Multispectral narrow band imaging (NBI) represents a promising screening tool for oropharyngeal cancer (OPC) due to its selective enhancement of tumor vasculature. However, most implementations of NBI for cancer screening have relied on qualitative observations of mucosa (e.g., presence of capillary loops), rather than quantitative measures based on image features. This preliminary study was designed to determine if specific narrow band signal characteristics of tumor vasculature may be used to train a machine learning algorithm for quantitative OPC detection. Adult patients with no tumor (5) and biopsy-proven oropharyngeal squamous cell carcinoma (25) were recruited, and consent was obtained to examine the oropharynx by endoscopy under white light endoscopy (WLE) and narrow band imaging (NBI). De-identified WLE and NBI laryngoscopy videos were then read frame-by-frame into Matlab 2011a and sampled regions of mucosa were processed to extract color (a, b values in L*a*b image space) and texture (pixel entropy) information. Color and texture image features of tumor and non-tumor sides for 5 patients (4 tumor, 1 healthy) were used to train a Naïve Bayesian classifier for the remaining 25 (21 tumor, 4 healthy), and performance under WB and NBI was compared by ROC analysis. Compared to WLE, mNBI significantly enhanced the performance of a Naïve Bayesian classifier trained on low-level image features of oropharyngeal mucosa (76.6% AUC vs 50.5% for WLE). These findings suggest that automated clinical detection of oropharyngeal carcinoma could be used to enhance surgical vision, improve early diagnosis, and allow for high-throughput screening.
Shamik Mascharak, Brandon Jackson Baird, and F. Christopher Holsinger, "Automated detection of oropharyngeal carcinoma using multispectral imaging: a prospective feasibility study (Conference Presentation)," Proc. SPIE 10469, Optical Imaging, Therapeutics, and Advanced Technology in Head and Neck Surgery and Otolaryngology 2018, 104690H (Presented at SPIE BiOS: January 27, 2018; Published: 14 March 2018); https://doi.org/10.1117/12.2300894.5751470598001.
Conference Presentations are recordings of oral presentations given at SPIE conferences and are published as part of the conference proceedings. They include the speaker's narration along with video of the slides and animations. Most also include full-text papers. Interactive, searchable transcripts and closed captioning are now available for 2018 presentations, with transcripts for prior recordings added daily.
Search and browse our growing collection of more than 16,000 conference presentations, including many plenary and keynote presentations.