, was increased in cancerous tissues, relative to that in benign and normal tissues. Using this variable in a linear decision algorithm, cancerous tissues were discriminated from benign and normal tissues with an accuracy of 100%. Lohmann and Kunzel [40
] found that the fluorescence emission spectra of breast tissue at 340 to 380 nm excitation was associated primarily with connective tissue fibers in between lactiferous ducts and lobular complexes.
Chwirot et al.
] presented the first and only report on fluorescence imaging of stomach cancers, in vitro
. Specifically, they measured fluorescence images of 21 resected specimens of stomach cancers and normal tissues, in vitro
at six visible wavelengths, at 325 nm excitation. Using the fluorescence intensities at 440 nm and 395 nm, normalized to that at 590 nm and a linear decision surface, cancerous tissues were classified with a sensitivity of 96%.