南方医科大学学报

南方医科大学学报 ›› 2024, Vol. 44 ›› Issue (2): 397-404.doi: 10.12122/j.issn.1673-4254.2024.02.23

• • 上一篇    下一篇

基于临床-影像组学列线图模型鉴别局灶性机化性肺炎与肺腺癌

刘云泽，李宬润，郭俊唐，刘 阳   

  1. 中国人民解放军总医院研究生院，第一医学中心胸外科，北京 100853
  • 发布日期:2024-03-14

A clinical-radiomics nomogram for differentiating focal organizing pneumonia and lung adenocarcinoma

LIU Yunze, LI Chengrun, GUO Juntang, LIU Yang   

  1. Graduate School, Department of Thoracic Surgery of First Medical Center, Chinese PLA General Hospital, Beijing 100853, China
  • Published:2024-03-14

RichHTML

1

PDF

43

摘要： 目的 探讨临床-影像组学组合模型对于局灶性机化性肺炎和周围性肺腺癌的鉴别诊断价值。方法 回顾性分析2019年1月~2022年12月解放军总医院第一医学中心胸外科术后病理证实为局灶性机化性肺炎的60例患者，根据倾向评分匹配，1∶2选择出120例肺腺癌患者，收集其临床和影像资料。临床相关资料采用Logistic回归筛选独立危险因素，影像组学相关资料采用随机抽样的方法将患者按照8∶2的比例分为训练集（144例）与测试集（36例），采用Python3.7数据包提取1835个特征，经过统计学处理，结合临床资料建立模型，在训练集和测试集中应用受试者操作特征（ROC）曲线对模型进行验证，评价该模型针对局灶性机化性肺炎与周围型肺腺癌的鉴别诊断效能，并建立列线图模型。结果 统计学分析，发现“过敏史”为机化性肺炎的独立危险因素（P=0.016），血液学结果无明显差异（P>0.05）。在影像组学特征提取和降维后，筛选出30个对于鉴别两种疾病有意义的影像组学标签，其中对于鉴别意义前3位的影像组学标签为“GraylevelNonUniformity，灰度游程矩阵中的灰度非均匀性”、“SizeZoneNonUniformity，灰度级大小区域矩阵特征中的尺寸区域非均匀性”、“shape-Sphericity，原始图像形状的球形度”。根据临床-影像组学特征建立的预测模型在训练集中受试者工作特征（ROC）曲线下面积（AUC）为0.909（95% CI：0.8550~0.9627）、测试集为0.901（95% CI：0.8030~0.9989）。模型在训练集敏感度、特异性及准确率分别为85.4%、83.5%、84.0%，测试集分别为94.7%、70.6%、83.3%。结论 基于临床-影像组学的列线图模型对局灶性机化性肺炎与周围型肺腺癌有较好的鉴别诊断效能，有望降低因诊断不明导致的过度治疗，使患者获益。

关键词: 局灶性机化性肺炎；肺腺癌；影像组学；列线图

Abstract: Objective To evaluate the performance of a clinical-radiomics model for differentiating focal organizing pneumonia (FOP) and lung adenocarcinoma (LUAD). Methods We retrospectively analyzed the data of 60 patients with FOP confirmed by postoperative pathology at the First Medical Center of the Chinese PLA General Hospital from January, 2019 to December, 2022, who were matched with 120 LUAD patients using propensity score matching in a 1∶2 ratio. The independent risk factors for FOP were identified by logistic regression analysis of the patients' clinical data. The cohort was divided into a training set (144 patients) and a test set (36 patients) by random sampling. Python 3.7 was used for extracting 1835 features from CT image data of the patients. The radiographic features and clinical data were used to construct the model, whose performance was validated using ROC curves in both the training and test sets. The diagnostic efficacy of the model for FOP and LUAD was evaluated and a diagnostic nomogram was constructed. Results Statistical analysis revealed that an history of was an independent risk factor for FOP (P=0.016), which was correlated with none of the hematological findings (P>0.05). Feature extraction and dimensionality reduction in radiomics yielded 30 significant labels for distinguishing the two diseases. The top 3 most discriminative radiomics labels were GraylevelNonUniformity, SizeZoneNonUniformity and shape-Sphericity. The clinical-radiomics model achieved an AUC of 0.909 (95% CI: 0.855-0.963) in the training set and 0.901 (95% CI: 0.803-0.999) in the test set. The model showed a sensitivity of 85.4%, a specificity of 83.5%, and an accuracy of 84.0% in the training set, as compared with 94.7% , 70.6% , and 83.3% in the test set, respectively. Conclusion The clinical-radiomics nomogram model shows a good performance for differential diagnosis of FOP and LUAD and may help to minimize misdiagnosis-related overtreatment and improve the patients' outcomes.

Key words: focal organizing pneumonia; lung adenocarcinoma; radiomics; nomogram