Multi-class classification algorithm for optical diagnosis of oral cancer

Proceedings of the North American Society of Head and Neck Pathology Companion Meeting, New Orleans, LA, March 12, 2023: Oral Cavity Dysplasia: Why Does Histologic Grading Continue to be Contentious?

Head and neck squamous cell carcinoma (HNSCC) is the world's 6th most common malignancy. Oral cavity SCC (OCSCC) represents approximately one third of the HNSCC cases diagnosed annually in the United States. Despite therapeutic advances, OCSCC is frequently lethal, with a modest 5-year survival. Because OCSCC is often preceded by premalignant lesions, it is an ideal disease for screening initiatives. The conventional visual and tactile exam (CVTE), coupled with a tissue biopsy, remains the gold standard. However, CVTE alone cannot reliably differentiate between reactive/inflammatory and dysplastic lesions. Further, the histologic diagnosis of dysplasia is subjective in nature and a highly imperfect predictor of malignant transformation. This prognostic uncertainty creates a significant clinical management dilemma-watchful waiting with increased patient psychological and economic burdens versus unnecessary aggressive treatment. As such, the development and validation of novel diagnostic platforms such as Artificial Intelligence (AI) and prognostic molecular biomarkers may help address these critical unmet clinical needs.

Review of the Use of Artificial Intelligence in Early Diagnosis and Prevention of Oral Cancer

The global occurrence of oral cancer (OC) has increased in recent years. OC that is diagnosed in its advanced stages results in morbidity and mortality. The use of technology may be beneficial for early detection and diagnosis and thus help the clinician with better patient management. The advent of artificial intelligence (AI) has the potential to improve OC screening. AI can precisely analyze an enormous dataset from various imaging modalities and provide assistance in the field of oncology. This review focused on the applications of AI in the early diagnosis and prevention of OC. A literature search was conducted in the PubMed and Scopus databases using the search terminology “oral cancer” and “artificial intelligence.” Further information regarding the topic was collected by scrutinizing the reference lists of selected articles. Based on the information obtained, this article reviews and discusses the applications and advantages of AI in OC screening, early diagnosis, disease prediction, treatment planning, and prognosis. Limitations and the future scope of AI in OC research are also highlighted.

Design, fabrication and testing of 3D printed smartphone-based device for collection of intrinsic fluorescence from human cervix

Fluorescence spectroscopy has the potential to identify discriminatory signatures, crucial for early diagnosis of cervical cancer. We demonstrate here the design, fabrication and testing of a 3D printed smartphone based spectroscopic device. Polarized fluorescence and elastic scattering spectra are captured through the device using a 405 nm laser and a white LED source respectively. The device has been calibrated by comparison of spectra of standard fluorophores (Flavin adenine dinucleotide, fluorescein, rhodamine, and porphyrin) with the corresponding spectra collected from a commercial spectrometer. A few cervical tissue spectra have also been captured for proof of its applicability as a portable, standalone device for the collection of intrinsic fluorescence spectra from human cervix.

The Effectiveness of Artificial Intelligence in Detection of Oral Cancer

Aim

The early detection of oral cancer (OC) at the earliest stage significantly increases survival rates. Recently, there has been an increasing interest in the use of artificial intelligence (AI) technologies in diagnostic medicine. This study aimed to critically analyse the available evidence concerning the utility of AI in the diagnosis of OC. Special consideration was given to the diagnostic accuracy of AI and its ability to identify the early stages of OC.

Materials and methods

From the date of inception to December 2021, 4 databases (PubMed, Scopus, EBSCO, and OVID) were searched. Three independent authors selected studies on the basis of strict inclusion criteria. The risk of bias and applicability were assessed using the prediction model risk of bias assessment tool. Of the 606 initial records, 17 studies with a total of 7245 patients and 69,425 images were included. Ten statistical methods were used to assess AI performance in the included studies. Six studies used supervised machine learning, whilst 11 used deep learning. The results of deep learning ranged with an accuracy of 81% to 99.7%, sensitivity 79% to 98.75%, specificity 82% to 100%, and area under the curve (AUC) 79% to 99.5%.

Results

Results obtained from supervised machine learning demonstrated an accuracy ranging from 43.5% to 100%, sensitivity of 94% to 100%, specificity 16% to 100%, and AUC of 93%.

Conclusions

There is no clear consensus regarding the best AI method for OC detection. AI is a valuable diagnostic tool that represents a large evolutionary leap in the detection of OC in its early stages. Based on the evidence, deep learning, such as a deep convolutional neural network, is more accurate in the early detection of OC compared to supervised machine learning.

Diagnostic tests for oral cancer and potentially malignant disorders in patients presenting with clinically evident lesions

BACKGROUND

Squamous cell carcinoma is the most common form of malignancy of the oral cavity, and is often proceeded by oral potentially malignant disorders (OPMD). Early detection of oral cavity squamous cell carcinoma (oral cancer) can improve survival rates. The current diagnostic standard of surgical biopsy with histology is painful for patients and involves a delay in order to process the tissue and render a histological diagnosis; other diagnostic tests are available that are less invasive and some are able to provide immediate results. This is an update of a Cochrane Review first published in 2015.

OBJECTIVES

Primary objective: to estimate the diagnostic accuracy of index tests for the detection of oral cancer and OPMD, in people presenting with clinically evident suspicious and innocuous lesions.

SECONDARY OBJECTIVE

to estimate the relative accuracy of the different index tests.

SEARCH METHODS

Cochrane Oral Health's Information Specialist searched the following databases: MEDLINE Ovid (1946 to 20 October 2020), and Embase Ovid (1980 to 20 October 2020). The US National Institutes of Health Ongoing Trials Register (ClinicalTrials.gov) and the World Health Organization International Clinical Trials Registry Platform were also searched for ongoing trials to 20 October 2020. No restrictions were placed on the language or date of publication when searching the electronic databases. We conducted citation searches, and screened reference lists of included studies for additional references.

SELECTION CRITERIA

We selected studies that reported the diagnostic test accuracy of the following index tests when used as an adjunct to conventional oral examination in detecting OPMD or oral cavity squamous cell carcinoma: vital staining (a dye to stain oral mucosa tissues), oral cytology, light-based detection and oral spectroscopy, blood or saliva analysis (which test for the presence of biomarkers in blood or saliva).

DATA COLLECTION AND ANALYSIS

Two review authors independently screened titles and abstracts for relevance. Eligibility, data extraction and quality assessment were carried out by at least two authors, independently and in duplicate. Studies were assessed for methodological quality using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2). Meta-analysis was used to combine the results of studies for each index test using the bivariate approach to estimate the expected values of sensitivity and specificity.

MAIN RESULTS

This update included 63 studies (79 datasets) published between 1980 and 2020 evaluating 7942 lesions for the quantitative meta-analysis. These studies evaluated the diagnostic accuracy of conventional oral examination with: vital staining (22 datasets), oral cytology (24 datasets), light-based detection or oral spectroscopy (24 datasets). Nine datasets assessed two combined index tests. There were no eligible diagnostic accuracy studies evaluating blood or salivary sample analysis. Two studies were classed as being at low risk of bias across all domains, and 33 studies were at low concern for applicability across the three domains, where patient selection, the index test, and the reference standard used were generalisable across the population attending secondary care. The summary estimates obtained from the meta-analysis were: - vital staining: sensitivity 0.86 (95% confidence interval (CI) 0.79 to 0.90) specificity 0.68 (95% CI 0.58 to 0.77), 20 studies, sensitivity low-certainty evidence, specificity very low-certainty evidence; - oral cytology: sensitivity 0.90 (95% CI 0.82 to 0.94) specificity 0.94 (95% CI 0.88 to 0.97), 20 studies, sensitivity moderate-certainty evidence, specificity moderate-certainty evidence; - light-based: sensitivity 0.87 (95% CI 0.78 to 0.93) specificity 0.50 (95% CI 0.32 to 0.68), 23 studies, sensitivity low-certainty evidence, specificity very low-certainty evidence; and - combined tests: sensitivity 0.78 (95% CI 0.45 to 0.94) specificity 0.71 (95% CI 0.53 to 0.84), 9 studies, sensitivity very low-certainty evidence, specificity very low-certainty evidence.

AUTHORS' CONCLUSIONS

At present none of the adjunctive tests can be recommended as a replacement for the currently used standard of a surgical biopsy and histological assessment. Given the relatively high values of the summary estimates of sensitivity and specificity for oral cytology, this would appear to offer the most potential. Combined adjunctive tests involving cytology warrant further investigation. Potentially eligible studies of blood and salivary biomarkers were excluded from the review as they were of a case-control design and therefore ineligible. In the absence of substantial improvement in the tests evaluated in this updated review, further research into biomarkers may be warranted.

Band-Selection of a Portal LED-Induced Autofluorescence Multispectral Imager to Improve Oral Cancer Detection

This aim of this study was to find effective spectral bands for the early detection of oral cancer. The spectral images in different bands were acquired using a self-made portable light-emitting diode (LED)-induced autofluorescence multispectral imager equipped with 365 and 405 nm excitation LEDs, emission filters with center wavelengths of 470, 505, 525, 532, 550, 595, 632, 635, and 695 nm, and a color image sensor. The spectral images of 218 healthy points in 62 healthy participants and 218 tumor points in 62 patients were collected in the ex vivo trials at China Medical University Hospital. These ex vivo trials were similar to in vivo because the spectral images of anatomical specimens were immediately acquired after the on-site tumor resection. The spectral images associated with red, blue, and green filters correlated with and without nine emission filters were quantized by four computing method, including summated intensity, the highest number of the intensity level, entropy, and fractional dimension. The combination of four computing methods, two excitation light sources with two intensities, and 30 spectral bands in three experiments formed 264 classifiers. The quantized data in each classifier was divided into two groups: one was the training group optimizing the threshold of the quantized data, and the other was validating group tested under this optimized threshold. The sensitivity, specificity, and accuracy of each classifier were derived from these tests. To identify the influential spectral bands based on the area under the region and the testing results, a single-layer network learning process was used. This was compared to conventional rules-based approaches to show its superior and faster performance. Consequently, four emission filters with the center wavelengths of 470, 505, 532, and 550 nm were selected by an AI-based method and verified using a rule-based approach. The sensitivities of six classifiers using these emission filters were more significant than 90%. The average sensitivity of these was about 96.15%, the average specificity was approximately 69.55%, and the average accuracy was about 82.85%.

Portable LED-induced autofluorescence spectroscopy for oral cancer diagnosis

Oral cancer is a serious and growing problem in many developing and developed countries. To improve the cancer screening procedure, we developed a portable light-emitting-diode (LED)-induced autofluorescence (LIAF) imager that contains two wavelength LED excitation light sources and multiple filters to capture ex vivo oral tissue autofluorescence images. Compared with conventional means of oral cancer diagnosis, the LIAF imager is a handier, faster, and more highly reliable solution. The compact design with a tiny probe allows clinicians to easily observe autofluorescence images of hidden areas located in concave deep oral cavities. The ex vivo trials conducted in Taiwan present the design and prototype of the portable LIAF imager used for analyzing 31 patients with 221 measurement points. Using the normalized factor of normal tissues under the excitation source with 365 nm of the central wavelength and without the bandpass filter, the results revealed that the sensitivity was larger than 84%, the specificity was not smaller than over 76%, the accuracy was about 80%, and the area under curve of the receiver operating characteristic (ROC) was achieved at about 87%, respectively. The fact shows the LIAF spectroscopy has the possibilities of ex vivo diagnosis and noninvasive examinations for oral cancer.

Diagnostic tests for oral cancer and potentially malignant disorders in patients presenting with clinically evident lesions

BACKGROUND

Oral squamous cell carcinoma is the most common form of malignancy of the lip and oral cavity, often being proceeded by potentially malignant disorders (PMD). Early detection can reduce the malignant transformation of PMD and can improve the survival rate for oral cancer. The current standard of scalpel biopsy with histology is painful for patients and involves a delay whilst histology is completed; other tests are available that are unobtrusive and provide immediate results.

OBJECTIVES

PRIMARY OBJECTIVE

To estimate the diagnostic accuracy of index tests for the detection of oral cancer and PMD of the lip and oral cavity, in people presenting with clinically evident lesions.

SECONDARY OBJECTIVE

To estimate the relative accuracy of the different index tests.

SEARCH METHODS

The electronic databases were searched on 30 April 2013. We searched MEDLINE (OVID) (1946 to April 2013) and four other electronic databases (the Cochrane Diagnostic Test Accuracy Studies Register, the Cochrane Oral Health Group's Trials Register, EMBASE (OVID) and MEDION (Ovid)). There were no restrictions on language in the searches of the electronic databases. We conducted citation searches and screened reference lists of included studies for additional references.

SELECTION CRITERIA

We selected studies that reported the diagnostic test accuracy of the following index tests when used as an adjunct to conventional oral examination in detecting PMD or oral squamous cell carcinoma of the lip or oral cavity: vital staining, oral cytology, light-based detection and oral spectroscopy, blood or saliva analysis (which test for the presence of biomarkers in blood or saliva).

DATA COLLECTION AND ANALYSIS

Two review authors independently screened titles and abstracts for relevance. Eligibility, data extraction and quality assessment were carried out by at least two authors, independently and in duplicate. Studies were assessed for methodological quality using QUADAS-2. Meta-analysis was used to combine the results of studies for each index test using the bivariate approach to estimate the expected values of sensitivity and specificity.

MAIN RESULTS

We included 41 studies, recruiting 4002 participants, in this review. These studies evaluated the diagnostic accuracy of conventional oral examination with: vital staining (14 studies), oral cytology (13 studies), light-based detection or oral spectroscopy (13 studies). Six studies assessed two combined index tests. There were no eligible diagnostic accuracy studies evaluating blood or salivary sample analysis.The summary estimates for vital staining obtained from the meta-analysis were sensitivity of 0.84 (95% CI 0.74 to 0.90) with specificity of 0.70 (0.59 to 0.79), with 14 studies were included in the meta-analysis. For cytology, sensitivity was 0.91 (0.81 to 0.96) and specificity was 0.91 (0.81 to 0.95) with 12 studies included in the meta-analysis. For light-based detection, sensitivity was 0.91 (0.77 to 0.97) and specificity was 0.58 (0.22 to 0.87) with 11 studies included in the meta-analysis. The relative test accuracy was assessed by adding covariates to the bivariate analysis, no difference in model fit was observed.

AUTHORS' CONCLUSIONS

The overall quality of the included studies was poor. None of the adjunctive tests can be recommended as a replacement for the currently used standard of a scalpel biopsy and histological assessment. Given the relatively high values of the summary estimates of sensitivity and specificity for cytology, this would appear to offer the most potential. Combined adjunctive tests involving cytology warrant further investigation.

Assessment of the sensitivity and specificity of tissue-specific-based and anatomical-based optical biomarkers for rapid detection of human head and neck squamous cell carcinoma

OBJECTIVES

We propose the use of morphological optical biomarkers for rapid detection of human head and neck squamous cell carcinoma (HNSCC) by leveraging the underlying tissue characteristics in aerodigestive tracts.

MATERIALS AND METHODS

Diffuse reflectance spectra were obtained from malignant and contra-lateral normal tissues of 57 patients undergoing panendoscopy and biopsy. Oxygen saturation, total hemoglobin concentration, and the reduced scattering coefficient were extracted. Differences in malignant and normal tissues were examined based on two different groupings: anatomical site and morphological tissue type.

RESULTS AND CONCLUSIONS

Measurements were acquired from 252 sites, of which 51 were pathologically classified as SCC. Optical biomarkers exhibited statistical differences between malignant and normal samples. Contrast was enhanced when parsing tissues by morphological classification rather than anatomical subtype for unpaired comparisons. Corresponding linear discriminant models using multiple optical biomarkers showed improved predictive ability when accounting for morphological classification, particularly in node-positive lesions. The false-positive rate was retrospectively found to decrease by 34.2% in morphologically- vs. anatomically-derived predictive models. In glottic tissue, the surgeon exhibited a false-positive rate of 45.7% while the device showed a lower false-positive rate of 12.4%. Additionally, comparisons of optical parameters were made to further understand the physiology of tumor staging and potential causes of high surgeon false-positive rates. Optical spectroscopy is a user-friendly, non-invasive tool capable of providing quantitative information to discriminate malignant from normal head and neck tissues. Predictive models demonstrated promising results for real-time diagnostics. Furthermore, the strategy described appears to be well suited to reduce the clinical false-positive rate.

Multispectral characterization of tissues encountered during laparoscopic colorectal surgery

AIMS

This study investigated the feasibility of automated differentiation between essential tissue types encountered during laparoscopic colorectal surgery using spectral analysis.

METHODS

Wide band (440-1830 nm) spectra were collected using an optical fiber probe and spectrometer from freshly explanted, ex vivo, human colonic specimens. These data were normalized at 810 nm (an isobestic wavelength for hemoglobin and oxy-hemoglobin) and mathematically analyzed using total principal component regression (TPCR).

RESULTS

929 spectra were collected from specimens of 19 patients, distinguishing 5 tissue types: mesenteric fat (MF, n=269), blood vessels (BV, n=377), colonic tissue (CT, n=213), ureter (UR, n=10) and tumorous tissue in colon (TT, n=60). For each individual tissue type the distinctive ability was determined against all other tissue types pooled as a group. Paired probability density function (PDF) of "tissue" (centered around label 1) versus "all other pooled tissues" (centered around label 0) and the cumulative distribution function (CDF) at label crossover value 0.5 was determined for each tissue type (MF: CDF=0.99 [SD=0.19]; BV: CDF=0.95 [SD=0.29]; CT: CDF=0.98 [SD=0.22]; UR: CDF=0.99 [SD=0.09]; TT: CDF=0.99 [SD=0.18]).

CONCLUSION

Automated spectral differentiation of blood vessel, ureter, mesenteric adipose tissue, colonic tissue and tumorous tissue in colon, is feasible in freshly explanted human colonic specimens. These results may be exploited for further steps toward multi- or hyperspectrally enhanced in vivo (laparoscopic) surgical imaging.

Prospective evaluation of a portable depth-sensitive optical spectroscopy device to identify oral neoplasia

A portable, depth-sensitive clinical spectroscopy device for noninvasive early diagnosis of oral cancer is described. We carried out a pilot study to evaluate the ability of the device to identify oral neoplasia using a previously developed diagnostic algorithm. A total of 79 oral sites in 33 subjects, including 28 patients with oral lesions and 5 healthy volunteers, were measured and analyzed. Measurements of 54 nonkeratinized oral sites yielded an area under the receiver operating characteristic curve of 0.90. Measurements of 25 keratinized oral sites yielded an area under the receiver operating characteristic curve of 0.83.

Noninvasive evaluation of oral lesions using depth-sensitive optical spectroscopy

BACKGROUND

Optical spectroscopy is a noninvasive technique with potential applications for diagnosis of oral dysplasia and early cancer. In this study, we evaluated the diagnostic performance of a depth-sensitive optical spectroscopy (DSOS) system for distinguishing dysplasia and carcinoma from non-neoplastic oral mucosa.

METHODS

Patients with oral lesions and volunteers without any oral abnormalities were recruited to participate. Autofluorescence and diffuse reflectance spectra of selected oral sites were measured using the DSOS system. A total of 424 oral sites in 124 subjects were measured and analyzed, including 154 sites in 60 patients with oral lesions and 270 sites in 64 normal volunteers. Measured optical spectra were used to develop computer-based algorithms to identify the presence of dysplasia or cancer. Sensitivity and specificity were calculated using a gold standard of histopathology for patient sites and clinical impression for normal volunteer sites.

RESULTS

Differences in oral spectra were observed in: (1) neoplastic versus nonneoplastic sites, (2) keratinized versus nonkeratinized tissue, and (3) shallow versus deep depths within oral tissue. Algorithms based on spectra from 310 nonkeratinized anatomic sites (buccal, tongue, floor of mouth, and lip) yielded an area under the receiver operating characteristic curve of 0.96 in the training set and 0.93 in the validation set.

CONCLUSIONS

The ability to selectively target epithelial and shallow stromal depth regions appeared to be diagnostically useful. For nonkeratinized oral sites, the sensitivity and specificity of this objective diagnostic technique were comparable to that of clinical diagnosis by expert observers. Thus, DSOS has potential to augment oral cancer screening efforts in community settings.

Autofluorescence and diffuse reflectance spectroscopy of oral epithelial tissue using a depth-sensitive fiber-optic probe

Optical spectroscopy can provide useful diagnostic information about the morphological and biochemical changes related to the progression of precancer in epithelial tissue. As precancerous lesions develop, the optical properties of both the superficial epithelium and underlying stroma are altered; measuring spectral data as a function of depth has the potential to improve diagnostic performance. We describe a clinical spectroscopy system with a depth-sensitive, ball lens coupled fiber-optic probe for noninvasive in vivo measurement of oral autofluorescence and diffuse reflectance spectra. We report results of spectroscopic measurements from oral sites in normal volunteers and in patients with neoplastic lesions of the oral mucosa; results indicate that the addition of depth selectivity can enhance the detection of optical changes associated with precancer.

Performance analysis of three-class classifiers: properties of a 3-D ROC surface and the normalized volume under the surface for the ideal observer

Classification of a given observation to one of three classes is an important task in many decision processes or pattern recognition applications. A general analysis of the performance of three-class classifiers results in a complex 6-D receiver operating characteristic (ROC) space, for which no simple analytical tool exists at present. We investigate the performance of an ideal observer under a specific set of assumptions that reduces the 6-D ROC space to 3-D by constraining the utilities of some of the decisions in the classification task. These assumptions lead to a 3-D ROC space in which the true-positive fraction (TPF) can be expressed in terms of the two types of false-positive fractions (FPFs). We demonstrate that the TPF is uniquely determined by, and therefore is a function of, the two FPFs. The domain of this function is shown to be related to the decision boundaries in the likelihood ratio plane. Based on these properties of the 3-D ROC space, we can define a summary measure, referred to as the normalized volume under the surface (NVUS), that is analogous to the area under the ROC curve (AUC) for a two-class classifier. We further investigate the properties of the 3-D ROC surface and the NVUS for the ideal observer under the condition that the three class distributions are multivariate normal with equal covariance matrices. The probability density functions (pdfs) of the decision variables are shown to follow a bivariate log-normal distribution. By considering these pdfs, we express the TPF in terms of the FPFs, and integrate the TPF over its domain numerically to obtain the NVUS. In addition, we performed a Monte Carlo simulation study, in which the 3-D ROC surface was generated by empirical "optimal" classification of case samples in the multidimensional feature space following the assumed distributions, to obtain an independent estimate of NVUS. The NVUS value obtained by using the analytical pdfs was found to be in good agreemen- t with that obtained from the Monte Carlo simulation study. We also found that, under all conditions studied, the NVUS increased when the difficulty of the classification task was reduced by changing the parameters of the class distributions, thereby exhibiting the properties of a performance metric in analogous to AUC. Our results indicate that, under the conditions that lead to our 3-D ROC analysis, the performance of a three-class classifier may be analyzed by considering the ROC surface, and its accuracy characterized by the NVUS.