Department of Therapeutic Radiology, School of Medicine, Yale University, New Haven, Connecticut, United States of America.
PLoS One. 2018 Oct 24;13(10):e0205264. doi: 10.1371/journal.pone.0205264. eCollection 2018.
The objective of this study is to train and validate a multi-parameterized artificial neural network (ANN) based on personal health information to predict lung cancer risk with high sensitivity and specificity. The 1997-2015 National Health Interview Survey adult data was used to train and validate our ANN, with inputs: gender, age, BMI, diabetes, smoking status, emphysema, asthma, race, Hispanic ethnicity, hypertension, heart diseases, vigorous exercise habits, and history of stroke. We identified 648 cancer and 488,418 non-cancer cases. For the training set the sensitivity was 79.8% (95% CI, 75.9%-83.6%), specificity was 79.9% (79.8%-80.1%), and AUC was 0.86 (0.85-0.88). For the validation set sensitivity was 75.3% (68.9%-81.6%), specificity was 80.6% (80.3%-80.8%), and AUC was 0.86 (0.84-0.89). Our results indicate that the use of an ANN based on personal health information gives high specificity and modest sensitivity for lung cancer detection, offering a cost-effective and non-invasive clinical tool for risk stratification.
本研究旨在训练和验证一种基于个人健康信息的多参数人工神经网络（ANN），以实现高灵敏度和特异性的肺癌风险预测。我们使用了 1997-2015 年全国健康访谈调查成人数据来训练和验证我们的 ANN，输入包括：性别、年龄、BMI、糖尿病、吸烟状况、肺气肿、哮喘、种族、西班牙裔、高血压、心脏病、剧烈运动习惯和中风史。我们确定了 648 例癌症和 488418 例非癌症病例。在训练集中，敏感性为 79.8%（95%CI，75.9%-83.6%），特异性为 79.9%（79.8%-80.1%），AUC 为 0.86（0.85-0.88）。在验证集中，敏感性为 75.3%（68.9%-81.6%），特异性为 80.6%（80.3%-80.8%），AUC 为 0.86（0.84-0.89）。我们的结果表明，基于个人健康信息的 ANN 用于肺癌检测具有较高的特异性和适度的敏感性，为风险分层提供了一种具有成本效益且非侵入性的临床工具。
