基于胶囊内窥镜图像的出血病灶检测算法研究

发布时间：2018-08-20 17:51

【摘要】：如今,肠胃疾病已经成为人类健康的一大威胁。机械式内窥镜作为传统肠胃疾病的检测手段,不仅操作不方便,也给患者带来身体上的痛苦。随着半导体技术、无线通信技术、集成电路技术等技术的发展,胶囊内窥镜问世了,并很快在肠胃疾病的检测方面取得得天独厚的优势。然而胶囊内窥镜一次检测产生的消化道图像数量达几万张之巨,由医护人员来逐一查看,给医务人员带来沉重负担的同时,还会增加误诊率。针对这些问题,本文在介绍了胶囊内窥镜和基于胶囊内窥镜图像的出血病灶检测技术的国内外研究现状的基础上,对基于胶囊内窥镜图像的出血病灶检测技术进行了深入的研究,主要的工作包括:图像的预处理、基于颜色特征的感兴趣区域提取、基于颜色相似性和连通域面积的分类识别,实现了基于胶囊内窥镜图像的出血病灶检测算法。传统的基于胶囊内窥镜图像的出血检测算法主要有两种,一种是将图像分割为固定大小的区域,这种机械的划分会破坏图像本身含有的边界信息,导致准确度不高;另一种是在整幅图像上进行模板运算,这种算法可以最大限度反映原始图像的信息,然而由于数据量大,导致检测算法速度过慢。本文兼顾算法的速度和图像的原始边界信息,首先在RGB颜色空间利用颜色边界盒进行感兴趣区域提取,减少图像的冗余信息,然后利用感兴趣区域的颜色相似性系数和连通域面积组成的分类器,对感兴趣区域进行分类,在保证检测准确率的前提下提高了检测速度。最后,本文通过实验验证算法,结果表明:算法的灵敏度达到了91%,特异性达到88%,基本实现了胶囊内窥图像出血病灶的自动检测,可应用于实践治疗中。
[Abstract]:Nowadays, gastrointestinal diseases have become a major threat to human health. As a traditional means of detecting gastrointestinal diseases, mechanical endoscopy is not only inconvenient to operate, but also brings physical pain to patients. However, the number of digestive tract images produced by capsule endoscopy is tens of thousands, which brings heavy burden to medical staff and increases the misdiagnosis rate. To solve these problems, capsule endoscopy and capsule-based endoscopy are introduced in this paper. Based on the research status of bleeding lesion detection technology in endoscopy image at home and abroad, the bleeding lesion detection technology based on capsule endoscopy image is deeply studied. The main work includes: image preprocessing, region of interest extraction based on color features, classification and recognition based on color similarity and connected area area. There are two kinds of traditional algorithms for detecting hemorrhage based on capsule endoscopy image: one is to divide the image into fixed size regions, which will destroy the boundary information contained in the image itself and lead to low accuracy; the other is the whole image. This algorithm can reflect the information of the original image to the greatest extent, but the speed of the detection algorithm is too slow because of the large amount of data. Considering the speed of the algorithm and the original edge information of the image, this paper firstly extracts the region of interest in the RGB color space using color boundary box to reduce the redundant information of the image. Finally, the algorithm is validated by experiments. The results show that the sensitivity and specificity of the algorithm are 91% and 88% respectively, which basically realize the capsule. Automatic detection of bleeding lesions can be applied to practical treatment.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：R57;TP391.41

【参考文献】