Endoscopy 2020; 52(S 01): S235
DOI: 10.1055/s-0040-1704736
ESGE Days 2020 ePoster Podium presentations
Saturday, April 25, 2020 15:00 – 15:30 Artificial Intelligence for colonoscopy and ePoster Podium 4 small: Bowel endoscopy
© Georg Thieme Verlag KG Stuttgart · New York

IMPROVING OPTICAL DIAGNOSIS OF COLORECTAL POLYPS USING COMPUTER-AIDED DIAGNOSIS (CADX)

QEW van der Zander
1   Maastricht University Medical Center, Gastroenterology and Hepatology, Maastricht, Netherlands
2   Maastricht University, GROW, School for Oncology and Developmental Biology, Maastricht, Netherlands
,
RM Schreuder
3   Catharina Hospital Eindhoven, Gastroenterology and Hepatology, Eindhoven, Netherlands
,
R Fonolla
4   Eindhoven University of Technology, Department of Electrical Engineering, Eindhoven, Netherlands
,
T Scheeve
4   Eindhoven University of Technology, Department of Electrical Engineering, Eindhoven, Netherlands
,
F van der Sommen
4   Eindhoven University of Technology, Department of Electrical Engineering, Eindhoven, Netherlands
,
P Aepli
5   Luzerner Kantonsspital, Luzern, Switzerland
,
B Hayee
6   King’s College Hospital London, London, United Kingdom
,
A Pischel
7   University Hospital Gothenburg, Gothenburg, Sweden
,
M Stefanovic
8   Diagnostični Center Bled, Ljubljana, Slovenia
,
S Subramaniam
9   Queen Alexandra Hospital, Portsmouth, United Kingdom
,
P Bhandari
9   Queen Alexandra Hospital, Portsmouth, United Kingdom
,
PHN de With
4   Eindhoven University of Technology, Department of Electrical Engineering, Eindhoven, Netherlands
,
AAM Masclee
10   Maastricht University, NUTRIM, School of Nutrition & Translational Research in Metabolism, Maastricht, Netherlands
,
EJ Schoon
3   Catharina Hospital Eindhoven, Gastroenterology and Hepatology, Eindhoven, Netherlands
› Author Affiliations
Further Information

Publication History

Publication Date:
23 April 2020 (online)

Also available at
 

Aims Optical diagnosis is the endoscopic prediction of histopathology of colorectal polyps detected at colonoscopy. Optical diagnosis remains challenging with accuracies of 71-90% in the Dutch bowel cancer screening program, exposing patients to risks of incorrect diagnosis. We propose a new methodology to improve the diagnostic accuracy of optically diagnosing colorectal polyps by Computer-Aided Diagnosis (CADx).

Methods We prospectively compared the optical diagnosis of colorectal polyps made by CADx with experts from the international BLI-expert group and Dutch novices. The optical diagnosis was first made based on intuition, with a time limit of 30 seconds. After a washout period of four weeks, the same set of polyps was optically diagnosed based on a clinical classification model; BASIC (BLI Adenoma Serrated International Classification). CADx classified colorectal polyps by exploiting machine-learning algorithms.

Results Colorectal polyps of the following histopathology were included: hyperplastic polyp (n = 15), adenoma (n = 39), sessile serrated adenoma (n = 4) and adenocarcinoma (n = 2). Five experts, with a mean colonoscopy experience of 16.0 years and nine novices (mean 2.3 year) participated. The CADx algorithm was based on benign versus premalignant polyps. A subgroup analyses for experts and novices was performed to allow for an adequate comparison with CADx. The diagnostic accuracy of experts (81.0%) was significantly higher in comparison to novices (64.2%) based on BASIC and based on intuition (78.6% vs 63.8%). CADx had a significantly higher overall diagnostic accuracy of 93.8% (p < 0.001). Sensitivity (91.7% vs. 62.7% and 52.6%) and specificity (100.0% vs. 95.3% and 93.3%) were also significantly higher for CADx compared to both experts and novices.

Conclusions The clinical classification model BASIC increased the diagnostic accuracy of experts and novices compared to intuitive optical diagnosis. CADx diagnosed colorectal polyps significantly better in comparison to experts and novices. These findings stress the need for further validation of CADx for future implementation into daily endoscopy practice.