Fluorescence Excitation Emission Matrices of Human Tissue: A System for in vivo Measurement and Method of Data Analysis

We describe a system capable of measuring spatially resolved reflectance spectra from 380 to 950 nm and fluorescence excitation emission matrices from 330 to 500 nm excitation and 380 to 700 nm emission in vivo. System performance was compared to that of a standard scanning spectrofluorimeter. This “FastEEM” system was used to interrogate human normal and neoplastic oral cavity mucosa in vivo. Measurements were made through a fiber-optic probe and require 4 min total measurement time. We present a method based on autocorrelation vectors to identify excitation and emission wavelengths where the spectra of normal and pathologic tissues differ most. The FastEEM system provides a tool with which to study the relative diagnostic ability of changes in absorption, scattering, and fluorescence properties of tissue.


INTRODUCTIO N
A growing number of clinical studies have demonstrated that¯uorescence spectroscopy can be used to distinguish normal and abnorm al human tissues in vivo in the skin, 1 head and neck, 2,3 genito-urinar y tract, 4,5 gastro-intestinal tract, 6,7 breast, 8 and brain. 9 References 10 ±13 provide recent reviews of this ® eld. It is well known that uorescence intensity and line shape are a function of both the excitation and emission wavelength in samples containing multiple chromophores, such as human tissue. A complete characterization of the¯uorescence properties of an unknown sample requires measurem ent of ā uorescence excitation emission matrix, in which thē uorescence intensity is recorded as a function of both excitation and emission wavelength. The ® eld of analytical chemistry has exploited the¯uorescence properties of different compounds to identify and quantify them in mixtures. Holland et al. 14 described a computer-based system to m easure¯uorescence and absorbance simultaneously from quinine bisulfate. Warner and colleagues 15 built an instrument to collect a matrix of 241¯uorescence spectra at 241 different excitation wavelengths in 16.7 ms and used it to analyze the¯uorescence properties of mixtures of hydrocarbons. Linear algebra methods were then applied to determine quantitative concentration and num ber of analyte data from these m easurements. 16±19 These m ethods assume a linear system and are much more successful in nonscattering, optically dilute samples Receiv ed 16 July 1998; accepted 3 November 1998. * Author to whom correspondence should be sent. than they are in human tissue. In tissue, chromophores are embedded in a scattering and absorbing medium affecting both the excitation light reaching the chromophores and the emission light leaving the sample. The relationship between the¯uorescence spectra of dilute and turbid samples has been explored 20 and found to be highly nonlinear. Studies have also been carried out investigating the perform ance of different linear algebra methods in the determination of concentrations and number of chromophores present in a homogeneous turbid sample. 21 However, it is not clear that they can be applied to inhomogeneous tissues.
Most clinical studies reported to date have measured uorescence emission spectra at only a sm all number of excitation wavelengths (typically one to three) due to clinical requirements imposed on the size, speed, and sensitivity of instrumentation. The choice of excitation wavelength has been based on factors which vary from study to study, but include laser availability, 22 predictions of chromophores thought to be present in normal and abnorm al tissues, and m easurements of¯uorescence excitation emission matrices (EEMs) of normal and abnormal tissues in vitro. 23,24 W hile in vitro m easurements of tissue EEMs are feasible with the use of comm ercially available scanning¯uorim eters, several studies have demonstrated that the optical properties of tissue change signi® cantly when tissue is examined in vitro due in part to interruption of the blood supply, 25 oxidation, 26 and small size of biopsies. 27 Thus, in vitro studies to select excitation wavelengths are of limited value.
It is well known that the absorption and scattering properties of tissues in vivo affect both the intensity and line shape of measured¯uorescence spectra. 28 The optical properties of tissue vary spatially, depending on its architecture, blood supply, and m etabolic state; however, spatially resolved m easurements of diffusely re¯ected light can be used to estimate tissue optical properties in vivo. 29 ±33 Several recent studies have suggested that differences in these optical properties, assessed by using diffuse re¯ectance spectroscopy, can be used to discriminate normal and abnormal human tissues in vivo in the urinary bladder 34 and the skin. 35 Furthermore, m easuring both uorescence and diffuse re¯ectance spectra m ay provide additional information of diagnostic value. 36,37 A system capable of measuring spatially resolved reectance spectra and¯uorescence excitation emission matrices in vivo would remove the limitations of many previous studies, potentially enabling prediction of those excitation wavelengths which provide greatest discrimination of normal and abnormal tissues, as well as a better understanding of the relative diagnostic ability of changes in absorption, scattering, and¯uorescence properties of tissue. Two ® ber-optic systems to record¯uorescence EEMs and re¯ectance spectra at a single spatial location have been previously described in the literature. 38,39 Zuclich and colleagues 39 developed a ® ber-optic system to record both diffusely re¯ected excitation light and¯uorescence emission by using a single 14 bit m ultichannel detector. Sequential m easurements were m ade, ® rst with white light to collect the diffuse re¯ectance spectrum and then at a series of excitation wavelengths to collect both elastic backscattering and any induced¯uorescence. Light was provided by a pulsed, ® ltered arc lamp from 300 to 600 nm in 10 nm increments, and following m easurem ent, all spectra were assembled to yield the diffuse re¯ectance spectrum and a¯uorescence EEM for a single spatial location. Collection of elastic backscattering and uorescence EEMs from nonscattering tissues such as the human lens required 1±2 m in. Typically, the elastic backscattering was at least 3±4 orders of m agnitude stronger than the detected¯uorescence, leaving a ver y limited dynamic range available to measure tissue¯uorescence.
More recently, Zangaro et al. 38 described a system to record excitation emission m atrices and diffuse re¯ectance from a single spatial location. A N 2 laser that pumps a series of 10 dye-containing cuvettes m ounted on a rotating wheel provides sequential excitation at 11 different wavelengths from 337 to 500 nm; multichannel detection at each excitation wavelength yields 11 emission spectra that are then assembled to yield the excitation emission matrix. White light illumination is then used to produce a single diffuse re¯ectance spectrum. Collection of a tissue EEM and re¯ectance spectrum requires approximately 600 m s at a signal-to-noise ratio (SNR) of 50:1 for colon tissue.
In this paper we describe a system for in vivo measurem ent of¯uorescence excitation emission matrices and spatially resolved diffuse re¯ectance spectra. An arc lamp coupled to a scanning spectrometer provides continuously tunable excitation light, which is coupled into a ® ber-optic probe. Resulting tissue¯uorescence is collected through the ® ber probe and delivered to an imaging spectrograph and charge-coupled device (CCD) camera. Fluorescence emission spectra are collected sequentially at 18 excitation wavelengths ranging from 330 to 500 nm, which are then assembled into a¯uorescence EEM. Subsequently, white light is coupled into the probe, and diffusely re¯ected light exiting the tissue at three spatial locations is detected from 380 to 950 nm by using the spectrograph and CCD. With this system, tissue EEMs and re¯ectance spectra with SNR in excess of 25: 1 can be collected in approximately 4 m in. We present EEMs and re¯ectance spectra of normal hum an oral cavity mucosa and a tumor of the oral cavity obtained in vivo. Finally, we describe a new method, based on excitation and emission autocorrelation vectors, to analyzē uorescence EEMs to determine excitation and emission wavelength regions where differences between normal and abnormal tissues are greatest. We illustrate the application of this method to the EEMs. Figure 1 illustrates a block diagram of the system (FastEEM system). The system consists of three m ain components: (1) an arc lamp, stepper motor-driven monochromator, and ® lter wheel, which provides monochromatic and broad band excitation; (2) a ® ber-optic probe that directs excitation light to the tissue and collects re-mitted¯uorescence from one location and diffusely reected light from three locations; and (3) a ® lter wheel, imaging spectrograph, and CCD camera that detects the spectrally resolved re¯ectance and¯uorescence signals. Excitation monochromator position, ® lter wheel position, spectrograph grating position, CCD operation, and data acquisition are controlled by a laptop personal computer mated to a docking station. The speci® cations of each sub-system are described below.

M ATERIALS AND M ETH ODS
The probe, illustrated in Fig. 2, consists of a total of 46 optical ® bers [200 m m diameter, numerical aperture (NA ) 5 0.2] arranged in two concentric bundles. The center bundle contains 25¯uorescence excitation ® bers and 12¯uorescence collection ® bers. The proximal ends of the¯uorescence excitation ® bers are arranged in two vertical lines at the exit slit of the excitation m onochromator to maximize the coupling of the light into the sample. The proximal ends of the¯uorescence collection ®bers are arranged in a single vertical line at the entrance slit of the imaging spectrograph. At the distal end of the probe, the ® bers that excite and collect¯uorescence are arranged randomly in a central bundle and placed in contact with a short piece of a thick quartz ® ber (2 mm diameter, 15 m m long, NA 5 0.2). The distal tip of this ® ber is placed in contact with the sample surface and ensures that the area from which¯uorescence is collected is the same as that directly illuminated.
The nine ® bers for illumination and collection of diffuse re¯ectance are arranged in a concentric ring around the thick quartz¯uorescence measurement ® ber. The distal ends of these ® bers are¯ush with the tip of the central ® ber and are placed in contact with the sample surface. White light from a port on the side of the lamp housing is coupled to the proximal end of a single illumination ® ber (80 m m, NA 0.2). Photons that scatter through the tissue and exit the surface are collected at ® ve different positions with seven collection ® bersÐ three located 180 8 from the illumination ® ber (3 mm distance), two located 908 from the illumination ® ber (2.1 mm ), and two located 458 from the illumination ® ber (1.1 mm ) as shown. The proximal ends of the re¯ectance collection ® bers are situated at the top of the vertical line of¯uorescence collection ® bers, separated by dum my ® bers, as shown in Fig. 2.
The light source for the instrum ent, which provides both quasi-monochromatic excitation for¯uorescence and broadband illumination for re¯ectance, is a 150 W ozone-free Xe arc lamp (Spectral Energy Corp., Westwood NJ) with a spherical rear re¯ector. A condenser system consisting of two plano-convex quartz lenses is used to couple light into a monochromator. The primar y condenser is 1.5 in. in diameter with an aperture ratio of f /1.5. The secondary condenser is also 1.5 in. in diameter, but is masked to provide num erical aperture matching to the monochrom ator. A manual shutter is located between the condensing optics and m onochromator and is closed to prevent¯uorescence excitation light from reaching the sample during re¯ectance measurem ents. The monochrom ator has an aperture ratio of f /3.6 (Spectral Energy, GM 252) and is used with an ion-etched holographic grating (ISA, Edison, NJ; 240 nm blaze, 1180 grooves/mm , dispersion 5 3.3 nm/mm ). An RS-232 controlled stepper m otor drives the m onochromator with a maximum stepping rate of 400 step/s (approximately 10 nm /s). A bandwidth of 6.6 nm is selected by setting the entrance slit of the m onochromator to 2.0 m m. Light is coupled from the m onochromator into the probe via a ® ber-optic adapter (Spectral Energy, GMA 257) consisting of a quartz plano-convex lens and a 53 quartz microscope objective. The light passing through the objective is focused onto a vertical line of 25 ® bers in two columns, placed at the focal plane of the objective. The re¯ectance excitation ® ber is attached to the lamp housing via a micropositioner. Broadband light exiting the lamp housing through an existing hole is coupled to the re¯ectance illumination ® ber by using a quartz planoconvex lens (NA 5 0.24). A ® ve-position illumination ® lter wheel placed between the lamp and the lens contains three long-pass ® lters with 50% transmission at 295, 515, and 715 nm. One of the ® lter positions is blocked and acts as a shutter to prevent white light from reaching the sample during¯uorescence measurem ents.
Light collected by¯uorescence and re¯ectance ® bers is coupled through an eight-position, computer-controlled collection ® lter wheel into a Chrom ex 250 IS (Albuquerque, NM) imaging spectrograph containing a holographic grating blazed at 380 nm with 150 grooves/mm and a reciprocal linear dispersion (RLD) of 20 nm/m m. The ® bers are projected onto an entrance slit (250 m m) that yielded a spectral resolution of 5 nm. A thermoelectrically cooled CCD camera operated at 2 30 8 C (Spectrasource HPC-1, Westlake Village, CA) was located at the back focal plane of the imaging spectrograph. Chip dimensions were 13.8 3 9.2 mm with 1536 3 1024 pixels (Kodak KAF-1600 grade 2), yielding a nom inal spectral range of 276 nm for a single grating position. Dark current is speci® ed as 0.25 electrons/pixel/s when operated at 2 30 8 C. The quantum ef® ciency of the lumogen coated chip ranges from a peak of 40% at 550 nm to a low of 15% at 250 nm.
The detector and imaging spectrograph were wavelength calibrated by m easuring the room light spectra that showed three mercury peaks at 404.7, 436, and 546 nm. The relation between pixels and wavelength was then linearly ® tted through these points.
Fluorescence and re¯ectance m easurements are obtained sequentially. Prior to¯uorescence m easurements, the white light port is closed and pixels illuminated by the¯uorescence ® bers are selected to be read from the CCD. Dark current and A/D conversion offset is measured with the same setting as the subsequent measurement but with a closed camera shutter. These are subtracted from all¯uorescence and re¯ectance m easurements. The ® rst excitation wavelength is selected by scanning the excitation m onochromator, the emission ® lter wheel is rotated to select the appropriate long-pass ® lter, and the spectrograph grating is adjusted to record signal over the desired emission wavelength range. The monochrom ator and camera shutters are then opened for the desired exposure time to record the¯uorescence emission spectrum (1.5 s). The excitation wavelength is then increm ented, and the process repeated until all desired excitation wavelengths have been measured. The excitation wavelengths were incremented from 330 to 500 nm in 10 nm steps. Table I   excitation wavelengths and corresponding long-pass ® lters and emission wavelength ranges used in this study. Following collection of¯uorescence spectra, diffuse reectance spectra are then m easured. For these measurements, the m onochromator shutter is closed, the emission ® lter wheel is set to the lowest ® lter position, and the pixels illum inated by the corresponding re¯ectance collection ® bers are selected to be read from the CCD. Dark current and A/D conversion offsets are m easured and stored for subtraction of the following m easurements. The re¯ectance spectrum is collected over three illumination wavelength ranges. Prior to measurem ent of each range, the appropriate long-pass ® lter is selected in the illumination ® lter wheel, and the spectrograph grating is adjusted to record signal over the desired wavelength range. The lamp and camera shutters are then opened for the desired exposure time to record the re¯ectance spectrum (0.4 ±4.8 s). The illumination wavelength range is then incremented, and the process repeated until all desired wavelength ranges have been m easured. Exposure times are determined empirically to achieve an SNR greater than 20. Table I contains a list of the illumination wavelength ranges and corresponding long-pass ® lters used for diffuse re¯ectance measurem ents. The high dynamic range of the re¯ectance m easurements, spanning over three orders of m agnitude, requires that each spatial position be read out individually from the CCD. This approach prevents saturation and blooming artifacts.
There are no accepted safety standards for illumination of m ucosal surfaces other than skin and cornea. However, we calculated the exposure of solar radiation that is equivalent to the exposure received when a m easurement is made with our system. The method compares the spectral irradiance (W /cm 2 nm) of the excitation source with solar irradiance data obtained from Ref. 40. The comparison consists of a point-wise division of the irradiance from the FastEEM system to the solar irradiance at the same wavelength. This ratio gives a relative solar exposure factor. The solar data are for a sunny day in San Diego, CA. Irradiation during¯uorescence excitation is less than 7 times solar exposure at all wavelengths. Given that¯uorescence excitation times were 1.5 s, this corresponds to exposure to solar radiation for less than 11 s in any given wavelength band. During diffuse re¯ectance measurements, the lamp exposure is maximum at 300 nm, where the relative exposure is a factor of 25 that of the sun. Since the total exposure time for this wavelength band is 14 s, the exposure corresponds to 350 s or less than 6 min. All other wavelengths have relative exposure factors of 10 or less, resulting in a shorter equivalent total solar exposure.
Prior to every patient m easurement, the probe output was measured with a calibrated power meter (Newport, Irvine, CA, 818-UV) at 400 nm excitation wavelength. An average output of 86 m W 6 12 m W was achieved at this wavelength with a bandwidth of 6.6 nm. Background uorescence spectra were measured with the probe dipped in a non¯uorescent bottle containing distilled water. This background EEM was subtracted from all subsequently acquired EEMs to correct for room lights and probe auto¯uorescence. The nonuniform spectral response of the system was corrected by using correction factors determined from m easurements of calibration sources; in the visible, we used a National Institute of Standards and Technology (NIST ) traceable tungsten ribbon ® lament lamp, and in the UV, we used a deuterium lamp (550C and 45D, Optronic Laboratories Inc., Orlando, FL). Variations in the intensity of the¯uorescence excitation light source at different excitation wavelengths were corrected by using m easurements of the intensity at each excitation wavelength at the probe tip with the use of a calibrated photodiode (818-UV, Newport). Background spectra to correct re¯ectance m easurements for room light contributions were m easured with all parameters set as for tissue measurem ents, except that the white light shutter was closed. These m easurements were subtracted from all subsequent re¯ectance spectra.
Fluorescence and re¯ectance standards were m easured before each patient m easurement. The¯uorescence intensity is reported relative to the¯uorescence intensity of a solution of 2 mg/L Rhodamine 610 (E xciton, Dayton, OH) in ethylene glycol at 460 nm excitation and 580 nm emission. Re¯ectance data are reported with respect to a 2.68% -by-volume solution of 1.072 m m diameter polystyrene m icrospheres (Polyscience Inc., Warrington, PA). The m icrosphere standard was used for its well-characterized optical properties. The total integrated re¯ectance of this standard was m easured on a double-beam spectrophotometer (U-3300 Hitachi, Tokyo, Japan) with an integrating sphere attachm ent (L absphere Inc., North Sutton, NH). This measurem ent was used to correct the reectance standard measurem ents made with the FastEEM system. Tissue spectra at each collection ® ber position were divided pointwise by the corrected standard re¯ectance spectrum at the corresponding ® ber position.
The EEMs were assembled of¯ine from each series of uorescence emission scans. Data processing and plotting were performed with Matlab (T he Math Works Inc., Natick, MA). Re¯ectance spectra were assembled from three wavelength areas, giving a range from 380 to 950 nm. The wavelength range was further reduced (380 ±800 nm ) to comply with the range of calibration m easurements of the re¯ectance standards on the U-3300. Re¯ectance data are reported between 380 and 595 nm, a range where the possible in¯uence of room lights in the m easurement was minimized.
System Validation. The system performance was assessed by using two¯uorescence standards. The ® rst standard was a 2 mg/L Rhodamine 610 (E xciton Inc., Dayton, OH) ethylene glycol solution, which is nonscattering but has peak¯uorescence intensity approximately twice the average intensity of human cervix. 41 The second standard mimics the optical properties of tissue and consists of 20 m M Flavin Adenine Dinucleotide (FAD, Kodak, Rochester, NY), 0.625 vol % polystyrene m icrospheres (Polyscience Inc., diameter 5 1.072 m m ). 42 Both standards were m easured with the FastEEM system and a scanning spectro¯uorimeter (SPEX, Fluorolog II, Edison, NJ). The EEMs m easured with the SPEX were considered as standards since the perform ance of the system is well documented (dynamic range 5 10 5 , spectral resolution 5 nm, corrected for nonuniform spectral response). The excitation light was incident perpendicular to the sampling cuvette, and the emitted light was collected at approximately a 208 angle with respect to the excitation light. A front focus arrangement with a 10 m m cuvette was used in the SPEX. Sixty m inutes was required to collect a full EEM from each sample with the SPEX.
Clinical Studies. In vivo data were obtained from a group of patients with known or suspected prem alignant or malignant lesions of the oral cavity. The studies were reviewed and approved by the Internal Review Board of the University of Texas at Austin and the Surveillance Com mittee at the UT M. D. Anderson Cancer Center (Houston). Inform ed consent was obtained from each person in the study. Before the probe was used, it was disinfected with Metricide (Metrex Research Corp.) in accordance with the standard clinical protocol. Background uorescence EEM and re¯ectance spectra were m easured by dipping the ® ber-optic probe in a non¯uorescent bottle ® lled with deionized water. These EEMs and spectra correspond to the system auto¯uorescence and were subtracted from all subsequently acquired EEMs for that patient. Next an EEM was measured from a Rhodamine calibration standard, and a re¯ectance spectrum was m easured from a polystyrene solution calibration standard. The probe was then guided to the tissue site to be examined and its tip positioned¯ush with the tissue. Ā uorescence EEM and re¯ectance spectra were obtained from sites within a lesion and a clinically normal site. Post-spectroscopy, a 2±4 m m biopsy of the tissue was taken from normal and abnormal sites where the probe measured spectra. These specimens were evaluated by an experienced pathologist, Bonnie Kem p, M.D., with the use of light m icroscopy and were classi® ed by using standard diagnostic criteria.
Data Analysis. One of the goals of instrument developm ent is to provide information for the identi® cation of excitation wavelengths suitable for the differentiation of tissue of differing pathological characteristics, as well as identi® cation of the chrom ophores responsible for the differences. W hile all such information is present in the EEMs collected, it can be dif® cult to extract due to the dimensionality of the data set. A m ethod was devised to separately characterize the excitation and emission characteristics of the data set. Given that the EEM has dimensions corresponding to (l x , l m ), the following autocorrelation vectors are de® ned: where x av (l x ) is the excitation autocorrelation vector and m av (l m ) is the emission autocorrelation vector. Essentially, the emission autocorrelation vector is the diagonal of the product of the EEM with its transpose, and the excitation autocorrelation vector is the diagonal of the product of the transpose of the EEM with the EEM. Note that in signal processing terms, the autocorrelation vectors, x av and m av , are a measure of the average signal of the EEM at each excitation or emission wavelength, respectively. In this way they provide qualitative inform ation about an EEM. An example with simulated data is presented in Fig. 3 to illustrate how autocorrelation vectors re¯ect changes in¯uorescence peak positions in EEMs. Two kinds of changes are simulated in the m odeled data: a shift in the excitation wavelength at which a¯uorescence peak appears, and a shift in the emission wavelength at which ā uorescence peak appears. The original peak in the EEM was m odeled as a single Gaussian at 380 nm excitation, 550 nm emission with a full width at half-maximum (FWHM) of 35 nm in emission and excitation wavelengths. The original peak was then shifted by 30 nm in excitation, as shown by arrow 1 in Fig. 3a. The shift in emission wavelength is shown by arrow 2 in Fig. 3a, and corresponds to a 30 nm shift in the emission peak of the original data. Three sets of autocorrelation vectors were computed: one for the EEM with the original peak, one for the EEM with the excitation wavelength-shifted peak, and one for the EEM with the emission wavelength-shifted peak. The autocorrelation vectors are shown in Fig.  3b. Comparing the vectors for the original EEM (row 1 in Fig. 3b) with the vectors from the EEM with the excitation wavelength-shifted EEM (row 2 in Fig. 3b), it is seen that the excitation autocorrelation vector is sensitive to the change in excitation wavelength but not in emission wavelength. Similarly, comparing the autocorrelation vectors for the original EEM with the vectors from the EEM with the emission wavelength shift in the peak (row 3 in Fig. 3b) shows that the emission autocorrelation vector is sensitive to the changes in emission wavelength but not excitation wavelength.
It is sometimes desirable to normalize the autocorrelation vectors to facilitate comparisons between different sets of measurem ents. We calculated normalized autocorrelation vectors by dividing these vectors by their root mean square (rms) value, in effect forcing the area of the vector to 1 unit of signal energy. The normalized emission autocorrelation vector is well suited for the identi-® cation of differential features in EEMs, such as the shifting or broadening of¯uorescence peaks. Figures 4A and 4B show¯uorescence EEMs of the nonscattering Rhodamine standard and the scattering FAD phantom obtained with the FastEEM system. Intensities are reported with respect to the Rhodamine inten-  sity measured at 460 nm excitation and 580 nm emission wavelength. Figures 5a and 5c show¯uorescence emission spectra of the Rhodamine standard obtained at 370 and 450 nm excitation with the SPEX and the FastEEM system as well as the¯uorescence background. Figures  5b and 5d show the same spectra for scattering FAD phantom obtained at the same excitation wavelengths. The spectra are normalized at their maximum. Note the presence of Rayleigh scattering peaks from the excitation source in the data taken with the SPEX. In general, from nonscattering samples (Figs. 5a, 5c) the FastEEM system collects less light above 600 nm than the SPEX. This result could be due to the different collection ef® ciencies of the FastEEM probe and the front-face collection geometry of the SPEX. Under scattering conditions and with lower¯uorescence signal, the in¯uence of back-ground¯uorescence becomes m ore critical. At 370 nm excitation wavelength, the FastEEM system m easures more¯uorescence below 500 nm. A comparison with the measured¯uorescence background, however, shows that the additional signal has the same shape as the background. We hypothesize that the background has been underestimated by measuring it in a nonscattering nonuorescent media.

RESULTS AND DISCUSSIO N
In vivo¯uorescence EEMs of the oral cavity were measured from 71 sites, and in vivo re¯ectance spectra were m easured from 49 sites. These were obtained from patients in two studies. The ® rst study involved patients with abnormal oral lesions identi® ed in a previous m edical examination (17 patients). The second study, including nine patients, involved normal volunteers. All sites interrogated spectroscopically in patients with lesions were biopsied and submitted for histopathological analysis. Spectra and biopsies were also obtained from a contralateral site with no lesion in these patients with abnormal lesions. These biopsies were also evaluated histopathologically. No biopsies were taken from the normal  volunteers. Here, we show representative EEMs from tissue found to be histopathologically normal and m alignant to illustrate spectral features detectable with the FastEEM system. Two EEM contour plots from a normal and an abnormal area of the tongue are presented in Figs. 6A and 6B, respectively. In the normal sample,¯uorescence is observed throughout the whole collection range, with a peak located at 330/380 (excitation/emission) and a ridge extending from 340/450 to 450/500 nm. Table II lists excitation-emission m axima pairs of endogenous tissue chromophores. Comparison of the obser ved peaks with Table II shows that these peaks are consistent with the emission of structural proteins such as collagen and elastin, pyridine nucleotides (NA DH), and¯avoproteins (FAD). The norm al site shows overall increased¯uorescence with respect to the abnorm al site shown in Fig. 6B. The abnormal site, assessed by a pathologist as being moderately differentiated squamous cell carcinoma, also shows broad¯uorescence throughout. Peaks are obser ved at 330/380, 350/460, 460/520, and 500/630 nm . A valley is seen at 420 nm excitation between 560 and 580 nm emission. This valley is seen to extend along the 420 nm excitation line as well as the 580 nm emission line. Table  III suggests that these features are produced by hemoglobin reabsorption. Hem oglobin reabsorption may also in part account for the shift in the peaks of the abnorm al EEM with respect to the normal EEM. A summ ar y of the excitation and emission m axima for the peaks observed in the normal and abnormal sites m easured is presented in Table IV.
Fluorescence emission spectra at three selected excitation wavelengths are shown in Fig. 7, illustrating changes in relative intensities of¯uorescence emission. For comparison purposes, each set (norm al/abnormal) was normalized to the maximum at 350 nm excitation. Figure 7a shows the emission spectra at 350 nm excitation. Fluorescence from the normal site is seen as a broad peak with a maximum at 455 nm. The peak from the abnorm al site is seen to be narrower and red-shifted. Examination of this spectrum at 410, 540, and 580 nm suggests that the change in line shape is due to oxygenated hemoglobin. The general line shapes of the¯uorescence obser ved at 410 nm excitation (Fig. 7b) are seen to be similar for both sites in the 450 ±575 nm emission range, with a broad peak at 500 nm. The abnormal site shows a signi® cantly lower¯uorescence intensity, as well as an extra, narrow¯uorescence peak at 640 nm, attributed to porphyrin¯uorescence. Figure 7c shows the emission spectra at 460 nm excitation. The normal site shows a broad peak at 520 nm and clear m odulation from he-   moglobin reabsorption at 540 and 580 nm. Fluorescence from the abnormal site shows an even m ore marked hemoglobin reabsorption; also the overall¯uorescence intensity is reduced. Figure 8 shows the emission and excitation autocorrelation vectors for the same measurem ents. Note that the plots have a logarithmic y axis. The emission autocorrelation vectors have a large broad peak at 460 nm corresponding to the m ain¯uorescence peak observed in the EEMs. The vectors show the effect of hemoglobin absorption around 410, 540, and 580 nm in the abnormal site and the presence of additional¯uorescence in the UV in the normal sample (Fig. 8A). This autocorrelation vector also highlights the peak at 610 nm in the abnormal sample. The excitation autocorrelation vectors show different line shapes. The cur ve corresponding to the normal site decreases steadily from 330 to 500 nm excitation. The curve from the abnormal site shows a peak at 350 and a m inimum at 410 nm. The latter illustrates the great- er in¯uence of hemoglobin reabsorption in the abnormal sample, also shown in Fig. 7.
The corresponding re¯ectance data are shown in Fig.  9. Position 1 corresponds to the collection ® bers closest to the source ® ber and position 3 to those furthest from the source ® ber, as shown in Fig. 2. Differences induced by the¯uorescence reabsorption of oxygenated hemoglobin in the normal site and abnormal site are shown. The modulation of the spectrum by the 540 and 580 nm absorption bands is seen to be signi® cantly stronger in the abnorm al sample; this observation is consistent with the increased reabsorption seen in the¯uorescence spectrum of the abnormal sample. The re¯ectance in the blue range (450 ±500 nm) of the abnormal site is consistently higher than that of the normal site. Below 450 nm the re¯ectance seems not to differ between the normal and abnormal samples.

CONCLUSIO N
The total data acquisition time for the data presented here was 2.5 min for a¯uorescence EEM and 1.5 m in for the spatially resolved re¯ectance measurem ents. However, only 29 s of this time represents¯uorescence collection. Actual re¯ectance collection time was 26 s. The most time-consuming process was changing the excitation wavelength by using the stepper m otor-controlled excitation spectrograph and changing the corresponding long-pass ® lter with the use of the remotely controlled ® lter wheel. Worm drive-based monochromators are available (DDD180, ISA) that require less than 10 s to scan our entire wavelength range in 10 nm steps and could substantially reduce the total m easurement time. Using a higher power lamp will further reduce acquisition time of both¯uorescence and re¯ectance.
We have demonstrated the acquisition of EEMs in combination with spatially resolved re¯ectance m easurements of tissue phantom s and in the oral cavity in vivo with good signal-to-noise ratio. The system features easy and arbitrar y selection of excitation wavelengths in the UV and visible range. The system is also portable and is capable of functioning in a hospital operating room. Probes used in the FEEM system incorporate channels to measure spatially resolved re¯ectance and¯uorescence and are made small enough (, 5 m m) to be used during endoscopic surgical procedures. Autocorrelation vectors x av and m av are a suitable m ethod to reduce the data set while preserving information about the wavelength bands carr ying information. On the basis of the representative data shown here,¯uorescence emission and excitation as well as re¯ectance data appear promising for the identi-® cation of tumors of the oral cavity. The FastEEM system is an ideal tool to identify a subset of the most promising optical features to identify pathological ® ndings in large clinical studies.