川崎病早期冠状动脉损伤的认知图构建

发布时间：2018-10-16 10:14

【摘要】：目的构建川崎病(Kawasaki Disease,KD)急性期并发冠状动脉损伤(Coronary Artery Lesion,CAL)的预测模型和认知图(Cognitive Map,CM),对临床上该领域研究方法进行探讨。方法通过查阅PubMed、Web of Science、EBSCCO、中国知网、万方、维普等数据库,提取既往研究中KD并发CAL危险因素的生物学指标,同时在数据库中提取KD患儿入院后第一次生化检查指标,删除缺失率超过5%的指标后对数据进行处理,采用B-P神经网络(Back Propagation Artificial Neural Network,BPANK)建立的预测模型来确定各指标广义权重(Generalize Weight,GW),利用贝叶斯网络(Bayesian Network,BN)挖掘各生物学指标与CAL的关系,结合GW确定在BPANN预测模型中与KD并发CAL有关的主要危险因素和各个因素间的关联强度,最后构建KD并发CAL的CM。结果BPANN预测模型结果显示训练集的分类一致率为94.64%,灵敏度为63.76%,特异度为99.85%;测试集分类一致率为94.63%,灵敏度为53.45%,特异度为99.59%。GW散点图中红细胞、白细胞在整个模型中的权重接近于0;BN模型结果显示红细胞、白细胞计数与CAL独立,其余各个节点间存在7条关系,分别为年龄与红细胞压积,年龄与中性粒细胞百分比,血红蛋白与CAL,红细胞压积与血红蛋白,CAL与中性粒细胞百分比,血小板计数与中性粒细胞百分比,中性粒细胞百分比与C-反应蛋白。结论根据患儿入院后第一次生化检查数据,采用BPANN方法建立KD患儿并发CAL的预测模型,提高了预测精度,对临床早期识别并发CAL的高危患儿具有参考意义。结合BN结果,构建年龄、血小板计数、红细胞压积、血红蛋白、C-反应蛋白、中性粒细胞百分比与CAL之间的CM,直观的显示了各指标间的关联关系及关联强度,对临床该领域研究方法进行了探讨。但由于BPANN模型灵敏度较低,还需纳入更多非生化指标因素,并与炎症、免疫相关疾病对照,结合数据挖掘的方法进一步研究。
[Abstract]:Objective to construct a predictive model and a cognitive map (Cognitive Map,CM) of acute coronary artery injury (Coronary Artery Lesion,CAL) in Kawasaki disease (Kawasaki Disease,KD). Methods the biological indexes of the risk factors of KD complicated with CAL were extracted from the databases of PubMed,Web of Science,EBSCCO, Wanfang and Weip, and the first biochemical indexes of the children with KD were extracted from the database. The data are processed after the deletion rate is more than 5%. The prediction model established by B-P neural network (Back Propagation Artificial Neural Network,BPANK) is used to determine the generalized weight (Generalize Weight,GW) of each index, and Bayesian network (Bayesian Network,BN) is used to mine the relationship between each biological index and CAL. Combined with GW to determine the main risk factors related to KD concurrent CAL and the correlation strength of each factor in the BPANN prediction model. Finally, the CM. of KD concurrent CAL was constructed. Results the results of BPANN prediction model showed that the classification consistency rate of training set was 94.64, the sensitivity was 63.76, the specificity was 99.85, the classification consistency rate of test set was 94.63, the sensitivity was 53.45 and the specificity was erythrocyte in 99.59%.GW scatter plot. The weight of leukocytes in the whole model was close to that in the model of 0 BNM. The results showed that the white blood cell count was independent of CAL, and there were 7 relationships between the other nodes: age and hematocrit, age and percentage of neutrophils. Hemoglobin and CAL, hematocrit and hemoglobin, CAL and neutrophil percentage, platelet count and neutrophil percentage, neutrophil percentage and C-reactive protein. Conclusion based on the data of the first biochemical examination after admission, the prediction model of KD complicated with CAL was established by using BPANN method, which improved the prediction accuracy, and has reference significance for early clinical identification of high-risk children with CAL. Combined with the BN results, the CM, of age, platelet count, hematocrit, hemoglobin, C-reactive protein, neutrophil percentage and CAL showed the correlation and intensity of each index. The methods of clinical research in this field were discussed. However, because of the low sensitivity of BPANN model, it is necessary to include more non-biochemical index factors, and compare with inflammation and immune-related diseases, combined with the method of data mining to further study.
【学位授予单位】：重庆医科大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R725.4

【相似文献】