结合白光和NBI图像实时智能识别技术的应用案例分析  

Case Analysis of Real-time Intelligent Recognition Technology Combining White Light and NBI Images

作　　者：Bill Dong 董兰芳[1] 马涛[1] 晋晶 张倩倩[2] 孔德润[2] 吴艾久 Bill Dong;DONG Lanfang;MA Tao;JIN Jing;ZHANG Qianqian;KONG Derun;WU Aijiu(University of Science and Technology of China,School of Computer Science and Technology,Anhui 230027,China;Department of Gastroenterology,First Affiliated Hospital of Anhui Medical University,Anhui Key Laboratory of Digestive Diseases,Anhui 230022,China;Hefei Zhongna Medical Instrument Co.,Ltd.,Anhui 230088,China)

机构地区：[1]中国科学技术大学计算机科学与技术学院,安徽230027 [2]安徽医科大学第一附属医院消化内科,安徽省消化病重点实验室,安徽230022 [3]合肥中纳医学仪器有限公司,安徽230088

出　　处：《电子技术（上海）》2022年第11期12-15,共4页Electronic Technology

基　　金：安徽省科技厅2022年重点研发计划项目(2022e07020048)。

摘　　要：结合白光和窄带成像(NBI)两种光源的胃镜图像,阐述基于深度学习识别算法的早期胃癌智能识别系统的设计,利用采集、制作的10942张胃部早癌医疗图像训练深度学习模型。针对收集到的大量早期胃癌图像的数据特点,对算法模型进行了置信度阈值调优,提升模型在多指标下的综合效果。结合医师在实际使用时的实时需求,提出一种视频时序投票方法对视频预处理及后处理策略进行优化。Combined with the gastroscopic images of white light and narrowband imaging(NBI),this paper describes the design of an intelligent recognition system for early gastric cancer based on depth learning recognition algorithm,and uses 10942 medical images of early gastric cancer collected and produced to train the depth learning model.According to the data characteristics of a large number of early gastric cancer images collected,the confidence threshold of the algorithm model is optimized to improve the comprehensive effect of the model under multiple indicators.According to the real-time requirements of doctors in actual use,a video timing voting method is proposed to optimize the video pre-processing and post-processing strategies.

关 键 词：图像分割 目标检测 深度学习 视频处理 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]