| 李蕊,杨晓菲,詹鹏超,谷艳博,陈岩,高剑波.基于光子计数CT定量参数机器学习模型预测肺癌病理亚型[J].中国医学影像技术,2026,42(3):342~346 |
| 基于光子计数CT定量参数机器学习模型预测肺癌病理亚型 |
| Machine learning model based on photon-counting CT quantitative parameters for predicting pathological subtype of lung cancer |
| 投稿时间：2025-09-14 修订日期：2026-03-11 |
| DOI：10.13929/j.issn.1003-3289.2026.03.005 |
| 中文关键词: 肺肿瘤 体层摄影术,X线计算机 病理学 诊断,鉴别 前瞻性研究 |
| 英文关键词:lung neoplasms tomography,X-ray computed pathology diagnosis,differential prospective studies |
| 基金项目:河南省医学科技攻关计划联合共建项目(LHGJ20240246)。 |
|
| 摘要点击次数: 68 |
| 全文下载次数: 38 |
| 中文摘要: |
| 目的 探讨基于光子计数CT(PCCT)定量参数构建的机器学习(ML)模型预测肺癌病理亚型的价值。方法 前瞻性纳入96例经术后病理证实的肺癌患者,包括51例腺癌(AC)(AC组)、24例鳞状细胞癌(SCC)(SCC组)及21例小细胞肺癌(SCLC)(SCLC组)。采用Boruta算法基于治疗前增强PCCT筛选组间差异有统计学意义的参数并构建决策树(DT)ML模型;绘制受试者工作特征曲线,计算曲线下面积(AUC),评估模型预测不同病理亚型肺癌的效能。结果 SCC组和SCLC组病灶的动脉期碘浓度(IC)、标准化IC(NIC)、40~70 keV CT值、K40~80 keV,以及静脉期NIC、40~80 keV CT值均低于AC组(P均＜0.05),而前二者差异均无统计学意义(P均＞0.05);SCLC组、SCC组及AC组静脉期IC及静脉期K40~80 keV依次升高(P均＜0.05)。基于18个PCCT参数构建的DT模型预测肺癌亚型的整体准确率为80.21%,其预测AC、SCC及SCLC的AUC分别为0.912、0.771及0.945。结论 基于PCCT定量参数构建的DT模型用于预测肺癌病理亚型具有较高价值。 |
| 英文摘要: |
| Objective To investigate the value of machine learning (ML) model based on quantitative parameters of photon-counting CT (PCCT) for predicting pathological subtype of lung cancer. Methods A total of 96 patients with pathologically confirmed lung cancer were prospectively enrolled, including 51 cases of adenocarcinoma (AC) (AC group), 24 cases of squamous cell carcinoma (SCC) (SCC group) and 21 cases of small cell lung cancer (SCLC) (SCLC group). Based on pre-treatment contrast-enhanced PCCT, valuable parameters were selected using Boruta algorithm, and a decision tree (DT) ML model was constructed. Receiver operating characteristic curve was drawn, the area under the curve (AUC) was calculated to evaluate the performance of this model for identifying each pathological subtype of lung cancer. Results In SCC and SCLC groups, the arterial phase iodine concentration (IC), normalized IC (NIC), CT values at 40—70 keV, K40—80 keV, as well as venous phase NIC and CT values at 40—80 keV were all lower than those in AC group (all P＜0.05), but no significant difference was found between SCC and SCLC groups (all P＞0.05). Venous phase IC and venous phase K40—80 keV increased sequentially in SCLC, SCC and AC groups (all P＜0.05). A total of 18 PCCT parameters were selected, and the overall diagnostic accuracy of DT model constructed based on these parameters for predicting pathological subtype of lung cancer was 80.21%, with AUC for identifying AC, SCC and SCLC of 0.912, 0.771 and 0.945, respectively. Conclusion DT model constructed based on PCCT quantitative parameters had high value for predicting pathological subtype of lung cancer. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
