Computational Algorithms for Automated Early Diagnosis of Oral Cancer Based on Multispectral Fluorescence Lifetime Imaging Endoscopy
Abstract
Oral cancer is one of the most common types of cancer in the US, killing around 8,000 people each year. Patients diagnosed at advanced stages have only a 40% chance of survival and commonly require painful and highly invasive surgery to remove parts of the oral cavity. In contrast, patients diagnosed early usually require minor surgery and have an 84% chance of survival. Therefore, early detection holds great promise for improving both the survival rate and quality of life of these patients. Unfortunately, only three in ten patients are diagnosed at early stages since benign oral lesions are often difficult to distinguish from early stage cancer. Moreover, tissue from a biopsy may register as benign, but the surrounding tissue that was not biopsied can be cancerous and remain undiagnosed, resulting in increased odds of local recurrence and lower survival rates. Hence, there is an urgent need for technologies for accurate, fast, and reliable screening of oral cancer.
This dissertation addresses these challenges in the diagnosis of oral cancer and precancer by making use of an optical technology called Fluorescence Lifetime Imaging (FLIM) endoscopy for the non-invasive imaging of clinically suspicious oral lesions in patients. Multispectral autofluorescence lifetime images of benign, precancerous, and cancerous oral lesions from 125 patients were acquired in vivo using a novel multispectral FLIM endoscope. These images were processed to generate widefield maps of biochemical and metabolic autofluorescence biomarkers of oral cancer and precancer. Statistical analyses applied to the quantified multispectral autofluorescence biomarkers indicated their potential to provide contrast between precancerous/cancerous vs. healthy oral tissue and precancerous/cancerous vs. benign oral tissue. Machine learning algorithms based on the most promising autofluorescence biomarkers of oral cancer and precancer were designed to discriminate precancerous/cancerous oral lesions vs. healthy oral tissue, and precancerous/cancerous vs. benign oral lesions. The results of this innovative study demonstrate the potentials of a computer-aided detection system based on endogenous multispectral autofluorescence endoscopy as a novel non-invasive clinical tool for oral cancer and precancer screening.
Subject
Oral cancer and dysplasiaFluorescence lifetime imaging (FLIM)
Autofluorescence biomarkers
Statistical analysis
Computer aided diagnosis (CAD).
Citation
Duran Sierra, Elvis de Jesus (2020). Computational Algorithms for Automated Early Diagnosis of Oral Cancer Based on Multispectral Fluorescence Lifetime Imaging Endoscopy. Doctoral dissertation, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /200768.