Automatic detection of various abnormalities in capsule endoscopy videos by a deep learning-based system: a multicenter study - 08/08/20

Doi : 10.1016/j.gie.2020.04.080

Tomonori Aoki, MD, PhD ¹, Atsuo Yamada, MD, PhD ^1,^⁎, Yusuke Kato, BSc ², Hiroaki Saito, MD ³, Akiyoshi Tsuboi, MD ⁴, Ayako Nakada, MD, PhD ¹, Ryota Niikura, MD, PhD ¹, Mitsuhiro Fujishiro, MD, PhD ⁵, Shiro Oka, MD, PhD ⁴, Soichiro Ishihara, MD, PhD ^6,⁷, Tomoki Matsuda, MD, PhD ³, Masato Nakahori, MD, PhD ³, Shinji Tanaka, MD, PhD ⁴, Kazuhiko Koike, MD, PhD ¹, Tomohiro Tada, MD, PhD ^2,^6,⁷
¹ Department of Gastroenterology, Graduate School of Medicine, The University of Tokyo, Tokyo, Japan
² AI Medical Service Inc, Tokyo, Japan
³ Sendai Kousei Hospital, Department of Gastroenterology, Miyagi, Japan
⁴ Department of Endoscopy, Hiroshima University Hospital, Hiroshima, Japan
⁵ Department of Gastroenterology and Hepatology, Nagoya University Graduate School of Medicine, Nagoya, Japan
⁶ Department of Surgical Oncology, Graduate School of Medicine, The University of Tokyo, Tokyo, Japan
⁷ Tada Tomohiro Institute of Gastroenterology and Proctology, Saitama, Japan

^∗Reprint requests: Dr Atsuo Yamada, Department of Gastroenterology, Graduate School of Medicine, The University of Tokyo, 7-3-1 Hongo, Bunkyo, Tokyo 113-8655, Japan.Department of GastroenterologyGraduate School of MedicineThe University of Tokyo7-3-1 HongoBunkyoTokyo113-8655Japan

Sous presse. Épreuves corrigées par l'auteur. Disponible en ligne depuis le Saturday 08 August 2020

Abstract

Background and Aims

A deep convolutional neural network (CNN) system could be a high-level screening tool for capsule endoscopy (CE) reading but has not been established for targeting various abnormalities. We aimed to develop a CNN-based system and compare it with the existing QuickView mode in terms of their ability to detect various abnormalities.

Methods

We trained a CNN system using 66,028 CE images (44,684 images of abnormalities and 21,344 normal images). The detection rate of the CNN for various abnormalities was assessed per patient, using an independent test set of 379 consecutive small-bowel CE videos from 3 institutions. Mucosal breaks, angioectasia, protruding lesions, and blood content were present in 94, 29, 81, and 23 patients, respectively. The detection capability of the CNN was compared with that of QuickView mode.

Results

The CNN picked up 1,135,104 images (22.5%) from the 5,050,226 test images, and thus, the sampling rate of QuickView mode was set to 23% in this study. In total, the detection rate of the CNN for abnormalities per patient was significantly higher than that of QuickView mode (99% vs 89%, P < .001). The detection rates of the CNN for mucosal breaks, angioectasia, protruding lesions, and blood content were 100% (94 of 94), 97% (28 of 29), 99% (80 of 81), and 100% (23 of 23), respectively, and those of QuickView mode were 91%, 97%, 80%, and 96%, respectively.

Conclusions

We developed and tested a CNN-based detection system for various abnormalities using multicenter CE videos. This system could serve as an alternative high-level screening tool to QuickView mode.

Le texte complet de cet article est disponible en PDF.

Abbreviations : CE, CNN, SSD, CEST

Plan

Introduction

Methods

Preparation of the training image set

Preparation of the small-bowel full CE videos as a test set

CNN algorithm

Outcome measures and statistics

Results

Characteristics of the small-bowel CE videos in the test set

Comparison of the detection capabilities for various abnormalities between the CNN and QuickView mode

CNN capability for classifying abnormalities into 4 categories

Discussion

DISCLOSURE: Dr Kato is an employee of AI Medical Service Inc. Dr Tada is a shareholder in AI Medical Service Inc. AI Medical Service Inc did not provide any funding for this study and was not involved in the study design, study conduct, data handling, statistical analysis, or interpretation of the results. All other authors disclosed no financial relationships.