基于人工神经网络的卵巢早衰预测模型研究

发布时间：2018-06-01 19:20

本文选题：原发性卵巢功能不全 + 卵巢功能早衰　；参考：《中国全科医学》2017年27期

【摘要】：目的建立基于人工神经网络(ANN)的卵巢早衰(POF)预测模型——多层向前神经网络模型,以期提高POF临床诊断总符合率。方法 2011年1—3月选取武汉市白玉山街所管辖的6个社区内符合纳入标准的妇女341例为研究对象。2011年5月—2016年6月,每隔4个月对研究对象进行1次来院随访,随访至其40岁。随访过程中2例研究对象行子宫切除术,2例服用性激素治疗,失访21例,均予以剔除,最终共纳入316例研究对象。采用无偏随机化分配法将316例研究对象分为训练样本(177例)、检验样本(44例)和坚持样本(95例)。设置输入参数为A型行为、腮腺炎病史、妇科手术史、使用促排卵药物史、婚育史、卵泡刺激素(FSH)、FSH/黄体生成素(LH)、抗苗勒管激素(AMH)、抑制素B(INHB)、窦状卵泡数(AFC)、收缩期峰流速(PSV)、阻力指数(RI);输出参数为"是否发生POF"。通过训练样本进行模型构建,检验样本对模型进行校正,坚持样本对模型进行稳定性检测。结果ANN经过剔除"冗余"后,自动构建出输入单元(12个)、单隐层(6个节点)和激活函数(hyperbolic tangent)、输出单元(2个)和激活函数(softmax)的模型。训练样本的交叉熵误差值为53.236,在预测误差未减少时终止测试,训练时间为0.42 s。影响权重在前5位的输入参数分别为AMH(26.3%)、INHB(24.1%)、AFC(21.7%)、A型行为(7.2%)、妇科手术史(6.5%)。多层向前神经网络模型预测训练样本、检验样本、坚持样本发生POF的灵敏度分别为97.8%、91.7%和92.0%,特异度分别为92.4%、84.4%和80.0%,总符合率分别为93.8%、86.4%和83.2%。在训练样本和检验样本的基础上,得到多层向前神经网络模型预测POF的受试者工作特征曲线下面积(AUC)为0.972。结论基于ANN构建的POF预测模型——多层向前神经网络模型具有较高临床诊断总符合率,不仅为临床高效诊断及优化检查提供理论基础和方法支持,而且为实现早防早治提供机会,值得临床推广。
[Abstract]:Objective to establish a multilayer forward neural network model for predicting premature ovarian failure (POF) based on artificial neural network (Ann) in order to improve the total diagnostic coincidence rate of POF. Methods from January to January 2011, 341 women who met the inclusion criteria were selected from 6 communities in Baayushan Street, Wuhan City. The subjects were followed up every 4 months from May 2011 to June 2016. They were followed up to 40 years old. During the follow-up, 2 cases were treated with sex hormone after hysterectomy, 21 cases were excluded, and 316 cases were included in the study. 316 cases were divided into training samples (177 cases), test samples (44 cases) and persistent samples (95 cases) by unbiased randomization method. The input parameters were: type A behavior, history of mumps, history of gynecological surgery, history of using ovulation promoting drugs, history of marriage and childbearing. Follicle stimulating hormone (FSH) / luteinizing hormone (LH), anti-mullerian hormone (AMH), inhibin (BINHBB), antral follicle number (POF), peak systolic velocity (PSV), resistance index (RI), and the output parameters were "whether POF occurred or not". The model is constructed by training samples, calibrated by test samples, and the stability of the model is detected by persisting samples. Results after eliminating "redundancy" in ANN, the models of input unit (12), single hidden layer (6 nodes) and activation function (hyperbolic tangent, output unit (2) and activation function) were automatically constructed. The cross-entropy error of the training sample is 53.236, and the test is terminated when the prediction error is not reduced, and the training time is 0.42 s. The input parameters in the first five positions of influence weight were AMH26.3 and INHB24.1, respectively. The AFCJ 21.7A behavior was 7.2cm, and the gynecologic operation history was 6.5cm. The sensitivity of POF was 91.7% and 92.0%, the specificity was 92.44.4% and 80.0%, respectively. The total coincidence rates were 93.864% and 83.2%, respectively. On the basis of training samples and test samples, a multilayer forward neural network model was obtained to predict the area under the operating characteristic curve of POF was 0.972. Conclusion the POF predictive model based on ANN, a multilayer forward neural network model, has a high total coincidence rate of clinical diagnosis, which not only provides theoretical basis and method support for clinical high efficiency diagnosis and optimization examination. And for early prevention and treatment to provide opportunities, worthy of clinical promotion.
【作者单位】：武汉钢铁(集团)公司第二职工医院妇产科;
【基金】：武汉市临床医学科研项目(WX15D15) 第四批武汉中青年医学骨干人才资助项目
【分类号】：R711.75;TP183

【相似文献】