基于多期CT图像的常见肝脏疾病计算机辅助诊断系统

发布时间：2019-01-12 16:09

【摘要】：随着社会发展,世界卫生组织等权威机构均观测到了肝脏疾病发病率的不断上升,其中肝癌已经成为致死率最高的肝脏疾病之一。除治疗手段方面的原因外,早期肝癌病理指标的不明显,亦会造成患者确诊与治疗的延迟。当前,肝癌的确诊依然很大程度上依赖于穿刺活检技术,其实施难度大,且患者体验、术后恢复等均存在一定的问题。于是非介入式的计算机辅助诊断在肝脏疾病的检测、发现、确诊中有着非常广阔的前景和重要的意义。本文致力于设计一个常见肝脏疾病的计算机自动辅助诊断系统,该系统根据数据处理流程可依次划分为：感兴趣区域(Regin Of Interest, ROI)提取模块、特征提取和特征选择模块、分类器模块。由于常见肝脏疾病在CT图像中的表现相似度较高,这给基于计算机的肝脏辅助诊断系统设计带来了一定的难度,因此本文使用多期腹部CT扫描数据作为系统的输入。系统首先结合了水平集方法和区域生长法提出病灶部分作为ROI；随后提取了基于灰度直方图、基于灰度共生矩阵的肝脏纹理统计特征和基于多期肝脏CT图的时序特征作为特征向量,并经过主成分分析法进行特征选择；最后将降维优化后的特征向量输入分类器模块,选取支持向量机作为分类器算法,将系统设计为三层级联的二分分类器,分别得到正常和非正常、肝囊肿和其它、肝血管瘤和肝癌的诊断准确率,并结合医学诊断的特殊性给出了接受者操作特性曲线值,两者共同作为判别系统性能的参考。经过实验数据验证,系统运行稳定且达到了较高的诊断准确率。其中对于最重要的判别指标——正常和非正常肝脏的分类准确率达到了99.49%,证明了本文方法的可靠和有效性。结合本文工作以及实际临床需求,未来肝脏疾病辅助诊断系统需要在肝脏分割和病灶提取方面进一步改进,以提高系统的诊断准确率。
[Abstract]:With the development of society, the World Health Organization and other authoritative organizations have observed the rising incidence of liver diseases, among which liver cancer has become one of the most fatal liver diseases. In addition to the treatment methods, the early pathological indicators of liver cancer are not obvious, but also lead to the delay of diagnosis and treatment. At present, the diagnosis of liver cancer still depends largely on the puncture biopsy technique, which is difficult to implement, and there are some problems in patient experience, postoperative recovery and so on. Therefore, non-interventional computer-aided diagnosis has broad prospects and important significance in the detection of liver diseases. This paper is devoted to the design of a computer-aided diagnosis system for common liver diseases. According to the data processing process, the system can be divided into three modules: region of interest (Regin Of Interest, ROI) extraction module, feature extraction and feature selection module. Classifier module. Because of the high performance similarity of common liver diseases in CT images, it is difficult to design a computer-based liver aided diagnosis system, so we use multi-phase abdominal CT scan data as the input of the system. Firstly, the level set method and the region growth method are combined to propose the lesion part as ROI;. Secondly, the feature vectors of liver texture statistics based on gray histogram, gray level co-occurrence matrix and temporal feature based on multi-phase liver CT are extracted as feature vectors, and the feature selection is carried out by principal component analysis (PCA). Finally, the optimized feature vector is input into the classifier module, and the support vector machine is selected as the classifier algorithm. The system is designed as a three-layer cascade binary classifier to obtain normal and abnormal, liver cyst and others, respectively. The diagnostic accuracy of hepatic hemangioma and liver cancer was analyzed, and the operating characteristic curve of the recipient was given in combination with the particularity of medical diagnosis, which could be used as a reference to judge the performance of the system. The experimental data show that the system runs stably and achieves high diagnostic accuracy. The classification accuracy of normal and abnormal liver is 99.49, which proves the reliability and validity of this method. In order to improve the accuracy of diagnosis, the future liver disease diagnosis system needs to be further improved in liver segmentation and focus extraction.
【学位授予单位】：厦门大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R575;TP391.41

【参考文献】