Award: Presidential Poster Award
Jasna I. Beard, MD1, Arslan Kahloon, MD1, Peter Siersema, MD, PhD2, Seth Gross, MD, FACG3, Harold Jacob, MD4
1University of Tennessee, Chattanooga, TN; 2Radboud University, Nijmegen, Gelderland, Netherlands; 3Tisch Hospital, New York University Langone Health, New York, NY; 4Gastroenterology Institute, Hadassah Medical Center, Kiryat Hadassah, Yerushalayim, Israel
Introduction: Missed lesions are considered a major cause of interval cancer of the colon. The new Automatic Polyp Detection System (APDS) from Magentiq Eye was developed to enhance the ability of endoscopists to detect polyps. It is designed for offline use with future real-time applications. The system uses recorded colonoscopy videos to assess endoscopist performance against the APDS. The APDS reviews the videos to generate a computerized automatic signature of polyps; a separate screen then shows the corresponding "heat map" contour of areas suspicious for presence of polyps (see Figure 1).
Methods: In this proof-of-concept study, the performance of the trained system was compared with a known testing database of 35 videos (each sequence approximately 30 seconds and 1000 frames). Overall, the testing database included 35,290 frames, with polyps found in the right colon (cecum and ascending colon, 10), transverse colon (5), and left colon (sigmoid and rectum, 20). Of the 35, 33 polyps were sessile and 2 were pedunculated; sizes ranged from less than 3mm (10), 3 to 6mm (16), 6 to 10mm (5), and over 10mm (4).
Our goal was to define the optimal system working point in terms of sensitivity and specificity to detect polyps present in at least 3 consecutive frames. Physicians reviewed videos for polyps that were missed by the APDS and compare it with those missed by the endoscopists.
Results: The best working point of the system showed 88% sensitivity with 98.4% specificity (0.016 false positive rate). This resulted in a system polyp detection rate (SPDR) of 85.8% with 5 missed polyps. Missed lesions were found in the ascending (2), transverse (2), and descending (1) colon. All were sessile with sizes of less than 3mm (1), 3 to 6 mm (3), 6 to 10mm (1). The physicians missed 4 polyps found in the transverse colon (3) and descending colon (1). All polyps missed by physicians were sessile with sizes of less than 3mm (1), 3 to 6 mm (2), and 6 to 10mm (1). Only one polyp was missed both by the APDS and the physician.
Discussion: Given that only one polyp was missed both by the APDS and the endoscopists, the APDS demonstrated the potential to behave as a complementary tool during colonoscopy in order to alert the endoscopist to lesions that may otherwise be missed, aiming to decrease the incidence of interval colon cancers.
Citation: Jasna I. Beard, MD; Arslan Kahloon, MD; Peter Siersema, MD, PhD; Seth Gross, MD, FACG; Harold Jacob, MD. P0219 - AUTOMATIC POLYP DETECTION SYSTEM: MAN VERSUS MACHINE TOWARDS A COLLABORATIVE COLONOSCOPY. Program No. P0219. ACG 2019 Annual Scientific Meeting Abstracts. San Antonio, Texas: American College of Gastroenterology.