VIDEO-BASED COMPUTER AIDED DETECTION SYSTEM DETECTS BARRETT’S NEOPLASIA WITH HIGH ACCURACY DURING LIVE ENDOSCOPIC PROCEDURES: A MULTI-CENTER PILOT AND FEASIBILITY STUDY

Background: Using EsoCheck (EC), a non-endoscopic balloon device for sampling the distal esophagus, coupled with EsoGuard (EG), a DNA based testing screening for Barrett’s esophagus (BE), we have previously reported high sensitivity and specificity for detection of BE among patients with known BE compared with controls. We now utilized this screening method in the population at risk for developing BE and esophageal adenocarcinoma (ECA).
Aim: Determine the accuracy of EG/EC in patients eligible for BE screening.
Method: We recruited veterans who meet American College of Gastroenterology (ACG) Guideline criteria for endoscopic BE screening at our VA hospital. Study participants completed EC/EG followed by esophagogastroduodenoscopy (EGD). We compared the yield of guideline based screening with a theoretical strategy where patients have EC/EG, and only those with a positive result then have an EGD.
Results: We enrolled 47 participants (89.3% male, median age 60.9 (SD 11.2)). 46/47 patients successfully swallowed the EC. In total, 10.6% (5/47) had an endoscopic diagnosis of BE. EC/EG was positive for 14 patients, negative for 23, and 9 patients did not have sufficient material for EG analysis. We found that 10.8% (4/37) patients had BE diagnosed by EGD, compared with 28.6% (4/14). There were no cases of BE diagnosed by endoscopy among patients with a negative EC/EG test. Compared with EGD the sensitivity and specificity of the test was 100% (95% CI 39.8% to 100%) and 69.7% (95% CI 51.3% to 84.4%), respectively. The PPV, NPV, and accuracy were 28.12% (95% CI 18.91% to 39.63%), 100.00%, 72.91% (95% CI 55.81% to 86.16%).
Conclusions: The performance of the EC/EG test was promising in the screening population and this study sets the ground for future larger trials of an EC/EG intervention for BE detection in the wider screening population.

Background and aims: The early diagnosis and intervention of esophageal squamous cell carcinoma (ESCC) is particularly important due to poor prognosis and the lack of effective therapies. Comprehensive study on early ESCC at single cell level is rare due to the requirements of fresh and high-quality specimen obtained from ESD. This study aims to systematically describe the cellular atlas of human intramucosal ESCC.
Methods: Five paired samples of intramucosal ESCC, para-ESCC esophageal tissue from endoscopically resected specimen, and peripheral blood mononuclear cells were adopted for scRNA-seq analysis. A computational pipeline scMetabolism was applied to quantify the metabolic diversity of single cells.
Results: A total of 164,715 cells were profiled. The abundance of CD8⁺ TEXs , Tregs and PD1⁺CD4⁺T cells suggested an exhausted and suppressive immune microenvironment. Several genes in immune cells such as CXCL13, CXCR5 and PADI4 were identified as new biomarkers for poor prognosis. Epithelial cells exhibited high intra-tumor heterogeneity and two evolutional trajectories during ESCC tumorigenesis initiated from proliferative cells, and then through an intermediate state, to two different terminal states of normal differentiated epithelial cells or malignant cells, respectively. Intercellular interaction analysis based on ligand-receptor pairs revealed malignant cells interacting with CAFs via the MDK-NCL pathway, which was varified by IHC and coculture, is an important hallmark in the change of tumor microenvironment, and serves as a sign of CAF activation to stimulate downstream pathways for facilitating tumor invasion, and therefore suggesting a potential early biomarker of ESCC progression.The proliferation of CAFs was significantly promoted by the stimulation of tumor supernatant and the alteration could be rescued by the inhibitor of MDK (MDK inhibitor, iMDK)
Conclusion: This study demonstrates the changes of cell subsets and transcriptional levels in human intramucosal ESCC, providing unique insights into the development of novel biomarkers and potential intervention strategies.

Background and Aim: Computer Aided Detection (CADe) systems have the potential to improve endoscopic detection of early neoplasia in Barrett’s Esophagus (BE) patients. We aimed to test a recently developed CADe system during live endoscopic procedures.
Method: The CADe system was developed using a large and heterogeneous BE training dataset originating from 15 international endoscopy centers, including 6.237 neoplastic (1.304 patients) and 7.595 non-dysplastic images (1.103 patients). Subsequently, it underwent rigorous external validation by means of multiple ex-vivo benchmarking studies. The CADe system displayed robust performance and significantly increased the neoplasia detection rate of endoscopists. In this pilot study, the CADe system was evaluated during endoscopic procedures of BE patients with a neoplastic lesion or with non-dysplastic Barrett’s esophagus (NDBE) in two tertiary hospitals. The protocol comprised a sequence of white light endoscopy videos obtained by a BE expert endoscopist with real-time evaluation and feedback by the CADe system. First, the Barrett’s segment was completely visualized with a standardized pullback video, starting at the gastric folds up to the maximum extent of the Barrett’s segment. Thereafter, every 2 centimeters of the Barrett’s segment, a 10 second overview video was recorded, starting in retroflexed position. Ground truth (the presence or absence of visible abnormalities requiring targeted biopsy) was established by the endoscopist before starting the protocol, followed by post-hoc histopathological confirmation (by targeted biopsies/endoscopic resection or acquisition of random biopsies). Outcome measure was the stand-alone performance of the CADe system in terms of sensitivity and specificity per patient.
Results: A total of 15 neoplastic and 15 NDBE patients were enrolled in the study. The CADe system correctly detected all neoplastic lesions on a per patient basis, resulting in a sensitivity of 100%. 14 out of 15 visible lesions were correctly diagnosed in the pullback videos. The missed neoplastic lesion was subsequently detected in the level videos. The CADe system incorrectly predicted neoplasia in 8 NDBE patients. Extrapolated to clinical practice, this would result in one additional targeted biopsy per patient whereas the mean Barrett length in this study would dictate 16 random biopsies. Histopathological examination confirmed neoplasia in 13 neoplastic cases (11x adenocarcinoma, 2x high-grade dysplasia), 2 cases did not contain dysplasia. In 4 cases in the non-dysplastic group, low-grade dysplasia was found in the random biopsy protocol.
Conclusion: This study is one of the first to evaluate a CADe system for real-time BE neoplasia detection in the endoscopy suite. The system correctly diagnosed all neoplastic lesions against the background of an acceptable number of false positive detections.