Smart electronic gastroscope system using a cloud–edge collaborative framework

Abstract

Abstract Computer-aided gastroscopic image analysis is capable of reducing the work intensity of gastroscopists, and it is of great significance for improving the sensitivity and specificity of upper gastrointestinal disease screening. However, most computer-aided gastroscope systems provide intelligent image analysis services that rely on public cloud platforms and suffer from high communication and computing costs. Moreover, these systems are normally unavailable for offline clinical practice. In this study, we propose a smart electronic gastroscope system based on a cloud–edge collaborative framework. In this system, edge computing platforms and cloud platforms work collaboratively to achieve real-time lesion localization and fine-grained disease classification of gastroscopic videos. In addition, we propose a novel approach called cloud–edge collaborative dynamic lesion detection for upper gastrointestinal disease inference. First, to assist real-time lesion detection in the offline mode or discover a suspicious frame in the online mode, we develop a Tinier–YOLO algorithm based on the k-DSC module in edge computing platforms. Second, to further improve the modeling performance, we integrate lesion ROI segmentation strategy into the YOLOv3 algorithm in the cloud platform. By testing clinical data, we prove that our approach exhibits superior performance in mAP and IOU of lesion detection and response time of service.