Esophageal adenocarcinoma (EAC) is among the most lethal malignancies with a 19% five-year survival rate and its incidence has increased several fold in the last decades. Barrett’s esophagus (BE) confers elevated risk for progression to EAC. Patients diagnosed with BE undergo periodic surveillance endoscopy with biopsies to detect dysplasia which can be treated by endoscopic eradication with radiofrequency ablation before it progresses to EAC. However, the majority of diagnosed EAC patients have not had prior screening endoscopy and present with advanced lesions that limit treatment options and result in poorer survival. The development of a rapid, low cost, well tolerated, non-endoscopic BE screening technique that can be performed in unsedated patients at points of care outside the endoscopy suite would improve BE detection and reduce EAC morbidity and mortality. Our program is a multidisciplinary collaboration among investigators at the Massachusetts Institute Technology and Veteran Affairs Boston Healthcare System / Harvard Medical School that integrates novel optical imaging and software design, preclinical studies in swine, clinical studies in patients, and advanced image processing / machine learning. Aim 1 will develop an omniview tethered capsule technology that generates a map of the esophageal mucosa over a multi-centimeter length of esophagus and a series of wide angle forward views to aid navigation as the capsule is swallowed or retracted. The images will resemble endoscopic white light or narrow band imaging, but will not suffer from perspective distortion present in standard endoscopic or video capsule images. This will facilitate development of automated BE detection algorithms as well as enhance their sensitivity and specificity. This aim will also perform imaging studies in swine as a translational step toward clinical studies. Aim 2 will determine reader sensitivity and specificity for BE detection versus standard endoscopy / biopsy and prepare data for developing automated BE detection. Patients undergoing screening as well as with history of BE undergoing surveillance will be recruited and unsedated capsule imaging will be performed on the same day prior to their endoscopy. Sensitivity and specificity for detecting BE will be assessed using multiple blinded readers and data sets suitable for developing automated BE detection algorithms will be developed. Aim 3 will develop image analysis methods for automated BE detection by investigating classifiers that operate on handcrafted features (colors and textures) and modern deep convolutional neural network methods for direct classification. If successful, this program will develop a rapid, low cost and scalable method for BE screening that would not require patient sedation, endoscopy, or tissue acquisition, and which could be performed in community primary care clinics. The procedure would be much faster and many times lower cost than endoscopy. Automated BE detection would ena...