AI-Assisted Microendoscopy for the Early Detection of Esophageal Cancer

Funding Agency:

NIH / National Cancer Institute

Collaborators:

Sharmila Anandasabapathy (Baylor CoM), Rebecca Richards-Kortum (Rice)

Overview:

Our group developed and evaluated a mobile, battery-powered, high-resolution microendoscope (mHRME) that provides 1100x-magnified images of the epithelium during evaluation for esophageal squamous cell neoplasia (ESCN). This low-cost device provides an “optical biopsy” (microscopic images of epithelium and cell nuclei),allowing real-time identification of neoplasia. In our 2015 single-arm trial in China and the USA with academic endoscopists, mHRME increased the specificity of standard Lugol’s endoscopy from 29% to 79%,while maintaining sensitivity (95%). In our original parent R01, we completed: (1) a randomized controlled trial (USA and China; n=918) of mHRME with visual interpretation inpatients undergoing screening or surveillance for ESCN; (2) software algorithms for automated detection of neoplastic images; and (3) a pilot study (n=41subjects) of the software-assisted mHMRE in Brazil. Despite COVID-related delays in China, our trial revealed significantly higher specificity for visual interpretation by experts versus novices in the surveillance arm (100% vs. 19%, p <0.05). Sensitivity was unaffected. In the screening arm, diagnostic yield (neoplastic biopsies/total biopsies) increased 3.6 times (8 to 29%); 16% of patients were correctly spared any biopsy, and 18% had a change in clinical plan.Using a small set of images, we then developed initial algorithms for interpretation of mHRME images to overcome a key barrier: need for user expertise. We conducted a single-arm pilot study evaluating an artificial intelligence-based mobile HRME (AI-mHRME) in 41 Brazilian participants undergoing endoscopic screening. This study(completed Aug. 2022) confirmed that AI-mHRME doubled diagnostic yield, improved endoscopist confidence, and had significant clinical impact (clinical plan changed in64%of cases). Algorithms performed with a sensitivity/specificity of 100%/85%.Endoscopists uniformly said they favored an AI-guided approach but were concerned about implementation(regulatory barriers, contextual fit, user training, and patient acceptance).In this competitive renewal, our goal is to build on our experience to date to optimize and evaluate (performance, clinical impact, barriers and acceptability) AI-mHRME, a low-cost (<$2,500) mobile device for the early detection of esophageal cancer in ethnically and socioeconomically diverse populations in the USA and Brazil (COVID related issues have precluded re-engaging in China). We will now leverage our prior work to optimize and evaluate our portable, AI-guided, low-cost ‘optical biopsy’ approach for esophageal cancer management in the USA and Brazil. Performance, effectiveness, and stakeholder/contextual data obtained will help facilitate implementation and dissemination of our innovative computer-assisted cancer prevention strategy in diverse global settings and in diverse patient populations across a variety of epithelial cancers (e.g. esophagus, stomach, colon, skin, oral, anal).