Journal of Cancer Research and Therapeutics

: 2010  |  Volume : 6  |  Issue : 4  |  Page : 497--502

Fluorescence spectroscopy for noninvasive early diagnosis of oral mucosal malignant and potentially malignant lesions

Pankaj Chaturvedi1, Shovan K Majumder1, Hemant Krishna2, Sidramesh Muttagi1, Pradeep K Gupta2,  
1 Laser Biomedical Applications and Instrumentation Division, Raja Ramanna Centre for Advanced Technology, Indore, India
2 Department of Head and Neck Surgery, Tata Memorial Hospital, Mumbai, India

Correspondence Address:
Shovan K Majumder
Laser Biomedical Applications and Instrumentation Division, Raja Ramanna Centre for Advanced Technology, Indore


Background: We report the results of a clinical in vivo study to evaluate the potential of fluorescence spectroscopy for differential diagnosis of oral mucosal malignant and potentially malignant lesions. Materials and Methods: The study involved 26 healthy volunteers and 144 patients enrolled for routine medical examination of the oral cavity at the outpatient department of the Tata Memorial Hospital, Mumbai. In vivo autofluorescence spectra were recorded using a N 2 laser based portable fluorimeter developed in-house. The different tissue sites investigated belonged to either of the four histopathologic categories: 1) squamous cell carcinoma (SCC), 2) oral sub-mucous fibrosis (OSMF), 3) leukoplakia (LP) and 4) normal squamous tissue. A multivariate statistical algorithm capable of direct multi-class classification was used to predict pathological designations. Results: With respect to histopathology as the «DQ»gold standard«DQ», the diagnostic algorithm was found to provide an accuracy of 82, 76, 81 and 85% based on leave-one-patient-out cross-validation in classifying the oral tissue spectra into four different pathology classes - SCC, OSMF, LP, and normal squamous tissue - respectively. When the algorithm was employed for delineating the normal oral tissues from all the abnormal oral tissues including SCC, OSMF and LP put together, a sensitivity of 98% and a specificity of 100% were obtained. Conclusion: The results suggest that it is possible to objectively classify the oral tissue into different pathology classes based on their in vivo autofluorescence spectra. Thus, the technique can potentially improve oral screening efforts in low resource settings where clinical expertise and resources are limited.

How to cite this article:
Chaturvedi P, Majumder SK, Krishna H, Muttagi S, Gupta PK. Fluorescence spectroscopy for noninvasive early diagnosis of oral mucosal malignant and potentially malignant lesions.J Can Res Ther 2010;6:497-502

How to cite this URL:
Chaturvedi P, Majumder SK, Krishna H, Muttagi S, Gupta PK. Fluorescence spectroscopy for noninvasive early diagnosis of oral mucosal malignant and potentially malignant lesions. J Can Res Ther [serial online] 2010 [cited 2022 Jun 30 ];6:497-502
Available from:

Full Text


Oral cancer is the world's sixth most common cancer, and global incidence and mortality rates are increasing. [1],[2] In the US, overall 5-year-survival rates for all oral squamous cell carcinoma (SCC) patients are barely over 50%, and for those with advanced disease, survival rates are 30-40% or less. [3] However, oral cancer is predominantly a disease of developing nations, particularly prevalent in India and other southeast Asian countries due to the habit of tobacco usage. In fact, one-fifth of the world's oral cancer patients live in the Indian subcontinent and it is one of the leading causes of cancer related deaths in Indian men. As per the Mumbai cancer registry, overall 5-year survival rates are only 30.5%. [4] Although patients with early disease have better chances for cure and functional outcome, most patients present with advanced tumors when treatment is more difficult, more expensive and less successful compared to earlier intervention. [3] The best way to improve outcomes is to improve early detection and diagnosis. Screening by visual examination of high risk population lacks accuracy, reproducibility and requires expertise. [5] The best means of obtaining a definitive diagnosis is a tissue biopsy that is subject to sampling error and patient discomfort. It would be ideal to have a noninvasive test that can accurately detect neoplastic or pre-neoplastic changes in the oral mucosa. Recent research has demonstrated that optical methods can present a viable approach for improving screening and detection of oral malignancies. [6]

A large body of work carried out over the past years [6],[7] has already established the potential of autofluorescence spectroscopy to tease out diagnostically useful information about human oral cavity tissues. Both fluorescence imaging and spectroscopic methods have been used with encouraging results. [7] However, most of these studies have looked at binary discrimination of normal and abnormal (neoplastic) tissue. We had earlier developed a multi-class diagnostic algorithm based on Total Principal Component Regression and showed its utility for simultaneous multi-class classification (high-grade SCC, SCC grade 1, leukoplakia, normal squamous tissue) of oral tissue. [8] Our current study was designed to follow up on the results of these past studies to further evaluate the potential of autofluorescence spectroscopy to classify the oral tissue spectra into four different categories - SCC, leukoplakia (LP), oral submucous fibrosis (OSMF) and normal squamous tissue - based on their autofluorescence spectra. We have also investigated the feasibility of the use of fluorescence spectroscopy to monitor the spectral changes in oral mucosa of healthy volunteers with a history of tobacco use.

 Materials and Methods

Ethical Committee approval was obtained and all the patients undergoing routine medical examination of the oral cavity at the Out Patient Department (OPD) were recruited in the in vivo study, regardless of gender or race. The final eligibility of each patient was determined by the participating doctor, based on the medical condition of the patient such that patient care was not compromised.

The study involved 26 normal volunteers with no history of the disease of the oral cavity and 144 patients enrolled for medical examination of the oral cavity. Informed consent was obtained from each patient as well as the normal volunteers who participated in this study. Age, sex, and details of smoking habit (if any) were also recorded for all subjects included in the study. The age of the patients ranged from 37 to 75 years with mean age and standard deviation of 50 years and 10, respectively, whereas for normal volunteers the age ranged from 24 to 58 years with mean and standard deviation being 38 years and 17, respectively. The overall ratio of male to female population was 2:1. As far as tobacco habits are concerned, 98% of the patients and 70% of the normal volunteers had habits of either smoking or chewing tobacco, and the remaining candidates were non tobacco users.

The patients included in this study had no history of malignancy or dysplasia, and were suspected by the examining physician of having either malignant or potentially malignant lesions on visual examination of the oral cavity. Patients having gone through any treatment like surgery, chemotherapy or radiotherapy or with recurrences were excluded from the study. Biopsies were taken subsequent to acquisition of spectra from the oral cavity sites suspected of being malignant or potentially malignant. However, as per the terms of the approval from the ethical committee of the hospital, no biopsies were available from the investigated sites of the patients with OSMF and the diagnosis of this condition was based on clinical findings only. The biopsy samples were fixed in formalin and were examined later by an experienced pathologist who was blinded to the results of the optical spectra. Histopathology was considered to be the "gold standard". All fluorescence spectra were categorized in accordance to their histological identities and grouped into SCC, LP, OSMF, or normal oral squamous tissue.

In vivo autofluorescence spectra were measured using a compact and portable spectroscopic system as illustrated in [Figure 1]. A sealed-off, high-pressure nitrogen laser, developed in-house was used as the excitation source for inducing tissue fluorescence. Light delivery to and collection from tissue was achieved with a fiber-optic probe consisting of seven 400 μm core diameter fibers (0.22 NA) arranged in a six-around-one configuration (Avantes Inc., Broomfield, CO, USA). The six surrounding fibers deliver laser light to the tissue surface while the central fiber collects fluorescence from the surface area directly illuminated by the excitation light. The fluorescence emission collected by the fiber-optic probe is then dispersed and detected with a chip-based spectrometer (model number S-2000, Ocean Optics, Dunedin, FL, USA). For this study, the nitrogen laser was operated at 10 Hz repetition rate, 5 ns pulse width, and an average pulse energy of 50 ± 5 μJ at the tissue surface. An integration time of 500 ms was used for each spectral measurement. The overall spectral resolution of the system was ~5 nm. The spectral data acquisition was computer controlled.{Figure 1}

All the spectral measurements were performed by the participating head and neck surgeon (SM) using a standard protocol which was maintained for all patients in this study. Prior to recording spectra from an individual, the fiber-optic probe was disinfected with CIDEX (Johnson and Johnson, Mumbai, India), washed with phosphate buffered saline (PBS) and cleaned dry with a piece of sterilized cotton. The mucosal surface was wiped with dry gauge to remove any saliva, blood or betel quid incrustations accumulated at the tissue surface. The probe tip was also wiped dry between consecutive measurements from different tissue sites in an individual. For recording the in vivo autofluorescence spectra, the tip of the fiber-optic probe was placed in gentle contact with the tissue surface and it was ensured that none of the patients or the normal volunteers complained of the probe being painful. A background spectrum was acquired with the probe placed in air. It was subtracted from all subsequently acquired spectra. From each site, three spectra were recorded and were averaged to yield a single spectrum per site. The overhead room lights in the OPD room were turned off temporarily during spectral acquisition to minimize the contribution of the ambient light in the acquired spectra. A set of reference spectra was also acquired from a fluorescence standard to correct for the interpatient variability due to variations in laser-pulse energy prior to each patient study. The fluorescence standard is a low-concentration Rhodamine 6G solution (2 mg/L) contained in a quartz cuvette. All the measured fluorescence spectra were processed to remove instrumentation-induced variations and to yield calibrated spectra. The resultant fluorescence spectra were further corrected for the non-uniform spectral response of the detection system. The minimum signal-to-noise ratio was ~100:1 at the fluorescence maximum.

The in vivo autofluorescence spectra were recorded in the 375-700 nm spectral range from a total of 477 tissue sites of 144 patients. Out of these, 88 sites were identified as OSMF by the examining doctor and from these no biopsies were taken. Of the remaining tissue sites, 299 were histopathologically characterized as SCC and 90 as LP. Spectra were also recorded from 283 sites from healthy squamous tissue of 26 normal volunteers. The normal volunteers had no clinically observable lesions of the oral cavity and also had no history of malignancy. The details of the histopathological distribution of the tissue sites included in the study are summarized in [Table 1]. Each site was treated separately and classified via the diagnostic algorithm developed.

The quantitative analysis of the spectra involved two steps: extraction of diagnostically relevant spectral information through Maximum Representation and Discrimination Feature (MRDF) [9] and classification via Sparse Multinomial Logistic Regression (SMLR). [10] In brief, MRDF is a feature extraction procedure that aims to find a set of nonlinear transformations on the input data that optimally discriminate between the different classes in a reduced dimensionality space. SMLR separates a set of labeled input data into its constituent classes based on the posterior probabilities of their class membership.{Table 1}

The inputs to these algorithms were the full processed spectra after normalization according to the scheme described previously by Majumder et al. [10] All analyses were performed using leave-one-patient-out cross-validation. In this method, the training of the algorithm was performed using N − 1 samples (where N = 170 patients), and test was carried out using the excluded sample. This was repeated N times, each time excluding a different patient. Thus, training was achieved using, in a sense, all patients, and at the same time independence between training and test set was maintained. Each spectrum was classified to the predicted class membership (pathology) with the highest posterior probability.


[Figure 2] shows the mean fluorescence spectra for SCC (n = 299), OSMF (n = 83), LP (n = 90), and normal squamous (n = 283) tissues, with the error bars representing ±1 standard deviation. It is apparent that the variation in the measured spectral intensity is comparable for all the tissue types. The percentage variation () in the spectral intensities from the different measurement sites was observed to lie in the range of 15-35% over the respective number of tissue sites included in the four histopathological categories. Here, is the mean intensity value from different measurement sites of one category and σ is one standard deviation.{Figure 2}

The spectra show a variety of differences between the different tissue types. The most prominent of these are seen in the wavelength region below 500 nm, particularly in the 390 and 460 nm spectral bands. The 390 nm spectral band is the most intense in OSMF tissues, while the intensity of the 460 nm band is the highest in the spectra from normal squamous tissues. However, past 500 nm, no visually obvious significant differences are observed between the tissue types excepting a small peak around 635 nm, found only in the spectra of cancerous tissues. The 390 nm band is generally attributed to collagen, while the differences in the longer wavelength band are associated with the NADH emission, which has its peak at 460 nm. The peak around 635 nm is believed to be due to endogenous porphyrins.

[Table 2] shows the results in the form of confusion matrix displaying comparisons of the pathological diagnosis with that of the MRDF-SMLR based spectroscopic diagnostic algorithm. The classification results were obtained based on leave-one-patient-out cross-validation of the entire data set. One can see that the algorithm provided an overall classification accuracy of 82.5% (627 out of 760). It proved most adept at classifying normal squamous tissues with a classification accuracy of 85%, though it fared worse in classifying other tissue types, and errors were spread among the various classes. SCC spectra were correctly classified in 82% of the sites, while LP and SMF spectra were classified correctly in 81 and 76% of the sites, respectively.{Table 2}

[Table 3] shows the confusion matrix for the performance of the MRDF-SMLR algorithm based on leave-one-patient-out cross-validation applied, in three-class classification mode, on the whole set of spectra separated into three categories (instead of four), normal, malignant (SCC) and potentially malignant, where the SMF and LP spectra were put together to form the third category. Although the overall discrimination accuracy is seen to be reduced marginally by 1.5% (from 82.5 to 81%), the accuracy with which the SMF and LP spectra together (belonging to the potentially malignant category) can be discriminated is found to improve to 84% with 150 out of 178 spectra of this category being classified correctly. However, the algorithm fared worse in classifying the normal squamous tissue spectra where the classification accuracy is found to be only 77%.{Table 3}

The discrimination results of the three-class algorithm (to separate the same three categories) improved when the normal category comprised spectra from only those healthy volunteers who did not have any known tobacco habits. This led to a substantial increase in the accuracy of discrimination of normal tissue spectra from 77 to 92%, with the overall classification accuracy having improved by 3% (from 81 to 84%) [Table 4]. In order to ascertain whether within the healthy volunteers themselves the set of spectra was different amongst those with tobacco habits and those without any tobacco habits, we applied the algorithm on the whole set of spectra (from the healthy volunteers), split into two separate groups, one with tobacco habits and the other without any tobacco habits. It was found that the normal volunteers with and without tobacco habits could be discriminated with up to 97% accuracy based on their autofluorescence spectra. The mean spectra of these two groups were also observed to have considerable difference in intensity as well as spectral distribution with the spectra of healthy volunteers with tobacco habits, in general, showing reduced fluorescence intensity throughout the spectral region. [Table 5] shows the results of the performance of the MRDF-SMLR algorithm applied, in binary classification mode, on the whole set of spectra separated into two categories, normal and abnormal, where SCC, SMF and LP spectra were put together to form the abnormal category and the spectra of all the healthy volunteers were put into the normal category. When "normal or healthy volunteers" with tobacco habits were excluded, discrimination results improved [Table 6]. It is apparent that the normal tissue sites can be discriminated from abnormal with 91% sensitivity and 90% specificity, or 94% positive predictive value and 86% negative predictive value, or an overall accuracy of 91%.{Table 4}{Table 5}{Table 6}


Although, recent studies have shown that visual screening can reduce the mortality in high-risk individuals and has a potential of preventing at least 37,000 oral cancer deaths worldwide, [5] the clinical risk stratification in a population by visual screening lacks accuracy, reproducibility and requires expertise in view of a vast number of lesions that manifest in the oral cavity with identical features. Consequently, over the years, several modalities like brush cytology and toluidine blue staining have been developed. [11],[12],[13],[14] Unfortunately, none of them have been shown to be ideal. They are either subjective, nonspecific, or have high costs, both in time and money. [15] The only means of obtaining a definitive diagnosis at present is via biopsy, which is subject to sampling error and patient discomfort. Hence, there is a need for a specific noninvasive device that can help in differentiation of high-risk lesions from the low-risk ones, thereby increasing the efficacy of community screening programs. This would not only improve the accuracy but also make it more acceptable, objective and reproducible. Optical based diagnostic aids are promising new technologies for improving screening and detection of epithelial malignancies in several organ sites. [6] During carcinogenesis, the optical properties of tissue are altered. Studies suggest that increased mitochondrial autofluorescence, decreased collagen autofluorescence, and decreased reflectance due to hemoglobin absorption occur during carcinogenic progression. These optical changes can be used for imaging based diagnosis using scattered light or fluorescence (ViziLite Plus, MicroLux DL, VELscope) or spectroscopic based diagnosis using reflectance and fluorescence spectroscopy. [16],[17],[18],[19] The reflectance and fluorescence spectroscopic approaches are particularly attractive as these can provide quantitative information about the biochemical and morphological states of tissue in a minimally invasive or noninvasive manner [6] using a very simple and low-cost instrument. Although different diagnostic algorithms have been developed and used for discriminating tissue sites based on their spectral signatures, [6],[7] most of these can only be used for binary classification. The ability to simultaneously classify oral tissue spectra into more than two classes is necessary as fluorescence spectroscopy moves closer to the clinic. The goal of the present study was to use a multivariate statistical algorithm capable of direct multi-class classification for objectively classifying the oral tissue spectra into four different pathology classes - SCC, OSMF, LP, and normal squamous tissue.

The primary basis for optical detection using spectroscopic techniques is an array of biochemical changes that take place as tissue undergoes neoplastic transformations. [6],[7] For example, SCC, OSMF, LP and normal oral tissues are known to show variable amount of collagen and elastin which is seen to get reflected in the fluorescence intensity of the 390 nm band characteristic of these connective tissue proteins [Figure 2]. Similarly, the differences in the concentrations and oxidation states of coenzymes such as NADH and FADH 2 due to differences in metabolic activities in normal and neoplastic oral tissues contribute to the changes in the fluorescence intensity of the broad 460 nm band believed primarily due to these fluorophores. [20],[21] Some of the changes found in the fluorescence spectra of normal and abnormal oral tissues also result from changes in the wavelength dependent absorption and scattering properties of tissues (that modulate the intrinsic tissue fluorescence). These changes are consistent with those seen in other studies and result from structural and metabolic changes associated with cancer. [6],[22],[23]

A critical evaluation of the diagnostic results listed in [Table 2],[Table 3],[Table 4],[Table 5],[Table 6] reveals some interesting points that are worth noting. In the four-class classification case, the LP and SMF spectra were classified correctly in 81 and 76%, respectively, of the sites. In the three-class mode, the classification accuracy of the SMF and LP spectra together (belonging to the potentially malignant category) is found to improve to 84%, though for the normal squamous tissue spectra the accuracy is found to reduce to 77%. The situation drastically improves with a substantial increase in the accuracy of discrimination of the normal oral tissues [Table 4] when the spectra of those with tobacco habits are excluded from the set of normal squamous tissue spectra belonging to the healthy volunteers. The results suggest that there is considerable difference in the oral mucosa of the healthy volunteers with and without tobacco habits. A further confirmation toward this is the fact that the spectra of the healthy volunteers without any tobacco habits could be discriminated from the abnormal oral tissues with 100% accuracy [Table 6]. In contrast, when the spectra from all the healthy volunteers were put together irrespective of tobacco habits the accuracy fell to 90% [Table 5].

It is pertinent to mention here that when it comes to distinguishing only normal from abnormal tissues, fluorescence spectroscopy is seen to have the potential to provide such discrimination with excellent accuracy. For example, the diagnostic results show that fluorescence spectroscopy can discriminate abnormal from normal oral tissues with sensitivity and specificity of 98 and 100%, respectively. Thus, if the objective is only to delineate abnormal from normal oral mucosa, as may be required for routine screening procedures, the fluorescence approach can serve as a method of choice. A further incentive toward using this approach is that given the currently available instruments, fluorescence is a significantly strong candidate for imaging techniques. In many applications, such as evaluating margin status, it is highly desirable to move to an imaging-based system rather than point spectroscopy to gather more information from an entire tissue surface in a much quicker time.

In terms of a clinical application, the most interesting statistic is likely negative predictive value. The physicians would like to be confident that any diagnosis of normal is an accurate one and they are not missing out any neoplastic lesion. A high positive predictive value would be desirable as well to avoid unnecessary biopsies during the clinical examination. From this data set [Table 6], the negative predictive value is 92% and the positive predictive value is 100%. Although there were over five times as many positive spectra as negative, predictive values often taken into account different population sizes, and the distribution of measurements in this sample set is a reasonable approximation of what might be encountered in actual medical use.

It should be noted that the diagnostic algorithms developed in this study were based on spectra from a limited number of patients assumed to be representative of the entire patient population. The patient selection criteria as well as the limited number of spectra in each pathologic category might influence the classification results obtained in this study. Therefore, further clinical studies in a larger patient population, which are already in progress, will be used to validate the classification estimates presented here.


The results strengthen the hypothesis that fluorescence spectroscopy has considerable potential for use as a diagnostic tool in a clinical setting and is an efficient technique for early detection of malignant and potentially malignant lesions of the oral cavity. The accuracy of diagnosis, using this technique, is found to be better than that got through visual examination. The technique also has considerable potential in objective discrimination of tobacco users and non users. Further work is in progress to test the efficacy of this technique in screening of oral mucosal malignant and potentially malignant lesions in a larger patient population.


The authors thank Dr. Sendhil Raja and his team at RRCAT, Indore, for help with instrumentation. The authors would also like to thank the nursing staff of the Head and Neck Surgery Department, Tata Memorial Hospital, for their active cooperation and help.


1Parkin MD, Bray F, Ferlay J, Pisani P, Global Cancer Statistics, 2002. CA Cancer J Clin 2005;55:74-108.
2Stewart BW, Kleihues P. World cancer report. Lyon: WHO international Agency for Research on Cancer. Lyon, France: IARC Press; 2003. p. 232-6. Available from: [Last accessed on 2003].
3Lopez AD, Mathers CD, Ezzati M, Jamison DT, Murray CJL, Global and regional burden of disease and risk factors, 2001: Systematic analysis of population health data, Lancet 2006; 367: 1747-57.
4Parkin DM, Whelan SL, Ferlay J, Teppo L, Thomas DB. Cancer incidence in five continents. Lyon: IARC Press; 2002. 232-49.
5Sankaranarayanan R, Ramadas K, Thomas G, Muwonge R, Thara S, Mathew B, et al. Effect of screening on oral cancer mortality in Kerala, India: A cluster-randomised controlled trial. Lancet 2005;365:1927-33.
6Ramanujam N. Fluorescence spectroscopy of neoplastic and non-neoplastic tissues. Neoplasia 2000;2:1-29.
7De Veld DC, Witjes MJ, Sterenborg HJ, Roodenburg JL. The status of in vivo autofluorescence spectroscopy and imaging for oral oncology. Oral Oncol 2005;41:356-64.
8Majumder SK, Gupta A, Gupta S, Ghosh N, Gupta PK. Multi-class classification algorithm for optical diagnosis of oral cancer. J Photochem Photobiol B 2006;85:109-17.
9Majumder SK, Gupta A, Gupta S. Non-linear multi-class pattern recognition for laser-induced fluorescence diagnosis of oral cavity cancer. Photo/Electrochemistry and Photobiology in the Environment, Energy and Fuel. In: Kaneco SB, Katsumata V, Katsumata H, editors. Research Signpost, 2009. p. 95-114.
10Majumder SK, Gebhart S, Johnson MD, Thompson R, Lin WC, Mahadevan-Jansen A. A probability-based spectroscopic diagnostic algorithm for simultaneous discrimination of brain tumor and tumor margins from normal brain tissue. Appl Spectrosc 2007;61:548-57.
11Fist S. The oral brush biopsy: Separating fact from fiction. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 2003;96:654-5.
12Eisen D. Brush biopsy ′saves lives′. J Am Dent Assoc 2002;133:688-90, 692.
13Warnakulasuriya KA, Johnson NW. Sensitivity and specificity of OraScan (R) toluidine blue mouthrinse in the detection of oral cancer and precancer. J Oral Pathol Med 1996;25:97-103.
14Ram S, Siar CH. Chemiluminescence as a diagnostic aid in the detection of oral cancer and potentially malignant epithelial lesions. Int J Oral Maxillofac Surg 2005;34:521-7.
15Johansson A, Kromer K, Sroka R, Stepp H. Clinical optical diagnostics- Status and perspective. Med Laser Appl 2008;23:155-74.
16Mashberg A, Samit AM. Early detection, diagnosis, and management of oral and oropharyngeal cancer. CA Cancer J Clin 1989;39:67-88.
17Oh ES, Laskin DM. Efficacy of the ViziLite system in the identification of oral lesions. J Oral Maxillofac Surg 2007;65:424-6.
18Ram S, Siar CH. Chemiluminescence as a diagnostic aid in the detection of oral cancer and potentially malignant epithelial lesions. Int J Oral Maxillofac Surg 2005;34:521-7.
19Majumder SK, Mohanty SK, Ghosh N, Gupta PK, Jain DK, Khan F. A pilot study on the use of autofluorescence spectroscopy for diagnosis of the cancer of human oral cavity. Curr Sci 2000;79:1089-94.
20Gillenwater A, Jacob R, Ganeshappa R, Kemp B, El-Naggar AK, Palmer JL, et al. Noninvasive diagnosis of oral neoplasia based on fluorescence spectroscopy and native tissue autofluorescence. Arch Otolaryngol Head Neck Surg 1998;124:1251-8.
21Majumder SK, Uppal A, Gupta PK. Autofluorescence spectroscopy of tissues from human oral cavity for discriminating malignant from normal. Lasers Life Sci 1999;8:211-27.
22Wagnières GA, Star WM, Wilson BC. In vivo fluorescence spectroscopy and imaging for oncological applications. Photochem Photobiol 1998;68:603-32.
23Richards-Kortum R, Sevick-Muraca E. Quantitative optical spectroscopy for tissue diagnosis. Annu Rev Phys Chem 1996;47:556-606.