| 陈疆红.人工智能肺结节辅助诊断系统预测亚实性结节恶性概率的效果研究[J].中国医学影像技术,2020,36(4): |
| 人工智能肺结节辅助诊断系统预测亚实性结节恶性概率的效果研究 |
| The efficacy of artificial intelligence-assisted pulmonary nodule diagnosis system on predicting the malignant probability of subsolid nodules |
| 投稿时间：2019-12-12 修订日期：2020-04-16 |
| DOI： |
| 中文关键词: 人工智能，深度学习，肺癌筛查，恶性概率，增强CT扫描 |
| 英文关键词:artificial intelligence, deep learning, lung cancer screening, malignant probability, enhanced CT scan |
| 基金项目:“使命”人才计划（SML20150101） |
|
| 摘要点击次数: 1573 |
| 全文下载次数: 522 |
| 中文摘要: |
| 目的 评价人工智能（artificial intelligence，AI）肺结节辅助诊断系统预测亚实性肺结节恶性概率的效能。方法 本研究共纳入86例患者。将亚实性结节按病理组织学结果分为三组：组1为浸润前病变；组2为微浸润腺癌；组3为浸润性腺癌。将患者术前胸部CT检查数据导入AI肺结节识别软件，记录AI软件自动计算的亚实性结节CT值、体积及恶性概率预测值。分别在平扫、增强后动脉期及延迟期行三组间CT值、体积及恶性概率预测值的多个独立样本非参数检验（Kruskal-Wallis H检验），之后分别行每组平扫与增强后的配对样本非参数检验（Wilcoxon检验）。结果 共88个亚实性结节，其中27、28及33个结节分别纳入组1、组2和组3。AI软件对组1结节在平扫、增强后动脉期、延迟期的恶性概率预测值分别为74.60%±19.76%、73.58%±17.18%、77.40±18.67%；组2分别为89.97%±8.55%、85.17%±13.92%、90.14%±7.70%；组3分别为94.25%±7.04%、94.18%±5.72%和94.01%±6.33%。三组结节的恶性概率预测值在三期扫描均有统计学差异（P＜0.001），且三组间CT值、体积比较亦有统计学差异。结论 目前应用的基于深度学习的AI肺结节辅助诊断系统可以协助我们判定肺腺癌亚实性结节的侵袭程度。 |
| 英文摘要: |
| Objective To evaluate the efficacy of artificial intelligence (AI)-assisted pulmonary nodule diagnosis system on predicting the malignant probability of subsolid pulmonary nodule. Methods Eighty-six patients were enrolled in this study. The resected subsolid nodules were divided into three groups according to histopathological results. Preinvasive lesions were classified into group 1 while microinvasive adenocarcinoma and invasive adenocarcinoma as group 2 and group 3, respectively. The preoperative chest CT data were loaded into AI pulmonary nodule diagnosis system for the measurements of CT value and volume, and the prediction of malignant probability for each subsolid nodule. Kruskal-Wallis H test was used to examine the differences of measured and CT value and volume, and the predicted malignant probability among these three groups based on either plain CT or enhancement CT. Then paired-samples Wilcoxon test was performed to analyze the difference of these measured and predicted values between plain CT and enhancement CT for each group. Results 88 subsolid nodules were enrolled, in which 27, 28, and 33 cases were included in group 1, group 2, and group 3, respectively. Of note, all the resected subsolid nodules were detected by the AI pulmonary nodule diagnosis system. The predicted malignant probability of subsolid nodules at plain CT，arterial phase, and venous phase were 74.60%±19.76%, 73.58%±17.18% and 77.40±18.67% for group 1, 89.97%±8.55%, 85.17%±13.92% and 90.14%±7.70% for group 2, 94.25%±7.04%, 94.18%±5.72% and 94.01%±6.33% for group 3, respectively. Statistically significant differences were observed in the measured CT value and volume, and the predicted malignant probability among these three groups in either of these scanning phases. Conclusions At present, the deep learning-based AI-assisted pulmonary nodule diagnosis system can help us determine the invasiveness of pulmonary adenocarcinoma manifested as subsolid nodule. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
