Endoscopy 2019; 51(04): S7
DOI: 10.1055/s-0039-1681189
ESGE Days 2019 oral presentations
Friday, April 5, 2019 08:30 – 10:30: Artificial intelligence Club A
Georg Thieme Verlag KG Stuttgart · New York

AUTOMATIC POLYP DETECTION IN COLONOSCOPY – GENERAL COMPARISON OF SYSTEM AND VIDEO ANALYSIS STATISTICS

H Jacob
1   Gastroenterology, Hadassah University Medical Center, Jerusaelm, Israel
,
Y Kopelman
2   Gastroenterology, Hillel Yaffe Medical Center, Hadera, Israel
,
G Oren
2   Gastroenterology, Hillel Yaffe Medical Center, Hadera, Israel
,
P Siersema
3   Gastroenterology, Radboundumc University, Rotterdam, Netherlands
,
A Cohen
4   Magnetiq Eye Ltd, Haifa, Israel
,
R Eliakim
5   Magnetic Eye Ltd., Haifa, Israel
,
M Zaltshendler
5   Magnetic Eye Ltd., Haifa, Israel
,
D Zur
4   Magnetiq Eye Ltd, Haifa, Israel
› Author Affiliations
Further Information

Publication History

Publication Date:
18 March 2019 (online)

 

Aims:

The Automatic Polyp Detection System (APDS) was developed to enhance the ability of endoscopists to detect polyps during screening colonoscopy. The aim of this study was to compare performance of the trained system with a known database of 35 video sequences.

Methods:

The testing database was collected from 12 physicians in 6 endoscopy centers during years 2014 – 2018. Each of the 35 video sequences represented consecutive frames a half-minute in length. Our goal was to define the best system performance (sensitivity and specificity working point, and the percentage of polyps detected by the system) in at least 3 consecutive frames under a given system specificity. In addition, the number of polyps that were estimated to be missed during the procedure according to a manual analysis of the video sequences was calculated.

Results:

The best working points of the system over the testing database was 88% sensitivity with 98.7% specificity. For working points with a specificity of 97.5% and below, all polyps in the testing database were detected by the system in at least 3 consecutive frames. On the other hand, according to the manual analysis of the recorded videos, 4 polyps out of the 35 polyps in the testing database (11.4%) were missed by the physicians during the procedures.

Conclusions:

The endoscopist could be alerted to the presence of a polyp with a specificity of 97.5%. The estimated miss rate of the physicians of polyps correlates with research work that showed a reduction in the miss rate when using behind folds imaging techniques. The use of the APDS can contribute to the reduction of the miss rate of polyps in daily clinical practice.