The Laryngoscope V C 2018 The American Laryngological, Rhinological and Otological Society, Inc. Detecting Oropharyngeal Carcinoma Using Multispectral, Narrow-Band Imaging and Machine Learning Shamik Mascharak, BS; Brandon J. Baird, MD; F. Christopher Holsinger, MD, FACS Objective: To determine if multispectral narrow-band imaging (mNBI) can be used for automated, quantitative detection of oropharyngeal carcinoma (OPC). Study Design: Prospective cohort study. Methods: Multispectral narrow-band imaging and white light endoscopy (WLE) were used to examine the lymphoepi- thelial tissues of the oropharynx in a preliminary cohort of 30 patients (20 with biopsy-proven OPC, 10 healthy). Low-level image features from five patients were then extracted to train na € ıve Bayesian classifiers for healthy and malignant tissue. Results: Tumors were classified by color features with 65.9% accuracy, 66.8% sensitivity, and 64.9% specificity under mNBI. In contrast, tumors were classified with 52.3% accuracy (P 5 0.0108), 44.8% sensitivity (P 5 0.0793), and 59.9% spe- cificity (P 5 0.312) under WLE. Receiver operating characteristic analysis yielded areas under the curve (AUC) of 72.3% and 54.6% for classification under mNBI and WLE, respectively (P 5 0.00168). For classification by both color and texture fea- tures, AUC under mNBI increased (80.1%, P 5 0.00230) but did not improve under WLE (below 55% for both models, P 5 0.180). Cross-validation with five folds yielded an AUC above 80% for both mNBI models and below 55% for both WLE models (P 5 0.0000410 and 0.000116). Conclusion: Compared to WLE, mNBI significantly enhanced the performance of a na € ıve Bayesian classifier trained on low-level image features of oropharyngeal mucosa. These findings suggest that automated clinical detection of OPC might be used to enhance surgical vision, improve early diagnosis, and allow for high-throughput screening. Key Words: Multispectral imaging, narrow-band imaging, machine learning, na € ıve Bayesian classification, oropharyngeal carcinoma, surgical vision, head and neck. Level of Evidence: NA. Laryngoscope, 00:000–000, 2018 INTRODUCTION In the United States, the landscape of head and neck cancer has been transformed with an unexpected rise in the number of oropharyngeal cancer cases. Reports from the United States 1,2 and abroad 3 have documented the precipitous rise in the incidence of oropharyngeal carci- noma (OPC) in individuals younger than typically expected for traditional head and neck cancer patients. 4 In 2017, nearly 50 thousand patients are expected to be diag- nosed with oral or pharyngeal cancer. 5 In a prospective case-control study, D’Souza and Gillison first demon- strated an association between human papillomavirus (HPV) and OPC. 6 Using the Surveillance, Epidemiology, and End Results repository, Chaturvedi reported that the incidence of HPV-positive oropharyngeal cancers increased by 225% from 1988 to 2004. 2 Patients with OPC due to HPV tend to present earlier in life when compared to those with tobacco-associated OPC. Often, the primary tumor is difficult to see, especially within the papillary lymphoepithelial folds of Waldeyer’s ring, and OPC is not diagnosed until the tumor metastasizes to the neck. Given these sweeping changes in the epidemiology of head and neck cancer, improvements in screening, preven- tion, and treatment are imperative. Multispectral narrow band imaging (mNBI) has recently been shown to differen- tially illuminate the intraepithelial papillary capillary loops of superficial tumors of the upper aerodigestive tract in esophageal, hypopharyngeal, and laryngeal carci- noma, 7–9 but the use of mNBI in OPC remains under- studied. 10 It is difficult to differentiate OPC against surrounding lymphoepithelial mucosa using standard white light endoscopy (WLE) due the irregular and innu- merable folds of Waldeyer’s ring, especially within the base of tongue. However, oropharyngeal tumors are known to be well vascularized. We therefore anticipated superior visualization of vascular features of OPC under mNBI. Prior studies utilizing mNBI endoscopy relied on qualitative assessments of the surrounding mucosal ves- sels by trained, blinded clinicians. 7–9,11–15 To improve reproducibility and increase sensitivity and specificity, we combined image processing and basic machine learning techniques to automate the assessment of oropharyngeal mucosa in 30 patients screened using WLE and mNBI endoscopy. Quantitative detection of OPC under mNBI Additional supporting information may be found in the online version of this article. From the Division of Head and Neck Surgery, Department of Oto- laryngology, School of Medicine, Stanford University (S.M., B.J.B., P .H., N.B., F .C.H.), Palo Alto, California, U.S.A. Editor’s Note: This Manuscript was accepted for publication February 1, 2018. The authors have no funding, financial relationships, or conflicts of interest to disclose. Send correspondence to F. Christopher Holsinger, MD, FACS, Stanford University, 875 Blake Wilbur Drive, Palo Alto, CA 94305-5820. E-mail: holsinger@stanford.edu DOI: 10.1002/lary.27159 Laryngoscope 00: Month 2018 Mascharak et al.: Multispectral Imaging of Oropharynx Cancer 1