IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996
113
Morphological Model of Human Colon Tissue Fluorescence George I. Zonios, Robert M. Cothren, Joseph T. Arendt, Jun Wu, Jacques Van Dam, James M. Crawford, Ramasamy Manoharan," and Michael S . Feld, Member, IEEE
Abstract- Fluorescence spectroscopy of tissue is a promising technique for early detection of precancerous changes in the human body. Investigation of the microscopic origin of the clinically observed tissue fluorescence can provide valuable information about the tissue's histology. The objective of this study was the development of a morphological model of colon tissue fluorescence which connects the clinically observed spectra with their underlying microscopic origins. Clinical colon tissue fluorescence spectra were modeled by measuring the intrinsic fluorescence properties of colon tissue on a microscopic level and by simulating light propagation in tissue using the Monte-Carlo method. The computed spectra were in good agreement with the clinical spectra acquired during colonoscopy, and exhibited the characteristic spectral features of the in vivo collected spectra. Our analysis quantitated these spectral features in terms of the intrinsic fluorescence properties of tissue and its general histological characteristics. The fluorescence intensity difference between normal and adenoma observed in vivo was found to be due to the increased hemoglobin absorption,the reduced mucosal fluorescence intensity,and the absence of submucosal fluorescence in adenomatous polyps. The increased red fluorescence in adenoma was found to be associated with the dysplastic crypt cell fluorescence.
not evident on gross endoscopic inspection, random sampling must be employed in the screening process. The fact that foci of dysplasia may be very small can lead to large sampling error [l], [2]. Flat lesions are more likely to be malignant than elevated adenomas of similar size [3]. For this reason, their early diagnosis may be more important than that of adenomatous polyps. Researchers have begun to explore the use of laser-induced fluorescence (LIF) spectroscopy as a tool for identifying dysplasia in human tissues. This technique studies the autofluorescence induced in the tissue by low intensity laser light. The features of the resulting spectra are determined by the tissue composition and architecture, and measurements can be performed in real time, without excising the tissue. This methodology provides a potentially important tool for guiding biopsy and performing wide area surveillance using fluorescence imaging techniques [4], [5]. Numerous studies have reported differences in the autofluorescence of normal and neoplastic tissues, over a wide range of excitation and emission wavelengths, both in vitro and in clinical settings [6]-[ 171. Initial in vitro [lo]-[13] and in vivo [14]-[17] LIF studies I. INTRODUCTION of colonic dysplasia have been promising. These studies OLONOSCOPY is the primary means of identifying examined adenomatous polyps as an example of dysplasia, and treating adenomatous' colonic polyps, which are a because these polyps are microarchitecturally identical to flat form of dysplasia? and are precursors to adeno~arcinoma.~ dysplasia and can be readily identified during colonoscopy. In Since these lesions are recognized by their raised structure, these investigations, excitation laser light delivery to tissue and colonoscopy is less effective in clinical situations in which return fluorescence collection, were implemented by means dysplastic changes are flat. Because areas of flat dysplasia are of an optical fiber probe which was passed through the Manuscript received February 23, 1995; revised October 16, 1995. This work was supported by the National Institute of Health (NIH) under Grant accessory channel of a colonoscope and was brought into R01 CA 53717 and was completed at the NM-supported (NIH P41-RR02594) contact with the tissue site under study. Fluorescence spectral Laser Biomedical Research Center at the Massachusetts Institute of Technol- features which differentiate normal mucosa and adenoma have ogy. Asterisk indicates corresponding author. G. I. Zonios, J. Wu, and M. S. Feld are with the George R. Harrison been identified and empirically correlated with histological diagnosis. Until now, the origins of these features have not Spectroscopy Laboratory, Massachusetts Institute of Technology, Cambridge, MA 02139 USA. been investigated. To our knowledge, the work presented here *R. Manoharan is with the George R. Harrison Spectroscopy Laboratory, Massachusetts Institute of Technology, Rm. 6-014, Cambridge, MA 02139 is the first to provide a detailed quantitative picture of the unUSA (e-mail:
[email protected]). derlying features of tissue fluorescence. Colon tissue contains R. M. Cothren and J. T. Arendt are with the Department of Biomedical various microscopic morphological structures which exhibit Engineering, The Cleveland Clinic Foundation, Cleveland, OH 44195 USA. fluorescence characteristic of their biochemical composition. J. Van Dam is with the Gastroenterology Division, Brigham & Women's Hospital, Boston, MA 02115 USA. The pathology of the tissue determines both the types of J. M. Crawford is with the Department of Pathology, Brigham & Women's microstructures present and their spatial distribution [ 181. It Hospital, Boston, MA 02115 USA. is important to identify the intrinsic and architectural factors Publisher Item Identifier S 0018-9294(96)01045-2. Adenoma: Dysplastic mucosal polypoid structure widely believed to be responsible for the clinically observed spectral differences precursor to adenocarcinoma. between normal tissue and adenomatous polyps. Once these Adenocarcinoma: Epithelial cancer with grandular growth patterns on a factors are identified, the potential of tissue fluorescence microscopic level. spectroscopy for diagnosing precancerous changes may be Dysplasia: Deranged development, characterized by proliferation and atypical cytologic alterations involving cell size, shape, and organization. assessed.
C
0018-9294/96$05.00 0 1996 IEEE
114
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996
Adenomatous Polyp
‘;I
Fig. 1. Layered structure and principal histologic components of normal, dysplastic, and adenomatous colon tissue. In normal colon, the top layer (mucosa) contains neatly arranged oval-shaped crypts which are packed with epithelial cells. In dysplastic tissue, the crypts become disorganized and irregular in shape and size. Dysplastic colon tissue may present a gross morphology similar to that of normal colon (flat dysplasia), or it may be characterized by additional gross morphological changes (adenomatous polyp). The structure surrounding the crypts is called the lamina propria. The submucosal layer begins approximately 450 pm below the tissue surface, it is mainly composed of collagen, and it contains no cells. Typical submucosal thickness is comparable to that of the mucosa. In this study, only the mucosal and submucosal layers were significant. The actual size of the polyps studied was much larger than the one shown here.
The objective of the present study was to develop a morphological model of colon tissue fluorescence which connects the features of the clinically observed spectra with their underlying biochemical and microstructural origins. The importance of this new approach in the analysis of tissue fluorescence lies in the fact that it contributes to the understanding of the tissue histology and microstructure which, in turn, is important for the early diagnosis of disease. We compared the results of this analysis to the data available from our previous clinical studies [ 161-[ 171. These studies employed 370-nm excitation light and collected fluorescence in the range 400-700 nm. Three main spectral features were observed: 1) both normal colon and colonic adenoma exhibited fluorescence intensity maxima at approximately 460 nm, 2) the fluorescence intensity of normal colon was larger than that of adenoma by a factor varying between two to nine times, and 3) adenoma exhibited increased fluorescence at wavelengths above 600 nm compared to that of normal tissue. Based on these spectral features, an empirical algorithm was developed, which was used to distinguish normal from adenoma. Very good diagnostic accuracy was obtained with 96% of the normal samples, and 90% of the adenoma samples classified correctly in a prospective manner [17]. This accuracy is comparable to that achieved by histologic examination. The first step in constructing a model was to identify the morphological structures which fluoresce. Fig. 1 illustrates the typical colon tissue morphology and gives definitions for the medical terminology used. The mucosa is the top layer consisting of tubular crypts lined with epithelial cells. The crypts are surrounded by lamina propria, a connective tissue structure mainly composed of collagen, which contains additional microstructures such as inflammatory cells, fibroblasts, and other biochemical structures. Beneath the mucosa there is the submucosa, a layer of connective tissue also primarily composed of collagen. Also shown in Fig. 1 are precancerous changes such as dysplasia and adenoma. Flu-
orescence photomicrography of normal colon tissue revealed that the lamina propria and the eosinophils (a type of white blood cell frequently found in tissue sections), are the only major fluorescing microstructures in the mucosal layer [18]. In adenomatous tissue, the same structures are found to fluoresce, with additional fluorescence from the dysplastic crypt cells. Finally, in both normal and adenoma, submucosa fluoresces quite strongly, compared to the mucosa. The model contains three components determined by tissue microstructure: 1) The intrinsic spectral lineshapes of each fluorescent microstructure, 2) the spatial distribution of the microstructural fluorescence intensities as a function of depth below the mucosal surface, and 3) the extent to which the excitation light and the return fluorescence are altered by scattering and absorption effects in tissue. The above quantities were measured by means of microspectrofluorimetry, quantitative microscopic imaging, and Monte-Carlo simulations of light propagation in tissue, respectively. The modeled spectra were obtained by combining these three quantities. U. MATERIALS AND METHODS
A. Tissue Specimens
The intrinsic fluorescence measurements (spectral lineshapes and density functions) were conducted on colon tissue sections prepared as follows: Tissue was obtained from colon resections which were performed to remove previously diagnosed invasive adenocarcinomas. Isolated adenomatous polyps 3-15 mm in diameter and histologically normal mucosa found in the resected colons were sampled. Sections for both microscopic imaging and microspectrofluorimetric studies were prepared in the same way. Each sample was frozen in OCT embedding-medium (International Equipment Company, Needham MA) at -2OOC and 5-pm-thick frozen sections were cut in a cryostat, perpendicular to the mucosal surface
~
115
ZONIOS et al.: MORPHOLOGICAL MODEL OF HUMAN COLON TISSUE FLUORESCENCE
comprising a complete cross section of mucosa, submucosa, and muscle wall. The fluorescence of sections from each tissue sample was studied, and adjacent serial sections to these studied were stained with hematoxylin and eosin (H&E) for histological examination. Sections used in the fluorescence studies were mounted on slides coated with polychloroprene, and they were cover-slipped under phosphate buffered saline (PBS) at pH 7.4. Polychloroprene was used to improve tissue adhesion and to facilitate removal of the coverslip so that the section could be stained and examined microscopically after spectroscopy. Polychloroprene, OCT, and PBS did not contribute to the observed tissue fluorescence, and there was no evidence that they altered the intrinsic fluorescence of the tissue samples studied. The remaining tissue blocks were stored, uncut, in an air-tight container at -80°C until used for experimental purposes. The measurement of colon tissue optical parameters (reduced scattering coefficient and absorption coefficient) was performed on separate normal and adenomatous polyp samples which were obtained from surgical resections in the same way as the samples used in the fluorescence studies. Samples were frozen in isopentane-cooled liquid nitrogen at - 17OOC and cut into 480-pm-thick sections perpendicular to the tissue surface using a microtome. In order to study the intrinsic fluorescence of fresh colon tissue and in particular the fluorescence of dysplastic crypt cells for reasons explained in Section 111-A, fresh specimens were prepared from polyps utilizing the cytology brushing technique [20]. Polyps removed during colonoscopy were sampled using a cytology brush. Specimens removed by the brush were placed on a polychloroprene coated microscope slide and cover-slipped under PBS saline. Specimens were studied within one hour of preparation.
excitation source fi
L
I
microscope
I
U
-
oeyepiece_1
I
Fig. 2. Experimental setup used for microscopic fluorescence spectral acquisition and microscopic fluorescence imaging. For spectral acquisition, the light source was an argon-ion laser at 364 nm and the detector a spectrograpNOMA system, while for microscopic imaging the light source was a xenon lamp and the detector a CCD camera.
Prior to spectroscopic measurement, the individual morphological structures to be studied were identified visually through the eyepiece of the microscope using white light transmitted through a dia-illumination port beneath the stage. The image of the collection aperture was superimposed on the sample's image to allow specific tissue morphological structures to be isolated for study. Identification of the morphological structures was confirmed by comparison with the corresponding structures in the H&E-stained serial section. Eosinophils which were difficult to locate in the H&E sections were identified by their small size and bright fluorescence [ 181.
B. Microspectrofluorimetry System
The microspectrofluorimeter (Fig. 2) consisted of a LeitzOrthoplan fluorescence microscope coupled to a laser source, a spectrograph, and an optical multichannel analyzer (OMA) [21]. The UV beam from an argon-ion laser at 363.8 nm was transmitted through a 200-pm core, 0.2-NA optical fiber and was collimated before it entered the epi-illumination port of the microscope. This excitation beam was reflected by a dichroic mirror and was delivered to the sample through a 40X objective (NA= 0.65). The resulting fluorescence was collected by the objective, passed through the dichroic mirror, and reimaged so that it could be viewed visually, photographed, or have its spectrum recorded using the spectrograph/OMA system. Variable collection and excitation microscope apertures ranging from 3 x 3 pm to 15 x 15 pm were used, depending on the fluorescence structure under study. A 380-nm long-pass barrier filter in front of the detector was used to block the excitation light. Wavelengths were assigned to individual diodes in the OMA diode array by means of a miniature pen-style mercury lamp, and the spectral response of the system was corrected using a standard NIST-traceable calibrated tungsten white light source. The spectral resolution of the spectrograph/OMA system was approximately 4 nm.
C. Fluorescence Microscopy Imaging System
The quantitative imaging studies employed a setup similar to the one used for microspectrofluorimetry(Fig. 2). Excitation light was provided by a 75-W short-arc xenon lamp fitted with a 380-nm narrow-band interference filter and delivered to an Olympus IMT-2 inverted microscope by an incoherent quartz optical fiber bundle. Light was directed through the microscope objective to the sample by means of a UV dichroic mirror. The resulting fluorescence was collected by the objective, directed through a long-pass barrier filter, and imaged onto the active area of a Princeton Instruments thermoelectrically-cooledCCD camera operated at -4OOC. Images were collected at 16-b dynamic resolution with a typical SNR of 60: 1, obtained using an integration time of 210 s. The 518 x 384 pixel images provided 1.2-pm resolution at 20x magnification. The tissue sections were oriented so that the mucosal surface was located perpendicular to the horizontal axis of the camera and then a phase-contrast reference image was acquired as follows. A uniform, weakly fluorescent 100-pmthick sample of doped Plexiglas was placed on the microscope stage, and images were acquired in five different orientations (usually by rotating the sample and displacing it slightly). A
EEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996
116
background image was also acquired at the same exposure conditions (exposure time, etc.) with no sample on the stage. After subtracting the background from each, the five images were averaged pixel-by-pixel to produce the reference image. Following, two fluorescence images were acquired under 380nm illumination using 405-nm and 455-nm long-pass barrier filters and 20x magnification. Images were also acquired from a blank microscope slide with a cover slip to allow for background correction of the camera dark charge. Images were processed and analyzed using the Khoros software system with several custom extensions. Each fluorescence image was background and flat-field corrected by subtracting the background image and dividing by the reference image. The 455-nm-barrier image was appropriately scaled and subtracted from the 405-nm-banier image to produce a single image containing fluorescence from a 50-nm-wide band centered at 430 nm. This technique allowed for quantitative correlation of intensities between the imaging and microspectrofluorimetry systems without requiring that the CCD camera be calibrated for its spectral response.
-t
mucosa1 surface
~
depth, 2
D. Tissue Fluorescence Model
Tissue was modeled as a turbid medium with embedded fluorescent microstructures. The tissue matrix was characterized by the absorption and scattering coefficients pa and ps and the scattering anisotropy coefficient g. Each individual microstructure was labeled by the index i and had intrinsic jhorescence $z (A). The fluorescence intensity spatial distribution of a certain microstructure was given by the JEuorescence density function D z ( z ) , with z being the depth from the tissue surface. Tissue turbidity was incorporated by means of the transfer function T(A,, A, z ) , defined below, which was calculated using Monte-Carlo simulations of light propagation in tissue. The intrinsic fluorescence q51(X) was normalized at 430 nm so that all fluorescence relative intensity information was contained in the density function Di ( 2 ) . The model calculated the bulk fluorescence spectrum, F ( X ) , as it is observed in vivo, in terms of $z(A), Ox(.) and T(A,, A , z )
Fig. 3. Intermediate steps in the fluorescence density function calculation showing the original fluorescence image, the structure mask, the masked image, and the final fluorescence density function, calculated as a function of depth z m tissue.
a specific structure. The mask is then correlated with the original image to produce a masked image, which is a twodimensional map of the fluorescence intensity distribution of the structure under study. The masked image pixel intensities are then averaged by integrating along the direction parallel to the tissue surface to produce the fluorescence density function as a function of depth in tissue. F. Tissue Optical Parameter Measurement
Z
J
The constant k is an overall normalization coefficient depending on the input light intensity, the efficiency of the light deliverykollection system, and the fluorescence quantum yield of tissue. This study is concerned with the relative intensity
of normal and adenomatous colon tissue fluorescence, and no attempt was made to evaluate k . E. Fluorescence Density Function
The fluorescence density function Di ( 2 ) represents the fluorescence intensity of a specific microstructure at 430 nm as a function of depth in tissue. Fig. 3 shows the individual steps in its calculation. First, the 430-nm band-pass fluorescence image was segmented using a directed version of Kittler’s optimal thresholding algorithm [22] in order to produce a structure mask. This mask contains all pixels belonging to
The transmission and reflection spectra of 450-pm-thick samples of normal mucosa and adenomatous polyps were measured in the range 300-700 nm using a Shimadzu UV265 spectrophotometer equipped with an integrating sphere (ISR-260, Shimadju Corp., Kyoto, Japan). The tissue scattering and absorption coefficients were calculated by computing the Kubelka-Munk coefficients [23] and transforming these into transport theory absorption and reduced scattering coefficients, pa and p: respectively [24]. The scattering coefficient ps was computed from p*.’,by means of the relation p s = pL(1 - g) where g is the scattering anisotropy coefficient (0 < g < I). We used g = 0.9, which is typical for tissues with histology similar to colon [25]. G. Transfer Function The transfer function accounts for scattering and absorption of excitation and fluorescence light at wavelengths A, and
ZONIOS et al.: MORPHOLOGICAL MODEL OF HUMAN COLON TISSUE FLSJORESCENCE
A, respectively. It also accounts for the specific light deliverykollection geometry defined by the optical probe used. Consider the fluorescence emanating from a layer of thickness dz located at depth z below the surface of a semi-infinite slab of tissue. The quantity T(A,, A, 2) dz is a measure of the fluorescence light arising from this layer, as it is collected at the tissue surface by the optical fiber probe. Monte-Carlo numerical simulations [26] were used to calculate the transfer function for normal and adenomatous tissues. Normal colon was assumed to be composed of two layers, corresponding to the mucosa (0450) pm and submucosa (450-900 pm). Tissue layers below the submucosa were ignored on the basis of the penetration depth of 370-nm excitation light in normal colon being about 400 pm only. Adenoma was modeled as a single mucosal layer of infinite thickness, because polyp mucosa is typically much thicker than the penetration depth of the excitation light, so that submucosal fluorescence arriving at the adenomatous tissue surface is negligible. The simulations trace the paths of incident excitation photons and the paths of the fluorescence photons as they traverse the tissue through several scattering and absorption events. Input parameters are the tissue transport coefficients ( p s ,pa,g), the tissue fluorescence quantum yield e, and the quantities defining the excitation and emission geometry (probe position relative to tissue surface, diameter, numerical aperture (NA), index of refraction). The transfer function was found to be relatively insensitive to the exact value of g in the range g = 0.75 - 0.95. The optical properties of the submucosa were assumed to be similar to the mucosal ones. The calculations modeled the excitatiodcollection geometry which was used in the in vivo study [17]. The optical fiber probe was composed of a central 0.22-NA, 200-pm core diameter excitation optical fiber surrounded by nine concentric 100-pm core diameter 0.22-NA collection fibers. The probe tip was formed by a cylindrical fused-silica optical shield [271 approximately 2.5-mm long, which was in contact with the tissue. In the simulations, a 2.5-mm-thick transparent layer of material with refractive index n = 1.5 was placed between the fiber tips and the tissue surface to simulate the shield. Uniformly distributed light in the range 0-loo was launched from a 200-pm-diameter central spot, and return fluorescence was collected over this same range of angles in a concentric ring of outer and inner diameters 400 and 200 pm, respectively. A tissue refractive index of n = 1.4 was assumed [26].
m 400
500
600
117
700
Wavelength (nm) Fig. 4. Typical intrinsic fluorescence spectra of the various morphological fluorescent structures found in colon tissue, (a) normal lamina propria, (b) adenomatous lamina propria, (c) eosinophil (measured on normal sample), (d) normal submucosa, (e), (f) dysplastic crypt cell. All spectra have been measured on frozen tissue sections, except of the dysplastic crypt cell spectrum in (f) which is measured on fresh tissue obtained using cytology brushing.
normal and adenomatous tissue [Fig. 4(c)]. Fig. 4(d) shows a typical spectrum of normal submucosal collagen with the characteristic collagen fluorescence peak near 430 nm. A typical dysplastic crypt cell spectrum is shown in Fig. 4(e), exhibiting two distinct peaks, one near 440 nm and one near 520 nm. The peak position of the lamina propria spectra was found to vary between 410 and 440 nm in normal and between 420 and 440 nm in adenoma. Eosinophil spectra showed limited variability. In contrast, dysplastic cell spectra were characterized by wide variations in the relative intensities of the two spectral peaks at 440 and 520 nm. Crypt cells did not exhibit measurable fluorescence in normal colon [18]. The 440-nm and 520-nm peaks observed in the crypt 111. RESULTS cell spectra are attributed to NADH [28] and flavins [29], respectively. The concentration of these chemical substances A. Intrinsic Fluorescence Spectra in cells is affected by the metabolism and the viability of the cells. Cellular flavin fluorescence is in general much weaker Fig. 4 shows typical intrinsic fluorescence spectra &(A) of the microstructures found in normal and adenomatous than cellular NADH fluorescence [30], so that the anticipated colon tissue based on measurements made on 14 normal 520-nm peak is much smaller than the 440-nm one. This and 14 adenomatous samples. Spectra collected from normal observation, together with the fact that there was no evidence and adenomatous lamina propria [Fig. 4(a) and (b)] exhib- for the 520-nm peak in the in vivo polyp spectra, suggested ited similar spectra, characteristic of collagen fluorescence that the 520-nm peak may be an artifact due to tissue section [19], with peak intensities around 420 nm. Eosinophil spectra preparation which involves cutting, as well as freezing and exhibited a broad lineshape peaking near 520 nm in both thawing of the tissue. To test this hypothesis we studied the
EEE TRAIYSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996
118
0.1 L
I
t
submucosa because this is located too far from the tissue surface and does not contribute to the total fluorescence. The eosinophil density function was very low compared to the rest of the microstructures, in both normal and adenomatous tissue [Figs 5(b) and (c)]. The fluorescence density function sample-to-sample variations were in general of the order of 2&30% max. for all microstructures.
/---I
C. Tissue Optical Parameters
L
dysplastic crypt 4 1 s eosinophils
fluorescence spectra of dysplastic crypts cells obtained from fresh polyp samples using cytology brushing. Fig. 4(f) shows a typical spectrum obtained, which does not contain the 520nm peak and contains additional red fluorescence (small peak in 650-700 nm region) which is also observed in the in vivo data [17]. Dysplastic crypt cell spectra obtained in this way using the cytology brushing method were used in producing the modeling results presented below, while crypt cell spectra measured on frozen sections were dismissed as being distorted by tissue sample preparation artifacts.
The average scattering and absorption coefficients from seven normal and seven adenomatous colon samples are shown in Fig. 6. Peaks at 420, 540, and 580 nm in the absorption curves are characteristic of hemoglobin absorption bands. The bumps in the scattering coefficient curves corresponding to the absorption bands of hemoglobin are probably due to inaccuracies of the method used in calculating them [24]. The adenomatous tissue absorption coefficient was approximately twice that of normal, while the scattering coefficient was less than that of normal. D. Transfer Function
Fig. 7(a) shows the transfer functions for normal and adenomatous tissue integrated over depth z from the tissue surface. The characteristic dips at 420, 540, and 580 nm are indicative of hemoglobin absorption. The integrated transfer function for adenomatous tissue was weaker than that of normal, due to the increased hemoglobin concentration in adenomatous polyps [Fig. 6(b)]. Fig. 7(b) shows the transfer function dependence on depth x, calculated at X = 460 nm. Most of the fluorescence signal comes from the top 450pm tissue layer (mucosa). This limits the strong submucosal fluorescence contribution in normal tissue, which arises below the top 450 pm. The transfer function in adenoma was weaker compared to that of normal, indicating an overall absorption increase in adenoma. E. Modeled Spectra
The in vivo spectra of normal and adenomatous colon were modeled by combining the quantitative information provided by the intrinsic fluorescence, the fluorescence density, and the transfer function, according to (1). Fig. 8 shows the B. Fluorescence Density Functions computed spectra of normal and adenomatous colon tissue Fig. 5 shows the average fluorescence density functions, versus the corresponding average spectra collected in vivo. D, ( z ), obtained from 11 normal and 11 adenomatous samples. The peak intensity of the in vivo spectra has been scaled in The fluorescence of normal colonic tissue was dominated by order to facilitate the comparison with the calculated spectra. lamina propria1 and submucosal fluorescence, with the submu- The constant k in (1) was not evaluated since we were only cosa fluorescing stronger by a factor of approximately eight interested in the relative fluorescence intensity differences [Fig. 5(a)]. The average mucosal depth was approximately between normal and adenoma. 450 pm which is comparable to the penetration depth of the The modeled spectra contain all the characteristic spectral features observed in the clinical data. The intrinsic fluores370-nm excitation light used in the clinical studies. Lamina proprial fluorescence was also predominant in ade- cence peak position of lamina propria, submucosa, and crypt nomatous tissue [Fig. 5(c)], with intensity reduced to approxi- cell spectra, typically in the range 400-430 nm which is shifted mately half that of normal tissue, probably due to the crowding to longer wavelengths around 460 nm in the in vivo spectra, of dysplastic crypts. Although of lower intensity, dysplastic was very well predicted. The peak intensity ratio between norcrypt cells contributed significantly to the total fluorescence mal and adenoma was approximately four compared to a range intensity due to the relatively large volume they occupy in the of 2-9 observed in vivo. The adenomahormal spectral ratio mucosa. No density function was calculated for adenomatous is shown in Fig. 9 illustrating a primary spectral difference
ZONIOS et al.: MORPHOLOGICAL MODEL OF HUMAN COLON TISSUE FLUORESCENCE
119
normal mucosa
+ ?
normal
100
15
adenomatous mucosa N,
i= d n o r m a l submucosa
I
300
I I
I 400
I
I I
I I
500
I 600
I
I
I I
600
500
700
1
0
Wavelength (nm)
(a)
"" h =460nm
0 Wavelength (nm)
depth, (rm)
(b)
CO)
Fig. 6. Colon tissue optical parameters as a function of wavelength A, (a) scattering coefficient ps(X),(b) absorption coefficient pa(A).Average values for measurements performed on several different pieces of tissue are shown here. Typical standard deviations were in the range k30-50%.
Fig. 7. Colon tissue transfer functions, (a) integrated over depth z for normal mucosa, (0-450 pm), normal submucosa (450-900 pm), and adenomatous mucosa (infinite thickness), as a function of emission wavelength A, (b) as a function of depth z for emission wavelength X = 460 nm.
between normal and adenoma in the wavelength range above 600 nm where the adenomatous spectrum contains increased fluorescence. This spectral feature was also very well predicted by our model. The model was also used to predict the fluorescence spectra of "flat" dysplastic lesions. In this case, the lesion was assumed to have the gross architecture of normal tissue (i.e., normal mucosal thickness), but the intrinsic fluorescence, microarchitecture and optical parameters of adenoma [31]. The computed spectra, obtained by assuming a 450-pm-thick mucosal layer with underlying submucosa, is shown in Fig. 10. This result predicts that fluorescence from flat dysplasia is less intense than normal colon tissue fluorescence, but more intense than adenomatous polyp fluorescence. The spectrum exhibits a rather sharp rising slope in the 420-460 nm range which is due to the increased hemoglobin absorption. IV. DISCUSSION
We have successfully developed a methodology for modeling clinical fluorescence spectra of adenomatous and normal colonic tissue, based on microscopic fluorescence analysis. The model predicts the features of the clinical spectra and provides a basis for understanding their morphological and
biochemical origins. This establishes a link between the observed spectra and the underlying biological information. The computed spectra exhibit the three characteristic features found in the in vivo spectra. The interpretation of these features according to our analysis is as follows: 1) For both normal and adenoma, the peak intensity occurs at 460 nm, even though the intrinsic fluorescence of collagen, which is the dominant fluorophore, peaks at around 420 nm. This shift is due to the effect of the large hemoglobin absorption band at 420 nm, accounted for in the transfer function. 2) The intensity of adenoma is smaller than that of normal colon. This is due to three factors: a) the mucosal collagen fluorescence is decreased in adenoma due to enlargement of the crypts, which displaces the lamina propria, b) adenomatous tissue exhibits increased absorption, due to increased hemoglobin content, and c) the submucosa contributes to the fluorescence in normal tissue, but not in adenoma, because of the increased mucosal thickness of a polyp. Approximately one-half of the total peak intensity difference is due to the submucosal contribution in normal which is absent in the adenoma.
120
Wavelength (nm)
Fig. 8. Calculated spectra for normal and adenoma vs. the average in vivo spectra. Only selected data points were calculated for the modeled spectra because of the numerical computer time-intensive method used (see Section 11-G). The experimental data consist of more than 400 data points so they are best shown here as a continuous line (same convention also used in Fig. 9).
lI
0 model
-clinical data
Wavelength (nm)
Fig. 9. Calculated adenomdnormal spectral ratio (selected wavelengths) compared to average clinical adenomdnormal spectral ratio.
3) Red fluorescence is increased in adenoma. This addi-
tional red fluorescence is primarily associated with the intrinsic fluorescence of the dysplastic crypt cells. The above model can be compared to a previous model developed by Schomacker et al. [15] to describe colon tissue fluorescence excited using 337-nm light. In this model, the observed spectra were assumed to be a linear combination of biochemicals thought to be present in the tissue. The fluorescence lineshapes of commercially available compounds were used to construct the modeled spectra. The distribution of fluorophores within the tissue and their in situ lineshapes were not taken into account, while tissue absorption and scattering effects were also not considered. This study concluded that the decrease in fluorescence intensity in adenomatous polyps
Wavelength (nm)
Fig. 10. Calculated fluorescence spectrum of flat dysplasia compared to the average normal and adenoma clinical spectra.
was entirely due to polyp architecture. In contrast, the present study which utilized 370-nm excitation, indicates that the observed spectral differences are due to a combination of factors related to intrinsic, biochemical, and microarchitectural tissue features, as well as polyp architecture. This result is important, in the sense that it opens the prospect of detecting nonpolypoid dysplastic lesions using fluorescence spectroscopy. The results of this study suggest several directions for future investigations aimed at improving the technique. The spectral features differentiating normal and adenomatous tissue can be enhanced in a number of ways. The increased red fluorescence in dysplastic tissue may be due to hemoglobin breakdown products such as porphyrin derivatives, which fluoresce in the range 600-700 nm and are known to accumulate in tumor cells [32]. It is possible to map the excitation-emission properties of the dysplastic crypt cells to investigate this assumption and perhaps select an excitation wavelength which further enhances red fluorescence. Optical probe design can also be improved to modify the contribution of the strongly fluorescing submucosa. The effective penetration and collection depth of light in tissue can be controlled by varying the optical fiber numerical aperture, as well as the distance between the excitation and collection fibers. By optimizing the probe to collect preferentially from the mucosa only, nondiagnostic information from the submucosa may be minimized. Other issues to be investigated include the validation of the in vitro measurements used in the model, as well as a more accurate measurement of the tissue optical parameters which are usually affected by various types of errors [25]. The model developed here connects the observed clinical fluorescence spectra with the microstructural properties of the tissue. Crypt volume, hemoglobin content, degree of disease, and mucosal thickness are all reflected in the fluorescence spectra [33]. By applying the model to the analysis of individual spectra, it may be possible to extract information related to these parameters. This could provide valuable histopathological information which is presently not available without biopsy.
ZONIOS et al.: MORPHOLOGICAL MODEL OF HUMAN COLON TISSUE FLUORESCENCE
In conclusion, we have developed a model which described Observed tissue fluorescence in Of the microscopic features of the tissue. This model enabled us to quantify the intrinsic and architectural factors responsible for fluorescence differences in normal tissue and adenoma. The results suggest that nonpolypoid dysplasia may also be distinguishable from normal. This method of analysis should be useful in employing fluorescence spectroscopy for guiding biopsy and, eventually, in SUrVeilhnCe Of large areas Of mucosa for the detection of dysplasia. Studies in this direction are in progress in our laboratories.
REFERENCES J. E. Lennard-Jones, J. K. Ritchie, B. C. Morson, and C. B. Williams, “Cancer surveillance in ulcerative colitis: Experience over 15 years,” Lancet, pp. 149-152, 1983. E. Rosenstock, R. G. Farmer, R. E. Petras, M. V. Sivak, G. B. Rankin, and B. H. Sullivan, “Surveillance for colonic carcinoma in ulcerative colitis,” Gust. End., vol. 89, pp. 1342-1346, 1985. M. Karita, D. Cantero, and K. Okita, “Endoscopic diagnosis and resection treatment for flat adenoma with severe dysplasia,” Amer. J. Gastroenterol., vol. 88, pp. 1421-1423, 1993. P. S. Anderson, S. Montan, and S. Svanberg, “Multispectral system for medical fluorescence imaging,” ZEEE J. Quantum Electron., vol. 23, pp. 1798-1805, 1987. S. Lam, C. MacAulay, J. Hung, J. LeRiche, A. E. Profio, and B. Palcic, “Detection of dysplasia and carcinoma in situ with a lung imaging fluorescence endoscope device,” J. Thoracic and Cardiovasc. Surg., vol. 105, pp. 1035-1040, 1993. R. R. Alfano, G. C. Tang, A. Pradhan, W. Lam, D. S. J. Choy, and E. Opher, “Fluorescence spectra from cancerous and normal human breast and lung tissue,” ZEEE J. Quantum Electron., vol. 23, pp. 1806-1811, 1987. Y. Yuanlong, Y. Yanmin, L. Fuming, K. Yufen, and M. Paozhong, “Characteristic autofluorescence for cancer diagnosis and its origin,” Lasers Surg. Med., vol. 7, pp. 528-532, 1987. J. Hung, S. Lam, J. C. LeRiche, and B. Palcic, “Autofluorescence of normal and malignant bronchial tissue,” Lasers Surg. Med., vol. 11, pp. 99-105, 1991. W. S. Glassman, C. H. Liu, G. C. Tang, S. Lubicz, and R. R. Alfano, “Ultraviolet excited fluorescence spectra from non-malignant and malignant tissues of the gynecological - tract,” Lasers Life Sci., vol. -. 5, pp. 46-58, 1992. P. N. Yakshe, R. F. Bonner, P. Cohen, M. B. Leon, and D. E. Fleischer, “Laser induced fluorescence spectroscopy may distinguish colon cancer from normal human colon,” Gastrointest. Endoscopy, vol. 35, p. 184, 1989. C. R. Kapadia, F. W. Cutruzolla, K. M. O’Brian, M. L. Stetz, R. Enriquez, and L. I. Deckelbaum, “LIF spectroscopy of human colonic mucosa,” Gastroenterol., vol. 99, pp. 150-157, 1990. R. R. Richards-Cortum, R. P. Rava, R. E. Petras, M. Fitzmaurice, M. Sivak, and M. S. Feld, “Spectroscopic diagnosis of colonic dysplasia,” Photochem. Photobiol., vol. 53, pp. 777-786, 1991. R. Marchesini, M. Brambilla, E. Pigoli, G. Bottiroli, A. C. Croce et al., “Light-induced fluorescence spectroscopy of adenoma, adenocarcinomas and nonneoplastic mucosa in human colon. I: Zn vitro measurements,” J. Photochem. Photobiol., B: Biol., vol. 14, pp. 219-230, 1992. K. T. Schomacker, K. J. Frisoli, C. C. Compton, T. J. Flotte, J. M. Richter, T. F. Deutsch, and N. S. Nishioka, “Ultraviolet laserinduced fluorescence of colonic polyps,’’ Gastroenterol., vol. 102, pp. 1155-1160, 1992. K. T. Schomacker, K. J. Frisoli, C. C. Compton, T. J. Flotte, J. M. Richter, N. S. Nishioka, and T. F. Deutsch, “Ultraviolet laser-induced fluorescence of colonic tissue: Basic biology and diagnostic potential,” Lasers Surg. Med., vol. 12, pp. 63-78, 1992. R. M. Cothren, R. Richards-Kortum, M. V. Sivak Jr., M. Fitzmaurice, R. P. Rava, G. A. Boyce, M. Doxtader, R. Blackman, T. B. Ivanc, G. B. Hayes, M. S. Feld, and R. E. Petras, “Gastrointestinal tissue diagnosis by laser-induced fluorescence spectroscopy at endoscopy,” Gastrointest. Endoscopy, vol. 36, pp. 105-1 11, 1990. R. M. Cothren, M. V. Sivak, J. Van Dam, R. E. Petras, M. Fitzmaurice, J. Crawford, J. Wu, J. F. Brennan, R. P. Rava, R. Manoharan, and
[18]
[NI
[20]
-r211-
1221 [23] [24] [25] [26] [27]
[28] [29] [30] [31] [32] [33]
121
M. S. Feld, “Detection of dysplasia at colonoscopy using laser-induced fluorescence: A blinded study,” Gastrointest. Endoscopy, in press, 1995. T. J. Romer, M. Fitzmaurice, R. M. & h e n , R. R. Richards-Cortum, M. S. Feld, R. Petras, M. V. Sivak Jr, and J. R. Kramer, “Laser Induced fluorescence microscopy of normal colon and dysplasia in colonic adenomas: Implications for spectroscopic diagnosis,” Am. J. Gastroenterol., vol. 90, pp. 81-87, 1995. G. Bottiroli, A. c. Croce, D. Locatelli, R. Marchesini, E. Pignoli, S. Tomatis, C. Cuzzoni, S. Di Palma, M. Delfante, and P. Spinelli, “Natural fluorescence of normal and neoplastic human colon: A comprehensive “ex vivo” study,” Lasers Surg. Med., vol. 16, pp. 48-60, 1995. J. T. Hanson, C. Thoreson, and J. F. Morrissey, “Bmsh cytology in the diagnosis of upper gastrointestinal malignancy,” Gastrointest. Endoscopy, vol. 26, pp. 33-35, 1980. J. J. Baraga, R. R. Rava, M. Fitzmaurice, L. Tong, P. Taroni, C. Kittrell, and M. S. Feld, “Characterization of the fluorescent morphological structures in human arterial wall using ultraviolet-excited microspectrofluorimetry,” Atherosclerosis, vol. 88, pp. 1-14, 1991. J. Kittler, J. Illingworth, “Minimum error thresholding,” Pattem Recog., vol. 19, pp. 4 1 4 7 , 1986. A. Ishimaru, Wave Propagation and Scattering in Random Media, Vol. 1. New York Academic, 1978, p. 191. M. J. C. Van Gemert, and W. M. Star, “Relations between the KubelkaMunk and the transport equation models for anisotropic scattering,” Lasers Life Sci., vol. 1, pp. 287-298, 1987. W. Cheong, S. A. Prahl, and A. J. Welch, “A review of optical properties of biological tissues,” ZEEE J. Quantum Electron., vol. 26, pp. 2166-2185, 1990. J. Wu, M. S. Feld, and R. P. Rava, “Analytical model for extracting intrinsic fluorescence in turbid media,” Appl. Opt., vol. 32, pp. 3585-3595, 1993. T.. F. Brennan, G. I. Zonios, T. D. Wang, R. P. Rava, G. B. Hayes, R. R. Dasari, and M. S. Feld, “Portable laser spectrofluorimeter system for in vivo human tissue fluorescence studies,’’ Appl. Spectroscopy, vol. 47, pp. 2081-2086, 1993. J. Eng, R. M. Lynch, and R. S. Balaban, “Nicotinamide adenine dinucleotide fluorescence spectroscopy and imaging of isolated cardiac myocytes,” Biophys. J., vol. 55, pp. 621-630, 1989. B. Chance, B. Schoener, R. Oshino, F. Itshak, and Y. Nakase, “Oxidation-reduction ratio studies of mitochondria in freeze-trapped samples,” J. Biol. Chem., vol. 254, pp. 47644771, 1979. J. E. Aubin, “Autofluorescence of viable cultured mammalian cells,” J. Histochem. Cytochem., vol. 27, pp. 3 6 4 3 , 1979. R. A. Wolber, D. A. Owen, “Flat adenomas of the colon,” Hum. Pathol., vol. 21, pp. 70-74, 1991. D. M. Harris, J. Werkhaven, “Endogenous porphyrin fluorescence in tumors,” Lasers Surg. Med., vol. 7, pp. 467472, 1987. R. Manoharan, G. Zonios, R. Cothren, J. Arendt, J. Van Dam, and M. S. Feld, “Laser-induced fluorescence spectroscopy of colonic dysplasia: Prospects for optical histologic analysis,” SPZE Advances Fluorescence Sensing Technol. ZZ, vol. 2388, pp. 417422, 1995.
George I. Zonios was born in Greece, in 1968. He received the B.S. degree in physics from the University of Ioannina, Greece, in 1990. He is currently completing the requirements for the Pb.D. degree in physics. From 1990 through 1992 he was a Research Assistant in the Laboratory for Nuclear Science at the Massachusetts Institute of Technology, where he worked in the design and implementation of a pionelectroproduction experiment to test tundamental aspects of nuclear theory. In 1992 he joined the M.I.T. George R. Harrison Spectroscopy Laboratory, Cambridge, MA, where he has been investigating human tissue fluorescence spectroscopy as a diagnostic tool in medicine His interests include optical applications in medicine and biology, biomedical instrumentation, and the physical processes and techniques applied in investigating the composition, structure, and functioning of the human body.
122
E E E TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996
Robert M. Cothren received the S.B. degree in 1981 and the M.S. degree in 1985 from the Department of Chemical Engineering at the Massachusetts Institute of Technology, Cambridge, and the Ph.D. degree in medical engineering from the Harvard-M.I.T. Health Sciences and Technology Program in 1987. He joined The Cleveland Clinic Foundation, Cleveland, OH in 1987, where he currently holds a Staff Appointment in the Department of Biomedical Engineering of the Research Institute. He directs an independent laboratory for in vivo tissue spectroscopy and is an active member of the Biomedical Image Processing group. His research interests include the use of fluorescence spectroscopy for the detection of precancerous changes in vivo, quantitative light and fluorescence microscopy, and automated analysis of clinical MR and intra-vascular ultrasound imaging. He helps direct highend computing and networking development within the department, and is active in graduate-student education through a relationship with The Ohio State University.
Joseph T. Arendt received the B.S. degree in electrical and computer engineering from The University of Wisconsin, Madison, in August 1986 and the M.S.E.E. degree from Ohio State University, Columbus, in May 1995. From 1986 to 1991, he was Assistant Staff Memher at M.I.T. Lincoln Laboratory in the Signature Analysis Group. Most of his work there involved analyzing signals from IR detectors. Since 1992, he has been conducting research at the Cleveland Clinic Foundation in tissue autofluorescence using spectroscopy and microscopy. He is using this research to pursue a doctorate in biomedical engineering from Ohio State University. U
Jun Wu was born in Beijing, China, in 1968, and received the B.S. degree in physics from Peking University, Beijing, China. He is currently pursuing the Ph.D. degree in medical engineering and medical physics from the Harvard-M.I.T. Division of Health Science and Technology He joined the M.I.T. George R. Harrison Spectroscopy Laboratory, Cambridge, MA, in 1990 and has been studying the photon migration approach to model the light propagation in biological tissues.
Jacques Van Dam was bom in Amersfoort, the Netherlands and emigrated to the United States in 1953. He received the B.A. degree from Rutgers University, Piscataway, NJ, in 1975, the M.S. degree from Hahnemann Medical College in 1981, and the M.D. and Ph.D. degrees from Georgetown University School of Medicine, Washington D.C., in 1984 and 1988, respectively. He completed his medical internship, residency, and clinical and research fellowships at Harvard Medical School, Cambridge, MA, at the New England Deaconess Hospital, Beth Israel Hospital, and the Massachusetts General Hospital, respectively. He is currently an Assistant Professor of Medicine at Harvard Medical School and the Director of Endoscopic Gastrointestinal Oncology and the Associate Director of Endoscopy at the Brigham and Women’s Hospital. He became a Visiting Scientist at the Laser Biomedical Research Laboratory of the Massachusetts Institute of Technology, Cambridge, MA, in 1992 and was named chairman of the Medical Advisory Committee for the LBRC in the same year. He is currently a Co-investigator (and Principal Investigator for the Brigham and Women’s Hospital) for an NIH RO1 Grant “Real Time In vivo Diagnosis of Dysplasia by Fluorescence.”
James M. Crawford is a Connecticut native and received the B.A. degree from DartmouthHanover, NH,in 1975, and the M.D. and Ph.D. degrees from Duke UniversityDurham, NC, in 1982. His graduate thesis, under Prof. J. J. Blum, involved mathematical modeling of intermediary metabolism in the liver. He trained in anatomic pathology at Brigham and Women’s Hospital, Boston, MA, specializing in gastro-intestinal and liver pathology. A postdoctoral fellowship, 1985 through 1987, was devoted to examining bilirubin trafficking through the liver. He is currently Assistant Professor of Pathology at Harvard Medical School, Cambridge, MA, and serves as Staff Pathologist at Brigham and Women’s Hospital. His basic research now focuses on mechanisms by which bile acids regulate hepatocellular bile formation, both at the level of intracellular membrane trafficking and secretion of organic solutes into bile. His long-time interests in mathematical analysis and gastrointestinal pathology are brought together through his collaborative participation in projects conducted through the Massachusetts Institute of Technology Laser Biomedical Center at the George R. Harrison Spectroscopy Laboratory.
Ramasamy Manoharan received the B.S. and M.S. degrees in chemstry from Madurai Kamaraj University, India, in 1982 and 1984, respectively. In 1989 he received the PhD. degree in physical chemstry from Indian Institute of Technoloy, Kanpur. From 1989 to 1990, he was a Postdoctoral Fellow in the Chemistry Department at the University of Rhode Island, Kmgston. His work there involved the application of UV resonance Raman spectroscopy for zn sztu characterization of biopolymers in bactenal cells and spores. In 1991, he joined the George R. Harrison Spectroscopy Laboratory at Massachusetts Institute of Technology, Cambridge, where he currently holds a Research Scientist appointment. He is the Group Leader of the Biomedical Spectroscopy Programs at the M.I.T. Laser B i o m e d d Research Center H s research interests are in the area of lasers and optics for biology and medicine. €Tis current projects include noninvasive detection and imaging of biological tissues using laser-induced fluorescence, IR absorption, near-infrared Raman and UV resonance Raman spectroscopic techniques. Dr. Manoharan was awarded the Alexander von Humboldt research fellowship from Germany in 1990.
Michael S. Feld (M’84) was bom in New York City in 1940. He received the S.B. degree in humanities and sciences and the S.M. degree in physics in 1963, and the Ph.D. degree in physics in 1967, all from Massachusetts lnstitute of Technology, Cambridge. He is a Professor of physics at M.I.T. and directs the George R. Harrison Spectroscopy Laboratory. His research interests are in the field of laser physics, especially the interaction of intense light fields with atomic and molecular systems, laser spectroscopy, superradiance, laser-nuclear physics, and medical applications of laser light. His publications include both theoretical and experimental topics. He has also done research in biological applications of lasers and in the history of science. In 1992, he was the Wolk Visitor and Lecturer at Colgate University. He is a Research Member of the Joint Faculty of the Harvard-M.I.T. Division of Health, Science and Technology, and an Adjunct Staff Member in the Department of Cardiovascular Research of the Cleveland Clinic Foundation. Dr. Feld is a Fellow of the American Physical Society, the Optical Society of America, the American Association for the Advancement of Science, the Society of Sigma Xi, and the American Association for Lasers in Surgery and Medicine. He was Alfred P. Sloan Research Fellow, 1973-1976; he received the M.I.T. Minority Community Distinguished Service Award in 1980, the Gordon Y. Billard Award in 1982, the Thompson Award in 1991 for the development of biomedical Raman spectroscopy, and the Vinci of Excellence (France), in 1995, for development of the single atom laser.
