【作者】 孟潘庆； 贾玉生； 郑树； 余捷凯；

【Author】 Meng Pan-Qing Jia Yu-Sheng Zheng Shu Yu Jie-Kai Surgical oncology department, Taian Central Hospital of Shandong, Taian 271000, Shandong Province, China Cancer Institute, the Second Affiliated Hospital of Zhejiang University, College of Medicine, Hangzhou 31009, Zhejiang Province, China

【机构】 泰安市中心医院肿瘤外科； 浙江大学肿瘤研究所；

【摘要】 目的:构建组织原位检测指标预测诊断大肠肿瘤的人工智能神经网络模型,探讨组织芯片技术与人工智能神经网络(ANN)结合应用的可行性。方法:应用组织芯片技术检测ST13等8种组织原位检测指标在大肠肿瘤演进过程各阶段的表达,同时采用人工智能神经网络构建相应的诊断模型。结果:采用Matlab 6．5软件中提供的Kruskal—wallis H秩和检验函数,对这8种指标在正常大肠组织、大肠腺瘤和大肠癌中阳性表达的差异进行统计学检验,其中ST13,Bcl-2, Survivin和HSF1mRNA的P<0．01,有显著性统计学差异, 将8种指标随机组合,分别构建相应的人工智能神经网络诊断模型,评价其各自的诊断效率,发现ST13、Bcl-2、Survivin 与HSF1mRNA组合的ANN—BP模型预测准确率最高,其对正常大肠组织、大肠腺瘤和大肠癌训练集的预测准确率分别高达92．895％,94．163％,92．013％,对该ANN—BP网络诊断模型的盲法测试结果也分别高达85．714％,79．412％,72％。结论:组织芯片技术与人工智能神经网络相结合,可以大大提高组织原位检测指标对大肠肿瘤的预测诊断效率,本研究所构建的4指标组合人工智能神经网络模型,对大肠肿瘤的预测诊断具有较高的准确率,可扩大样本进行更深入的应用性研究。更多还原

【Abstract】 Objective: To establish the colon tumour diagnostic models of 8 tumour related markers in tissue in situ by artificial neural network (ANN) and evaluate the feasibility of combined application of tissue microarray (TMA) technique and artificial neural network (ANN). Methods: Detect 7 kinds of tumour related protein (ST13 and so on) and HSFlmRNA by means of TMA technique, and establish the diagnostic models by ANN-BP. Results: By means of Kruskal-wallis H test available in Matlab 6.5, the expression of every 8 tumour related marker (proteins/mRNA) is evaluated in healthy colon, colon adenoma and colon carcinoma respectively, and the result shows that the expression in these 3 tissues of ST13, Bcl-2, Survivin and HSFlmRNA were significant difference (P < 0.01). Then random assortment of the 8 tumour relate markers are used to establish different diagnostic models, whose diagnostic sensibilities are evaluated by training and blinding test sets respectively. The diagnostic model established by the group of ST13, Bcl-2, Survivin and HSFlmRNA is found to be the best one among all the random groups. Its training set predicted veracity is 92.895% for healthy colon tissue, 94.163% for colon adenoma, and 92.013% for colon carcinoma, meanwhile its blinding test set predicted veracity is 85.714%, 79.412% and 72% respectively. Conclusion: Combined application of TMA technique and ANN could enhance the diagnostic efficiency of the tumour related markers in colon tissue in situ dramatically, The ANN-BP diagnostic model established by the group of ST13, PTEN, Bcl-2 and HSF1mRNA could identify the colon adenoma and carcinoma from the healthy colon tissues sensitively.更多还原