Artificial Intelligence for the Management of Colorectal Neoplasia Using Combined-Modality Spectroscopy and Enhanced Imaging

点击翻译按钮获取中文摘要

Objectives: The overarching objective of this proposal is to develop, validate, and deploy an artificial intelligence (AI)-based low-cost platform to make endoscopic prevention of colorectal cancer (CRC) more efficient. We seek to leverage our work in spectroscopic biopsy tools and automated endoscopic imaging interpretation to create an accurate and widely-adoptable, real-time histology (RTH) platform based on combined optical modalities and machine learning. At present, colonoscopic CRC prevention hinges on the complete removal and histopathological assessment of all polyps. This practice results in the removal of large numbers of polyps that have negligible malignant potential. As such, there is a widely-recognized need for simple, rapid, and low-cost methods for “smart” polyp assessment in real time to decrease biopsy costs and risks. To this end, major professional societies, led by The American Society for Gastrointestinal Endoscopy (ASGE), have endorsed the purely optical management of diminutive polyps and have put forth guidelines and acceptable performance thresholds (i.e. the PIVI statements) for eventual adoption. The past decade has seen an explosion in biophotonic technologies toward diagnosing and treating colorectal neoplasia more precisely. While several PIVI thresholds for diminutive colonic polyps have been met, prospective testing in non-academic settings has fallen short due to the barriers of operator skill and experience. Recent advances in machine learning/artificial intelligence, and their application to endoscopic imaging, have shown promise for automating RTH to overcome operator factors. Such capability would finally open the door to widespread adoption of cost-saving resect-and-discard and leave-behind paradigms for diminutive polyps. On this front, we will build on our work using elastic scattering spectroscopy (ESS) biopsy tools, which has shown great promise for RTH, combining it with computer- assisted diagnosis (CAD) of endoscopic images. We hypothesize that the novel combination of these complementary AI based technologies will lead to a highly-accurate, minimally-disruptive, and widely- deployable approach for RTH of colorectal polyps. The specific aims for the present project are: 1. Develop AI models for computer assisted RTH based on spectroscopy and endoscopic images; 2. Implement system enhancements and tool design for multisite deployment; 3. Perform a multisite clinical study using AI-based RTH based on the combination of ESS and CAD of endoscopic images. Methodology: First, we will conduct a clinical study at VA Boston in which we will collect ESS measurements and endoscopic images of polyps at colonoscopy. We will use this paired data, correlated to clinical features and histopathology to design and validate AI algorithms for computer assisted RTH of colorectal polyps (including serrated lesions) that utilize both sources of optical information. Concurrently, we will prototype and build the next-generation ESS system, based on a new design that dramatically reduces the hardware footprint and cost. We will also design and prototype reprocessable ESS probes for integration into standard polypectomy snares. Finally, we will conduct a multisite clinical study involving three other VA facilities where the work described above will be deployed. The primary endpoint of this aim will be to evaluate the performance of ESS and CAD of endoscopic images separately and in combination toward PIVI thresholds. As secondary endpoints, we will use the clinical study to evaluate and improve our clinical systems.