基于神经网络分析法的肺磨玻璃密度结节侵袭性CT分析预测模型研究

发布时间：2018-01-20 04:08

本文关键词： 磨玻璃密度结节神经网络多层感知器　出处：《临床放射学杂志》2017年08期 　论文类型：期刊论文

【摘要】：目的利用神经网络分析法构建肺磨玻璃密度结节(GGN)侵袭性的CT预测模型,探讨其预测的准确性。方法回顾性分析203例经手术病理证实为肺腺癌的肺GGN的CT影像特征。采集患者基本信息,统计肺结节密度(纯磨玻璃结节或混合磨玻璃结节)、是否含有内核、大小、实性成分比例;采用评分法对空泡征、胸膜牵拉征、血管集束征三个影像特征进行量化评分,利用单因素方差分析各CT特征在不同病理分组间的差异,利用神经网络法将病例随机分为培训组(103例)和检验组(100例),建立各CT特征与GGN病理之间的预测模型。结果203例肺GGN中AAH 20例,AIS 26例,MIA 74例,I-ADC 83例。四组病理类型间的结节性质、直径、实性成分比例以及三个影像特征通过单因素方差分析均存在显著性差异(P0.05)。基于此数据而使用神经网络的"多层感知器"(MLP)建立预测模型。培训组总体预测准确率为80.6%(AAH 92.9%,AIS 38.5%,MIA 91.2%,I-ADC81.0%)检验组预测总体准确率为72.0%(AAH 50.0%,AIS 46.2%,MIA 72.5%,I-ADC 82.9%),各自变量在模型中的重要性WTMW/WTLW(0.270,100%),影像特征评分(0.263,97.6%),WTMW(0.099,36.7%),WTLW(0.097,36.0%),胸膜牵拉征(0.085,31.5%),血管集束征(0.084,31.0%),空泡征(0.051,18.8%),内核(0.027,9.9%),结节密度(0.025,9.4%)。结论基于神经网络建立的GGN侵袭性CT预测模型可用于GGN病理侵袭性评估。
[Abstract]:Objective to establish a CT model for predicting the invasion of GGNN by neural network analysis. Methods the CT features of 203 cases of lung adenocarcinoma proved by operation and pathology were analyzed retrospectively. The basic information of the patients was collected. The density of pulmonary nodules (pure glassy nodules or mixed glassy nodules) was calculated to determine whether they contained the kernel, size, and proportion of solid components. Three imaging features, vacuole sign, pleural traction sign and vascular cluster sign, were scored quantitatively by scoring method. The differences of CT features among different pathological groups were analyzed by univariate variance analysis. The patients were randomly divided into training group (n = 103) and test group (n = 100). Results among the 203 cases of pulmonary GGN, 20 cases had AAH and 26 cases had GGN. 83 cases of I-ADC. The nodules and diameters between the four pathological types. There were significant differences in the proportion of real components and the three image features by single factor ANOVA (P0.05). Based on this data, a multilayer perceptron based on neural network (MLP) was used. The prediction model was established. The overall prediction accuracy of the training group was 80.6% and AAH 92.9%. The overall accuracy of prediction in the AIS 38.5 and MIA91.2 (I-ADC81.0) test group was 72.0 and AIS 46.2%. MIA 72.5 and I-ADC 82.9, the importance of their variables in the model WTM / WTL WN 0.270 / 100, image feature score of 0.263. 97.6% WTMW0.099 / 36.7T / WTLWN 0.0979 / 36.00, and 0.085 / 31.5 for pleural traction). The vascular cluster sign was 0.084% 31.0%, the vacuole sign was 0.051% and 18.8%, the nucleus was 0.027% 9. 9%, and the nodular density was 0. 025%. Conclusion the predictive model of GGN invasive CT based on neural network can be used to evaluate the pathological invasiveness of GGN.
【作者单位】：天津市天津医科大学肿瘤医院放射科;
【基金】：天津医科大学肿瘤医院科研项目(编号:Y1602-1)
【分类号】：R730.44;R734.2
【正文快照】： 近年来,随着低剂量多层螺旋CT(low dose CT,电压120 k V,自适应管电流,层厚1.25 mm,螺距LDCT)在早期肺癌筛查中的广泛应用,越来越多的0.984∶1,旋转时间0.6 s,视野(FOV)400 mm,重组肺磨玻璃密度结节(ground glass nodules,GGN)被早层厚1.25 mm,噪声指数N=14,自适应统计迭代重

【相似文献】