Project Summary/Abstract Endoscopic surveillance of Barrett’s esophagus (BE) is recommended for early diagnosis and treatment of neoplasia (i.e., esophageal adenocarcinoma and high-grade dysplasia). However, neoplasia is difficult to detect on regular white-light endoscopy (WLE; sensitivity 64%), and 26% of neoplasia is missed with WLE alone. On the other hand, confocal high-resolution microendoscopy (cHRME) is a low-cost, portable, reusable imaging technology that provides microscopic “optical biopsy” images of the esophageal mucosa at the time of endoscopy and has sensitivity of neoplasia detection upwards of 89% in the hands of experts. Despite these advantages, the dissemination of cHRME is limited by the availability of expert microendoscopists capable of interpreting these histopathology-like images. Artificial intelligence algorithms that automate interpretation of cHRME images could bridge this gap by providing a real-time computer-assisted diagnosis to users in community-based BE surveillance settings and reducing the need for expert review. My objective is to develop and validate an automated software algorithm for real-time BE neoplasia detection using cHRME. Furthermore, I will optimize the software algorithm by incorporating traditional clinical risk factors for comprehensive risk stratification. Thus, I propose the following Specific Aims: Aim 1: Technology Development: To develop a software algorithm that automates interpretation of cHRME images in BE neoplasia detection. Aim 2. Technology Evaluation and Optimization: (a) To validate the cHRME automated software algorithm for real-time neoplasia detection in BE; (b) To optimize the automated software algorithm by integrating demographic, lifestyle, and clinical risk factors for comprehensive BE neoplasia detection. Aim 3. Technology Acceptability: (a) To evaluate the acceptability and experiences of endoscopists using computer-assisted diagnosis; (b) To assess feasibility of the automated software algorithm in clinical BE neoplasia detection. The overarching goal of this proposal is to develop artificial intelligence algorithms that facilitate dissemination of novel, low-cost technologies into community settings for rapid, real-time, accurate neoplasia detection. Future longitudinal studies will focus on validation of comprehensive risk models that use macroscopic and microscopic metrics to predict future neoplasia risk in BE patients undergoing surveillance endoscopy. My long-term goal is to become an independently funded investigator in novel techniques for early gastrointestinal cancer detection. I have assembled an experienced mentoring committee comprised of senior, funded investigators with expertise in technology development, artificial intelligence algorithms, epidemiology, bioinformatics, and qualitative research. My career development plan includes additional formal research training in artificial intelligence methodology, machine learning, and clinical trials. With the support and protected time provided by the K23 Career Development Award, I will be able to complete my proposed research and career development goals and generate preliminary data to be competitive for independent research funding.

Project Narrative Esophageal cancer rates are increasing in the United States, and surveillance of Barrett’s esophagus (BE) may prevent esophageal cancer. I propose to develop and evaluate an artificial intelligence algorithm capable of diagnosing cancer within BE at the time of endoscopy using confocal high-resolution microendoscopy (cHRME), a novel technology capable of cellular imaging of the esophagus. This algorithm would provide an immediate, accurate computer-assisted diagnosis, which would allow this technology to be widely used during BE surveillance in community and low-resource settings with substantial public health impact.