WO2010025122A1 - Method for measuring disease probability - Google Patents

Abstract

Methods comprising providing a tissue sample, providing a first imaging device comprising a white light imaging device; providing a second imaging device comprising at least one imaging device selected from the group consisting of: a fluorescent imaging device, a narrowband reflectance imaging device, and a polarized reflectance imaging device; obtaining a first image of the tissue sample with the first imaging device, wherein the first image comprises a plurality of pixels; obtaining a second image of the tissue sample with the second imaging device, wherein the second image comprises a plurality of pixels; calculating a metric for at least one of the plurality of pixels of the second image; and calculating a disease probability using the metric calculated for the at least one of the plurality of pixels of the second image.

Description

METHOD FOR MEASURING DISEASE PROBABILITY

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims priority to U.S. Provisional Application No. 61/091 ,513, filed August 25, 2008, which is incorporated herein by reference. STATEMENT OF GOVERNMENT INTEREST

The present disclosure was developed with support under Grant Numbers R2 IDEl 6485, RO1CA103830, and P50CA097007, awarded by the National Institutes of Health. The U.S. government has certain rights in the invention.

BACKGROUND The present invention relates generally to imaging of biological tissue. In particular, the present invention relates to methods of imaging biological tissue for measuring disease probability.

Head and neck cancer, including cancers of the oral cavity, currently ranks as the sixth most common malignancy in the world. There were more than 270,000 new cases of oral cancer reported in 2002. Approximately 60% of these individuals present with stage III or IV disease, and about half will die within five years of diagnosis. Screening individuals at risk for oral cancer and its precursors has the potential to improve early detection, providing the opportunity to intervene when treatment is most effective. In addition, surveillance of patients who have survived their initial oral cancer is important to identify local recurrences and second primary oral tumors, which occur at a higher reported rate than for any other tumor.

Conventional oral examination using incandescent white light is the current standard of care for screening and surveillance for oral cancer and precancerous lesions. The sensitivity of visual examination is limited by several factors including the experience and index of suspicion of the examiners. In primary care situations, cases of malignancy may be seen rarely, and clinicians may have difficulty discriminating the sometimes subtle mucosal changes associated with premalignant lesions and early cancers from more common benign inflammatory conditions. Furthermore, it can be difficult to visualize the boundaries of neoplastic lesions using conventional oral examination making the choice of a biopsy location difficult.

Several new approaches have been proposed to address the limitations of the conventional oral examination, including the use of toluidine blue, brush cytology, reflectance visualization after acetic acid application, and illumination with a chemi-luminescent light source. While useful in certain situations, each of these approaches is associated with a high rate of false-positives. Moreover, current fluorescence imaging devices rely on qualitative observations to detect and delineate neoplastic oral lesions and therefore reliable screening with these instruments necessitates well-defined and standardized image interpretation criteria, and appropriate user training. This may not be feasible in many primary care situations. SUMMARY

In general, the present disclosure, is related to the application of digital image processing techniques to autofluorescence imaging of tissue to objectively identify and delineate the peripheral extent of neoplastic lesions. Among other things, this will provide a powerful tool in patient care locations where experts are not available or where physicians encounter few cases of malignant and pre-malignant neoplasia. Low-cost digital cameras with sufficient sensitivity to record tissue autofluorescence in near real time are now readily available, making clinical application of such automated image processing now feasible.

In certain embodiments, the methods of the present disclosure can delineate the presence and extent of neoplastic lesions within a field of view and provide results which correlate with the histopathologic assessment of extent of disease. Thus, quantitative autofluorescence imaging, according to certain embodiments of the present disclosure, may provide a non-invasive and objective method to improve screening and margin delineation of oral cancers and precancers.

DRAWINGS

Some specific example embodiments of the disclosure may be understood by referring, in part, to the following description and the accompanying drawings.

Figure 1 shows autofluorescence and white light images of the buccal mucosa of a typical study patient. A. White light image showing regions of interest of histopathologically confirmed normal tissue and invasive carcinoma. B. Fluorescence image at 365 nm excitation. C. Fluorescence image at 405 nm excitation. D. Fluorescence image at 450 nm excitation. Figure 2 shows A. Scatter plot of normalized red-to-green ratios at 405 nm excitation for the 102 ROI sites in the training set. The horizontal line indicates the threshold used to obtain 95.9% sensitivity and 96.2% specificity. Note that 2 additional abnormal data points had a red- to-green fluorescence intensity ratio greater than 3 but are not shown on this plot. B. Receiver- operating characteristic (ROC) curve of the classifier based on the normalized red-to-green ratio. The operating point used for classification is indicated by a dot and arrow. C. Scatter plot of the red-to-green ratio for the 57 sites in the validation set with threshold selected from the training set indicated. Note that 3 additional abnormal data points had a red-to-green fluorescence intensity ratio greater than 3 but are not shown on this plot. D. ROC curve obtained for the validation set. The operating point is indicated and corresponds to the threshold chosen from the training set.

Figure 3 shows A. White light image of floor of mouth with histopathologically confirmed dysplasia and carcinoma in situ. B. 405 nm excitation fluorescence image showing areas with deceased autofluorescence. C. White light image with disease probability map showing the predictive probability of a neoplastic lesion superimposed. Letters indicate specific locations were pathology is known. The key to the right of C. indicates pathology. The histology slides below show tissue sections from these areas. Marking bar at the lower right- hand comer = 1 mm. Figure 4 A. and B. show images from a patient with an invasive carcinoma in the floor of mouth. A. White light image B. White light image with disease probability mapping showing the predictive probability of a neoplastic lesion. C. and D. show images from a patient with a region of severe dysplasia on the tongue. E and F show images from a patient with a region of moderate dysplasia on the gingiva. G. and H. show images from inner lip of a normal volunteer. While the present disclosure is susceptible to various modifications and alternative forms, specific example embodiments have been shown in the figures and are herein described in more detail. It should be understood, however, that the description of specific example embodiments is not intended to limit the invention to the particular forms disclosed, but on the contrary, this disclosure is to cover all modifications and equivalents as illustrated, in part, by the appended claims.

DESCRIPTION

The present invention relates generally to imaging of biological tissue. In particular, the present invention relates to methods of imaging biological tissue for measuring disease probability. While much of the description and examples herein pertains to the imaging of oral tissue in humans, such is not intended to limit the scope of the present invention. Rather, the methods disclosed herein are applicable to a variety of subjects and tissue types.

The present disclosure provides, in certain embodiments, a method comprising providing a tissue sample; providing a first imaging device comprising a white light imaging device; providing a second imaging device comprising at least one imaging device selected from the group consisting of: a fluorescent imaging device, a narrowband reflectance imaging device, and a polarized reflectance imaging device; obtaining a first image of the tissue sample with the first imaging device, wherein the first image comprises a plurality of pixels; obtaining a second image of the tissue sample with the second imaging device, wherein the second image comprises a plurality of pixels; calculating a metric for at least one of the plurality of pixels of the second image; and calculating a disease probability using the metric calculated for the at least one of the plurality of pixels of the second image.

The tissue sample useful in the methods of the present invention may be any tissue sample suitable for imaging with the imaging devices useful in the methods of the present invention. In certain embodiments, such tissue samples may be human tissue. In certain embodiments, such human tissue may be from the oral cavity. In certain embodiments, such tissue may be abnormal tissue, such as precancerous or cancerous tissue. In other embodiments, such tissue may be abnormal tissue, such as from an early invasive disease. In other embodiments the tissue may be from a benign condition or from inflammation While the methods of the present invention, in certain embodiments, are capable of use without removal of the tissue from the subject, tissue samples removed from subjects, such as biopsies, may still be useful in the methods of the present invention, provided such removal does not render the tissue incapable of being imaged with the imaging devices useful in the methods of the present invention.

The first imaging devices useful in the methods of the present invention generally comprise a white light imaging device. Any suitable white light imaging device may be used in the methods of the present invention. In certain embodiments, such first imaging devices use commonly available broadband light sources, such as those in commercially available digital cameras. The choice of a suitable first imaging device may depend upon, among other things, the type and/or location of the tissue to be imaged.

The second imaging devices useful in the methods of the present invention include, but are not limited to, fluorescent imaging devices, narrowband reflectance imaging devices, and polarized reflectance imaging devices. An example of a suitable second imaging device is a multispectral digital microscope (MDM). In certain embodiments, the fluorescent imaging devices useful as second imaging devices in the methods of the present invention may be capable of producing fluorescent light with excitation wavelengths of one or more of 365 nm, 405 nm, and 450 nm. The choice of a suitable second imaging device may depend upon, among other things, the type and/or location of the tissue to be imaged. When obtaining images with the first and second imaging devices, it is preferred to align each device such that the captured images from each device are of substantially the same portion of the tissue sample. In certain embodiments, imaging processing techniques known to one of ordinary skill in the art may be used after the images are obtained to eliminate portions of the obtained images outside the portion of the tissue sample which is desired to be imaged. Furthermore, other image processing known to one of ordinary skill in the art may be used to account for any movement of the tissue sample during the imaging steps.

The resulting images comprise a plurality of pixels, and a variety of metrics may be calculated for at least one of the plurality of pixels. The desired metric may be known or easily determined by one of ordinary skill in the art, with the benefit of the present disclosure. Useful metrics may include, but are not limited to, the ratio of red pixel values to green pixel values, the ratio of red pixel values to blue pixel values, the ratio of green pixel values to blue pixel values. Such a metric may be calculated for any number of pixels in the image, including a single pixel, a plurality of pixels which make up the image of a region of particular interest in the tissue sample, and all of the pixels in the image.

After calculating a metric for at least one of the pixels in the second image, the metric(s) may be used to determine, among other things, a disease probability at the location of the tissue to which the individual pixel(s) correspond. Such a determination may be made by gathering a set of metrics from images of known abnormal (such as precancerous or cancerous) tissue and known normal tissue and comparing the values of the calculated metrics. A useful metric(s) may be found by demonstrating that such useful metric(s) are correlated with regions of abnormal tissue or regions of normal tissue. Such a metric may then be used in the methods of the present invention to determine the probability that individual portions of a tissue sample, represented by one or more pixels, and thus one or more values of the metric, are abnormal or normal. As with the metric calculations, such a determination may be made for any number of pixels in the image, including a single pixel, a plurality of pixels which make up the image of a region of particular interest in the tissue sample, and all of the pixels in the image.

In certain embodiments, the probability determination for one or more pixels of the image may be represented by a color-coded image. Such an image may be overlaid with the first image, obtained from a white light source, to provide a color-coded image of the tissue sample, showing the probability of certain regions(s) of the tissue sample being normal or abnormal. For example, in certain embodiments, quantitative fluorescence imaging devices that can show false color disease-probability maps based on red/green fluorescence intensity ratios at 405 nm excitation at the time of the examination may be used.

To facilitate a better understanding of the present invention, the following examples of specific embodiments are given. In no way should the following examples be read to limit or define the entire scope of the invention. EXAMPLES Methods Imaging Procedure

Autofluorescence images were obtained from the oral cavity of 56 patients with clinically abnormal lesions and 11 normal volunteers. Data were divided into a training set and a validation set. Data acquired from the first 39 patients and 7 normal volunteers imaged between June 2006 and January 2008 were allocated to the training set and used to develop an algorithm for detection of neoplasia. Data acquired from the subsequent 17 patients and 4 normal volunteers imaged between March and June 2008 formed a validation set and used to test the performance of this algorithm relative to histopathology. The clinical protocol was reviewed and approved by the Internal Review Boards at the University of Texas MD Anderson Cancer Center and Rice

University. All subjects enrolled in the study gave written informed consent. Patients were eligible if they had known or suspected precancerous or cancerous squamous lesions located in the oral mucosa. Patients may have had previous surgical, radiation, or chemotherapeutic treatments.

White light and autofluorescence images were obtained at 365 ran, 380 run, 405 nm, and 450 nm excitation using a Multispectral Digital Microscope (MDM). This device is described in detail elsewhere [24], but briefly, the MDM is a wide-field optical microscope which collects digital autofluorescence and reflectance images with a color CCD camera from a variable field of view, ranging in size from approximately 1 to 7 cm. Patients were imaged either in an outpatient clinic or in the operating room under general anesthesia prior to surgery. A physician positioned the patient and microscope so that the suspicious lesion or area of interest was clearly in the field of view of the device. Clinically normal areas distant from or contralateral to the lesion were also imaged.

Following imaging in the clinic, suspicious lesions were biopsied. In the operating room, previously biopsied lesions were surgically resected.

Histopathologic Correlation

Biopsies and resected tissues were evaluated using standard histopathologic analysis by a board certified pathologist (AEN, MDW). The location of biopsies and resected lesions was recorded using digital photography so that pathology results could later be correlated to multispectral imaging results. In addition, the locations of gross anatomical features were noted in both autofluorescence images and histology specimens to aid in correlation. The resulting histopathology sections were evaluated to provide a diagnosis along the entire length of the epithelium, also noting any submucosal abnormalities in each slide. Histopathology diagnosis included the following categories: normal, mild dysplasia, moderate dysplasia, severe dysplasia/carcinoma in situ, and invasive carcinoma. For the purposes of diagnostic algorithm development, two major categories were defined: normal tissue (including inflammation and hyperplasia) and neoplastic tissue (including dysplasia, carcinoma in situ and cancer).

Analysis and Statistical Methods Images were preprocessed to subtract signal from ambient room light and translated so that white light and fluorescence images of the same field of view were spatially registered. 276 measurements corresponding to 159 unique regions of interest (ROIs) sites of clinically normal and suspicious regions of tissue were selected from white light images by a head and neck surgeon (AMG) blinded to the results of the autofluorescence imaging. In some cases, repeat measurements were obtained from the same ROI site to help ensure image data was collected without motion artifacts; often both the first and repeat measurements were included in the analysis. These repeat measurements account for the difference between the number of measurements and the number of ROI sites. Heterogeneity in pathologic diagnoses may occur within relatively small areas of diseased oral mucosa [25, 26], so ROIs were stringently selected from suspicious areas using one of following four criteria: 1) areas corresponding to the same size and location as a biopsy with a pathological diagnosis, 2) ROIs from locations which could be correlated to a histopathology slide with a corresponding pathological diagnosis, 3) areas within well-defined exophytic tumors confirmed by pathological diagnosis and 4) ROIs from a location which was clinically normal and deemed by the physician to be sufficiently distant from the lesion.

Autofluorescence images from the training set were analyzed to determine whether specific image features could be used to classify an ROI as normal or neoplastic. The autofluorescence images and white light images were spatially registered so that the ROIs chosen in the white light images corresponded to the same region of tissue in the autofluorescence images. The training set included data from the first 39 patients and 7 normal volunteers and included measurements from 173 ROIs. Qualitatively, neoplastic ROIs were associated with a decrease in average green fluorescence intensity and often an increase in red fluorescence intensity. The mean ratio of red-to-green pixel intensities inside each of the ROIs was calculated from the fluorescence images at each excitation wavelength. Red and green pixel intensities were obtained from the collected Red-Green-Blue color images, created by the Bayer color mask on the CCD detector. A classifier was developed to distinguish neoplastic and normal ROIs using linear discriminant analysis with the single input feature of average ratio of red-to-green fluorescence. When more than one measurement corresponded to a ROI site, the mean of the feature values was used for classification. The classifier was trained using all of the ROI sites in training set and the prior probability input into the classifier was chosen to represent the percentage of abnormal to normal measurements in the data set. The classifier was developed after images were acquired from patients in the training set but before measurements were acquired from patients in the validation set. Classifier accuracy in the training set was assessed by plotting the receiver operating characteristic (ROC) curve, the area under the ROC curve (AUC), and the sensitivity and specificity at a particular operating point on the ROC curve [27, 28]. The positive and negative predictive values were also calculated at the operating point. Confidence intervals were calculated for operating characteristics using the Wilson 'score' method including a continuity correction. The algorithm was then applied to data from the validation set using the red-to-green ratio threshold found to produce the highest combination of sensitivity and specificity in the training set. The validation set was designed to rigorously test the algorithm and for most patients, ROI and biopsy pairs were collected on the clinical margins of the lesion in addition to directly on the lesion and in clinically normal areas. The validation set included 103 measurements from 57 unique ROIs in a second group of 17 patients and 4 normal volunteers.

An additional analysis step was explored to increase the performance of the classifier by normalizing the red-to-green ratio measurements for each patient. An additional unique and non- overlapping ROI of clinically normal tissue was chosen from the same anatomical site and in the same field of view for each of the ROIs described above. At each excitation wavelength, the mean red-to-green autofluorescence ratio was calculated in this ROI; the mean red-to-green ratios from the other ROIs were normalized by this value. This method provides a way to compensate for anatomical and patient to patient variations in red-to-green fluorescence intensity ratio. Identical statistical analysis was performed using this measured feature with both the training set and the validation set. The method utilizing the magnitude of the red-to-green fluorescence intensity ratio is termed the raw red-to-green method and the method utilizing a normalized red-to-green fluorescence intensity ratio is termed normalized red-to-green method. Disease Probability Maps

The classification algorithms described above provided a relationship between the magnitude of the red-to-green fluorescence intensity ratio for a particular region of interest within the image and die probability of that region having a diagnosis of abnormal. This relationship was used to predict the probability of a diagnosis of dysplasia or cancer for each pixel in an image, given the red-to- green fluorescence intensity ratio at that pixel. The posterior probability values at each pixel in the image were computed and pixels which corresponded to a 50% or greater probability of being classified as dysplastic or cancerous were color coded and digitally overlaid onto the white light images. This method provides a means to illustrate areas of tissue with the highest probability of being neoplastic. The assumption was made that the region of interest method described above could be generalized on a pixel by pixel basis. Disease probability maps were compared to histologic images of tissue resected from the field of view to confirm the accuracy of this method. Results

Tables 2 and 3 summarize the anatomic site and histopathologic diagnoses of the 159 sites included in this analysis. The most common sites were tongue, bucca mucosa and floor of mouth, followed by palate, lip, and gingiva. The training set contained 52% normal, 28% dysplastic, and 20% invasive carcinoma sites. The normal histopathologic category could include tissue with hyperkeratosis, hyperplasia, and/or inflammation as long as there was no dysplasia or carcinoma. The normal sites in the training set, based on available pathology (not including normal volunteers and normal sites where no biopsy was taken), included 7 sites (13.2% of normal sites) with hyperplasia and hyperkeratosis, 4 sites (7.5% of normal sites) with hyperkeratosis, and 3 sites (5.7% of normal sites) with hyperplasia and/or fibroadipose tissue. The validation set included 3 sites (8.6% of normal sites) with hyperplasia and hyperkeratosis, 1 sites (2.9% of normal sites) with hyperplasia, 1 site (2.9% of normal sites) with a submucosal hemorrhage, and 1 site (2.9% of normal sites) with marked inflammation and osteonecrosis. The abnormal histopathology category could include dysplasia and carcinoma. In the training set 59.2% of the abnormal sites were premalignant (mild, moderate, or severe dysplasia), in the validation set 68.2% of the abnormal sites were premalignant.

Table 1. Anatomic sites of ROIs in the training and validation set. Note: Percentages may not add up to 100 % because of rounding.

Anatomical Site No. of sites in No. of sites m training sci {%) validation set (%)

Tongue 37 (36.3) 19 (33J)

Buccal mucosa 12 ( 1 1.8) ]> {26 J)

Floor of mouth 22 (21 6) 4 (7.0)

Gingiva 2 12J⁾) 7 ( 12.3)

Lip 14 (13 7} 4 (7.0)

Palate 15 ( 14,7) 8 (14,0)

Toial 102 ( K)O) 57 ( 100) Table 2. Pathology diagnosis of ROI sites in training and validation set.

Diagnosis No, of sues in Mo, of sites in training set (%) validation set (%)

Normal 53 (52.0) 35 (61.4j Mild dysplasia 11 (10J) 5 (8.8) Moderate dysplasia 6 (5.9) 4 (7.0) Severe dysplasia/CIS 12 01.8) 6 {10.5) Imasivc carcinoma 20 (19.6) 7 (12.3) Total 102 (JOt)) 57 (IiX))

Figure 1 shows white light and autofluorescence images from the buccal mucosa of a patient with pathologically confirmed invasive carcinoma. The white light image (Fig. IA) shows two ROIs, one which corresponds to a pathologically confirmed invasive carcinoma, and the other which was clinically normal and outside of the pathologically confirmed clear resection margin. Figures IB- ID show autofluorescence images at different excitation wavelengths that were taken before surgery from the same field of view. The autofluorescence image obtained at 405 nm excitation qualitatively shows the greatest visual contrast between the normal and neoplastic ROIs. This observation was typical for study patients.

Table 4 summarizes the performance of both diagnostic algorithms, based on either the raw or the normalized mean red to green fluorescence intensity ratios, for classifying lesions in the training set. At each excitation wavelength, the classifier that used the normalized red-to- green fluorescence intensity ratio (Normalizaed R/G ratio) had slightly higher AUC than the algorithm based on the raw red/green fluorescence intensity ratio (Raw R/G ratio). In all cases, the highest AUC was obtained at 405 nm excitation. The sensitivity and specificity values at the point on the ROC curve nearest the gold standard (Q-point) are also reported in Table 4.

A scatter plot of the normalized red-to-green ratio at 405 nm excitation for each of the 102 sites in the training set, as well as the threshold of 1.19 used in the classification algorithm is shown in Figure 2A. Of the 102 sites, 4 were misclassified including one site of fibroadipose tissue on the lower lip misclassified at abnormal, one hyperkeratotic site on the right buccal misclassified at abnormal, one cancer site on the right lateral tongue misclassified as normal, and one site on the left soft palate with focal ulceration and dysplasia misclassified as normal. Figure 2B shows the ROC curve for this classifier; the AUC is 0.988, and at the Q-point, the sensitivity is 95.9% (95% confidence interval (CI) 84.9% - 99.3%) and the specificity is 96.2% (95% CI 85.9% - 99.3%). The positive predictive value is 95.9% (95% CI 84.9% - 99.3%) and the negative predictive value is 96.2% (95% CI 85.9% - 99.3%). This operating point is indicated on the ROC curve.

Table 3. Classification results at each fluorescence excitation wavelength using both the Raw RJG Ratio method and the Normalized RJG ratio method in the training set.

Rmv R/G ratio Normalized R/G ratio

Fluorescence cxcitaiioii

Sensitivity (%) Specificity (%} AUC Sensitivit) {%) Specificity (%) wavelength AUC

365 rail .832 83.7 73.6 .856 83.7 86.8

380 ran ,891 89.8 77,4 ,V24 83,7 m.i

405 ran ,971 91.8 92.5 95.9 %.2

450 ITO 311 81.6 92.5 .*⁾⁽)<> 85,7

The algorithm using the normalized red-to-green fluorescence intensity ratio at 405 nm excitation was applied to the validation set. In Figure 2C, a scatterplot of the normalized R/G ratio for each site in the validation set is shown along with the threshold that had been previously selected for the training set. Figure 2D depicts the ROC curve with the operating point selected for the training set indicated. A 100% sensitivity (95% CI 81.5% - 99.6%) and 91.4% specificity (95% CI 75.8% - 97.8%) and an AUC of .987 were achieved at this operating point for the validation set. The positive predictive value is 88.0% (95% CI 67.7% - 96.9%) and the negative predictive value is 100% (95% CI 86.7% - 99.7%). Of the 57 sites in the validation set, 3 were misclassified as abnormal, including one site on the left buccal with hyperplasia, one site on the right buccal, and another site on the left buccal.

Figure 3 shows white light and 405 nm excited autofluorescence images from a study patient with moderate dysplasia and carcinoma in situ located in the floor of mouth. The white light image is also shown with an overlay of the calculated disease probability map; regions corresponding to a predictive probability of a neoplastic lesion greater than 50% are shaded as indicated by the color bar. The disease probability map indicates the probability that a particular pixel in the image corresponds to a neoplastic area of tissue. Histologic sections obtained at several areas in the tissue are also shown. Only one of these areas was included in the previous classification analysis. The disease probability map shows qualitative agreement with the presence of dysplasia and cancer in the histologic sections. Figure 4 shows representative white light images with and without superimposed disease probability maps from four study patients. Images in the first three rows correspond to patients with histologically confirmed neoplasia, while the image in the bottom row is from a normal volunteer with no clinically suspicious lesions. Although the lesion in Figure A is obvious, those in Figures B and C are less so, highlighting the potential to aid clinicians in identifying the presence of neoplasia and identifying optimal sites for further evaluation with biopsy. Images in Figures 4A and B are from a patient with an invasive carcinoma in the floor of mouth. Images in Figures 4C and D are from a patient with a region of severe dysplasia on the tongue. The images in Figures 4E and F are from a patient with a region of moderate dysplasia on the gingiva. In all three cases, the disease probability map delineates the suspicious regions identified clinically by an oral cancer specialist and are consistent with histopathologic sections obtained. Figures 4G and H are from the inner lip of a normal volunteer and the disease probability map does not indicate any lesions. Discussion The results of this study, among other things, illustrate how autofluorescence imaging may enhance the ability of clinicians to detect and delineate areas of oral dysplasia and carcinoma. Although all four illumination conditions tested allowed visualization of changes in autofluorescence with neoplasia, illumination with 405 nm wavelength produced the highest discriminatory capability. This corresponds to previous findings comparing illumination wavelengths for autofluorescence imaging in freshly resected oral cancer surgical specimens. While subjective interpretation of loss of autofluorescence has been shown to be useful, there are several important advantages associated with objective and quantitative analysis of changes in autofluorescence signal. First, quantitative analysis methods provide a rigorous and repeatable way to determine the threshold for demarcating a lesion, even for providers with less experience. Second, digital imaging allows the operator to save and process images, directly comparing data from multiple patients in a series or from a single patient over time. Third, ratios of fluorescence intensity values provide a way to reduce variations in images associated with spatial non- uniformities in illumination.

In the present study, the performance of a simple classifier based on the ratio of red-to- green autofluorescence intensity at 405 nm excitation was tested and found to discriminate neoplastic and non-neoplastic tissue with a sensitivity and specificity of 96% in the training set and 100% sensitivity and 91.4% specificity in the validation set. These results compare favorably with the performance of visual oral examination. Downer, et al identified eight prospective studies between 1980 and 2002 that involved conventional oral exam with gold standard verification provided by an expert observer. In four of the studies the screeners were general dentists and in four of the studies the screeners were trained health workers. Sensitivity ranged from 59% to 97%, specificity ranged from 75% to 99%, and meta-analysis resulted in a weighted pooled sensitivity of 85% and a specificity of 97%. Other reports of the performance of visual oral screening include Sankaranarayanan et al (sensitivity 77%, specificity 76%), Ramadas et al (sensitivity 82%, specificity 85%), and Nagao et al (sensitivity 92%, specificity 64%). The classifier in this study can be applied to entire images of the oral cavity to visualize areas with a high probability of being neoplastic; disease probability maps correlate well with histologic sections obtained from tissue in the field of view.

The results demonstrate, among other things, quantitative fluorescence imaging as an objective approach to non-invasively identify and delineate the mucosal extent of neoplastic lesions in the, for example, the oral cavity.

Notwithstanding that the numerical ranges and parameters setting forth the broad scope of the invention are approximations, the numerical values set forth in the specific examples are reported as precisely as possible. Any numerical value, however, inherently contain certain errors necessarily resulting from the standard deviation found in their respective testing measurements.

Therefore, the present invention is well adapted to attain the ends and advantages mentioned as well as those that are inherent therein. While numerous changes may be made by those skilled in the art, such changes are encompassed within the spirit of this invention as illustrated, in part, by the appended claims.

References:

1 Parkin, D.M., et al., Global cancer statistics, 2002. CA Cancer J Clin. 55(2): p. 74-108, (2005).

2 NCI, "SEER Program: Oral Cancer", p. 96-99, (2003). 3 Lippman, S.M. and W.K. Hong, Second malignant tumors in head and neck squamous cell carcinoma: the overshadowing threat for patients with early-stage disease. Int J Radiat Oncol Biol Phys. 17(3): p. 691-4, (1989).

4 Day, G.L. and WJ. Blot, Second primary tumors in patients with oral cancer. Cancer.

70(1): p. 14-9, (1992). 5 Lingen, M.W., et al., Critical evaluation of diagnostic aids for the detection of oral cancer. Oral Oncol. 44(1): p. 10-22, (2008).

6 Gillenwater, A., V. Papadimitrakopoulou, and R. Richards-Kortum, Oral premalignancy: new methods of detection and treatment. Curr Oncol Rep. 8(2): p. 146-54, (2006).

7 Oh, E.S. and DM. Laskin, Efficacy of the ViziLite system in the identification of oral lesions. J Oral Maxillofac Surg. 65(3): p. 424-6, (2007).

8 Rosin, M.P., et al., Visualization and other emerging technologies as change makers for oral cancer prevention. Ann N Y Acad Sci. 1098: p. 167-83, (2007).

9 Poh, C.F., et al., Fluorescence visualization detection of field alterations in tumor margins of oral cancer patients. Clin Cancer Res. 12(22): p. 6716-22, (2006). 10 De Veld, D.C., et al., The status of in vivo autofluorescence spectroscopy and imaging for oral oncology. Oral Oncol. 41(2): p. 117-31, (2005).

11 Svistun, E., et al., Vision enhancement system for detection of oral cavity neoplasia based on autofluorescence. Head and Neck- Journal for the Sciences and Specialties of the Head and Neck. 26(3): p. 205-215, (2004). 12 Utzinger, U., et al., Optimal visual perception and detection of oral cavity neoplasia. IEEE Trans Biomed Eng. 50(3): p. 396-9, (2003).

13 Lane, P.M., et al., Simple device for the direct visualization of oral-cavity tissue fluorescence. J Biomed Opt. 11(2): p. 024006, (2006). 14 Kulapaditharom, B. and V. Boonkitticharoen, Laser-induced fluorescence imaging in localization of head and neck cancers. Ann Otol Rhinol Laryngol. 107(3): p. 241-6, (1998).

15 Anderson-Engels, S., et al., In vivo fluorescence imaging for tissue diagnostics. Phys Med Biol. 42(5): p. 815-24, (1997).

16 Richards-Kortum, R. and E. Sevick-Muraca, Quantitative optical spectroscopy for tissue diagnosis. Annu Rev Phys Chem. 47: p. 555-606, (1996).

17 Svistun, E., et al., Vision enhancement system for detection of oral cavity neoplasia based on autofluorescence. Head Neck. 26(3): p. 205-15, (2004). 18 Pavlova, L, et al., Understanding the biological basis of autofluorescence imaging for oral cancer detection: high-resolution fluorescence microscopy in viable tissue. Clin Cancer Res. 14(8): p. 2396-404, (2008).

19 de Veld, D.C., et al., Clinical study for classification of benign, dysplastic, and malignant oral lesions using autofluorescence spectroscopy. J Biomed Opt. 9(5): p. 940-50, (2004). 20 Lam, S., C. MacAulay, and B. Palcic, Detection and localization of early lung cancer by imaging techniques. Chest. 103(1 Suppl): p. 12S-14S, (1993).

21 Park, S.Y., et al., Automated image analysis of digital colposcopy for the detection of cervical neoplasia. J Biomed Opt. 13(1): p. 014029, (2008).

22 Kara, M.A., et al., Endoscopic video-autofluorescence imaging followed by narrow band imaging for detecting early neoplasia in Barrett's esophagus. Gastrointest Endosc. 64(2): p. 176-85, (2006).

23 Matsuda, T., et al., Does Autofluorescence Imaging Videoendoscopy System Improve the Colonoscopic Polyp Detection Rate?-A Pilot Study. Am J Gastroenterol, (2008).

24 Rahman, M., et al., Low-cost, multimodal, portable screening system for early detection of oral cancer. J. Biomed. Opt. 13(3), (2008).

IZ Roblyer, D., et al., Multispectral optical imaging device for in vivo detection of oral neoplasia. J Biomed Opt. 13(2): p. 024019, (2008).

26 Slaughter, D.P., H.W. Southwick, and W. Smejkal, Field cancerization in oral stratified squamous epithelium; clinical implications of multicentric origin. Cancer. 6(5): p. 963-8, (1953). 27 Braakhuis, BJ., et al., A genetic explanation of Slaughter's concept of field cancerization: evidence and clinical implications. Cancer Res. 63(8): p. 1727-30, (2003).

28 Duda, R.O., P.E. Hart, and D.G. Stork, Pattern Classification. Second ed. 2001, New York: John wiley & Sons, Inc. 29 Zhou, X.-H., McClish, D. K., Obuchowski N. A., Statistical Methods in Diagnostic Medicine. 2002, New York: John Wiley and Sons.

30 Albert, A. and E.K. Harris, Multivariate interpretation of clinical laboratory data. 1987, New York: Marcel Dekker inc.

31 Newcombe, R.G., Two-sided confidence intervals for the single proportion: comparison of seven methods. Stat Med. 17(8): p. 857-72, (1998).

32 Downer, M. C, et al., A systematic review of test performance in screening for oral cancer and precancer. Oral Oncol. 40(3): p. 264-73, (2004).

33 Sankaranarayanan, R., et al., Early findings from a community-based, cluster- randomized, controlled oral cancer screening trial in Kerala, India. The Trivandrum Oral Cancer Screening Study Group. Cancer. 88(3): p. 664-73, (2000).

34 Ramadas, K., et al., Interim results from a cluster randomized controlled oral cancer screening trial in Kerala, India. Oral Oncol. 39(6): p. 580-8, (2003).

35 Nagao, T., et al., Outcome following a population screening programme for oral cancer and precancer in Japan. Oral Oncol. 36(4): p. 340-6, (2000). 36 Zeng, H., MacAulay,C, Real-time endoscopic fluorescence imaging for early cancer detection in the gastrointestinal tract. Bioimaging. 6, (1998).

37 Kurie, J.M., et al., Autofluorescence bronchoscopy in the detection of squamous metaplasia and dysplasia in current and former smokers. J Natl Cancer Inst. 90(13): p. 991-5, (1998). 38 Hirsch, F.R., et al., Fluorescence versus white-light bronchoscopy for detection of preneoplastic lesions: a randomized study. J Natl Cancer Inst. 93(18): p. 1385-91, (2001).

39 Inaguma, M. and K. Hashimoto, Porphyrin-like fluorescence in oral cancer: In vivo fluorescence spectral characterization of lesions by use of a near-ultraviolet excited autofluorescence diagnosis system and separation of fluorescent extracts by capillary electrophoresis. Cancer. 86(11): p. 2201-11, (1999). Chen, J.K., R.V. Katz, and DJ. Krutchkoff, Intraoral squamous cell carcinoma. Epidemiologic patterns in Connecticut from 1935 to 1985. Cancer. 66(6): p. 1288-96, (1990).

Claims

CLAIMSWhat is claimed is:

1. A method comprising: providing a tissue sample, providing a first imaging device comprising a white light imaging device; providing a second imaging device comprising at least one imaging device selected from the group consisting of: a fluorescent imaging device, a narrowband reflectance imaging device, and a polarized reflectance imaging device; obtaining a first image of the tissue sample with the first imaging device, wherein the first image comprises a plurality of pixels; obtaining a second image of the tissue sample with the second imaging device, wherein the second image comprises a plurality of pixels; calculating a metric for at least one of the plurality of pixels of the second image; and calculating a disease probability using the metric calculated for the at least one of the plurality of pixels of the second image.

2. The method of claim 1 wherein the second imaging device comprises a fluorescent imaging device.

3. The method of claim 1 further comprising overlaying the first image and the second image.

4. The method of claim 2 wherein the step of obtaining a second image of the tissue sample with the second imaging device comprises obtaining an image of the tissue sample with the fluorescent imaging device at an excitation wavelength of about 405 nm.

5. The method of claim 1 wherein the metric is the ratio of red pixel values to green pixel values.

6. The method of claim 1 wherein the tissue sample comprises at least one tissue selected from the group consisting of: an oral tissue, a bronchial tissue, a cervical tissue, an esophageal tissue, a colon tissue, a breast tissue, a gastrointestinal tract tissue, and a lung tissue.