论文部分内容阅读
目的借助计算机技术协助进行寄生虫虫卵的识别,建立适应自动化仪器中临床应用的虫卵分类器算法和相应流程。方法选择华支睾吸虫(Clonorchis sinensis)、猪带绦虫(Taenia solium)、蛲虫(Enterobius vermicularis)、蛔虫(Ascaris lumbricoides)、鞭虫(Trichuris trichiura)、曼氏迭宫绦虫(Spirometra mansoni)、阔节裂头绦虫(Diphyllobothrium latum)、十二指肠钩虫(Ancylostoma duodenale)、日本血吸虫(Schistosoma japonicum)、卫氏并殖吸虫(Paragonimus westermani)和布氏姜片吸虫(Fasciolopsis buski)等11种寄生虫的虫卵,分为训练组和测试组进行显微摄影,并使用基于VC++技术进行特征值提取。构建特征值数据库,使用多种分类算法对训练组数据库进行测试,选取分类效率最高的方法构建分类器,建立基于多特征融合的识别方法。结果获取了11种寄生虫虫卵图像,去除无法识别或含无效值的图片后,训练数据组虫卵图片为19 844张,测试组为3 721张。对虫卵的14种特征值进行采集,发现11种虫卵的大小、颜色均有显著差异。如11种虫卵中体积最小的华支睾吸虫虫卵的长度、宽度、面积、亮度的均值分别为292.24μm、192.64μm、43 416.61μm~2、53.84,而体积最大的布氏姜片吸虫虫卵则分别为945.31μm、610.88μm、536 002.60μm~2、100.54。在多特征融合检索时用动态生成权值的方法建立分类器,对训练样本集的区分率为88.89%(17 641/19 844),该分类器对测试样本集的识别率为91.83%(3 004/3 271),平均建模时间为0.01 s。结论建立了基于特征值融合方法的寄生虫虫卵分类器算法及相应流程,为其可行性的进一步研究打下了基础。
Objective To assist the identification of parasite eggs by computer technology and establish the algorithm and corresponding flow of the worm eggs to adapt to the clinical application of automated instruments. Methods Clonorchis sinensis, Taenia solium, Enterobius vermicularis, Ascaris lumbricoides, Trichuris trichiura, Spirometra mansoni, Diphyllobothrium latum, Ancylostoma duodenale, Schistosoma japonicum, Paragonimus westermani, and Fasciolopsis buski Eggs were divided into training group and test group for microscopic photography, and feature extraction based on VC ++ was used. The eigenvalue database is constructed, and a variety of classification algorithms are used to test the training set database. The classifier is selected based on the highest classification efficiency and the identification method based on multi-feature fusion is established. Results Eleven parasite eggs images were obtained. After removing unrecognized or invalid images, there were 19,844 eggs in the training data set and 3,721 in the test group. The 14 characteristic values of the eggs were collected, and the size and color of the 11 kinds of eggs were found to be significantly different. For example, the mean value of the length, width, area and brightness of the eggs with the smallest size among the 11 kinds of eggs is 292.24μm, 192.64μm and 43 416.61μm ~ 2,53.84, respectively. The largest volume of M. bovis The eggs were 945.31μm, 610.88μm, 536.002.60μm ~ 2,100.54, respectively. In the multi-feature fusion retrieval, a classifier was constructed by dynamically generating weights, the discriminant rate of the training sample set was 88.89% (17 641/19 844), and the recognition rate of the classifier to the test sample set was 91.83% (3 004/3 271) with an average modeling time of 0.01 s. Conclusion The parasitoid egg classifier algorithm and its corresponding flow based on eigenvalue fusion method are established, which lays the foundation for further research on its feasibility.