Value of LASSO-Logistic Regression Model Incorporating CT Features and GLCM Parameters in Differentiating Benign and Malignant Solitary Pulmonary Nodules
-
摘要:
目的 探讨基于计算机断层扫描(computed tomography,CT)征象及灰度共生矩阵(gray-level co-occurrence matrix,GLCM)参数构建最小绝对收缩和选择算法(least absolute shrinkage and selection operator,LASSO)-Logistic回归模型鉴别孤立性肺结节良恶性的价值。 方法 回顾性选取2021年12月1日至2025年6月30日期间铜陵市枞阳县人民医院收治的300例孤立性肺结节患者进行研究,按照7∶3的比例随机分为建模队列(n = 210)与验证队列(n = 90),依据病理结果将建模队列患者分为恶性组(n = 161)和良性组(n = 49),比较两组患者临床资料和影像学特征,采用LASSO回归分析影响因素的重要性并处理共线性问题,通过Spearman秩相关分析法分析上述因素间的关系,并将影响因素纳入Logistic回归分析,分析孤立性肺结节良恶性的影响因素。利用R软件建立列线图预测模型,并绘制受试者工作特征曲线(receiver operating characteristic curve,ROC)进行验证,通过校准曲线和决策曲线评估预测价值。 结果 建模队列与验证队列患者临床资料及影像学特征比较无差异(P > 0.05)。210例孤立性肺结节患者中,恶性病变161例(76.67%),纳入恶性组;余下良性病变49例(23.33%),纳入良性组。恶性组的热休克蛋白90α(heat shock protein 90 alpha,HSP90α)水平、分叶征占比、毛刺征占比、含磨玻璃影(ground-glass opacity,GGO)成分占比、非实性结节占比、GLCM参数-熵均高于良性组,平直边占比、GLCM参数-角二阶矩水平均低于良性组。LASSO回归筛选得到HSP90α、平直边、含GGO成分、密度、GLCM参数-角二阶矩、GLCM参数-熵共6项影响因素。相关性分析发现,HSP90α、含GGO成分、密度、GLCM参数-熵与孤立性肺结节良恶性情况之间呈现正相关(rs = 0.309、0.305、0.357、0.312),而平直边、GLCM参数-角二阶矩与孤立性肺结节良恶性情况之间呈现负相关(rs = −0.326、−0.358)。将各因素纳入Logistic预测模型,发现HSP90α、平直边、含GGO成分、密度、GLCM参数-角二阶矩、GLCM参数-熵是孤立性肺结节良恶性情况的影响因素,P均 < 0.05。以上述结果为基础构建列线图模型,并用验证队列进行验证,ROC曲线显示,该模型建模队列曲线下面积(area under the curve,AUC)值为0.945;验证队列的AUC值为0.935。校准曲线显示,校准曲线和参考曲线相近;同时阈值范围中预测模型净获益率较高。 结论 HSP90α、平直边、含GGO成分、密度、GLCM参数-角二阶矩、GLCM参数-熵可为孤立性肺结节良恶性鉴别诊断提供参考依据。 Abstract:Objective To investigate the value of a least absolute shrinkage and selection operator (LASSO)-Logistic regression model based on computed tomography (CT) imaging features and gray-level co-occurrence matrix (GLCM) parameters in differentiating benign and malignant solitary pulmonary nodules (SPNs). Methods A retrospective study was conducted on 300 patients with SPNs admitted to Zongyang County People's Hospital, Tongling City, from December 1, 2021 to June 30, 2025. The patients were randomly divided into a modeling cohort (n = 210) and a validation cohort (n = 90) at a ratio of 7∶3. Based on pathological results, patients in the modeling cohort were further classified into a malignant group (n = 161) and a benign group (n = 49). Clinical data and imaging features were compared between the two groups. LASSO regression analysis was used to evaluate the importance of influencing factors and address collinearity. Spearman rank correlation analysis was performed to examine the relationships among these factors. Logistic regression analysis was conducted to identify influencing factors for benignity and malignancy of isolated pulmonary nodules. A nomogram prediction model was established using R software, and receiver operating characteristic (ROC) curves were plotted for validation. Calibration curves and decision curves were used to assess predictive value. Results No significant differences in clinical data or imaging features were observed between the modeling and validation cohorts (P > 0.05). Among the 210 patients in the modeling cohort, 161 (76.67%) were malignant lesions (malignant group) and 49 (23.33%) were benign lesions (benign group). Compared with the benign group, the malignant group showed higher levels of heat shock protein 90 alpha (HSP90α), higher proportions of lobulation sign, spiculation sign, ground-glass opacity (GGO) component, and non-solid nodules, as well as higher GLCM entropy. Conversely, the malignant group had a lower proportion of flat margin and lower GLCM angular second moment (ASM) levels. LASSO regression identified six influencing factors: HSP90α, flat margin, presence of GGO component, density, GLCM-ASM, and GLCM entropy. Correlation analysis revealed that HSP90α, presence of GGO component, density, and GLCM entropy were positively correlated with malignancy (rs = 0.309, 0.305, 0.357, 0.312, respectively), while flat margin and GLCM-ASM were negatively correlated with malignancy (rs = −0.326, −0.358, respectively). Logistic regression analysis confirmed that HSP90α, straight edge, GGO component, density, GLCM parameter-angular second moment, and GLCM parameter-entropy were independent predictors of benign versus malignant SPNs (all P < 0.05). Based on these findings, a nomogram model was constructed and validated using the validation cohort. ROC curves demonstrated that the model achieved an area under the curve (AUC) of 0.945 in the modeling cohort and 0.935 in the validation cohort. Calibration curves showed close alignment between the predicted and reference curves. Additionally, the prediction model demonstrated high net benefit across the threshold range. Conclusion HSP90α, flat margin, GGO component, density, GLCM-ASM, and GLCM entropy can serve as reference indicators for differential diagnosis of benign and malignant isolated pulmonary nodules. -
Key words:
- Solitary pulmonary nodule /
- Tomography /
- Computed tomography /
- Prediction model
-
表 1 建模队列与验证队列患者临床资料及影像特征比较[($ \bar x \pm s $)/n(%)]
Table 1. Comparison of clinical data and imaging features between modeling cohort and validation cohort [($ \bar x \pm s $)/n(%)]
项目 建模队列
(n=210)验证队列
(n=90)χ2/t P 临床资料 性别 0.280 0.597 男 105(50.00) 42(46.67) 女 105(50.00) 48(53.33) 年龄/岁 57.05 ± 5.81 57.11 ± 5.76 0.082 0.935 体质量指数
(kg/m2)23.48 ± 2.54 23.50 ± 2.61 0.062 0.951 吸烟史 21(10.00) 10(11.11) 0.084 0.772 饮酒史 22(10.48) 10(11.11) 0.027 0.870 糖尿病史 26(12.38) 11(12.22) 0.001 0.969 高血压史 38(18.10) 16(17.78) 0.004 0.948 收缩压(mmHg) 129.01 ± 8.05 128.93 ± 8.12 0.079 0.937 舒张压(mmHg) 79.54 ± 6.01 80.01 ± 5.95 0.623 0.534 糖类抗原CA125(U/mL) 15.82 ± 3.60 15.61 ± 3.45 0.469 0.640 糖类抗原CA724(U/mL) 5.51 ± 0.94 5.60 ± 1.02 0.741 0.460 HSP90α(pg/mL) 208.63 ± 18.16 210.12 ± 18.14 0.651 0.515 影像特征 结节位置 0.003 0.959 上叶 130(61.90) 56(62.22) 其他 80(38.10) 34(37.78) 结节最大径/cm 1.78 ± 0.63 1.81 ± 0.65 0.374 0.708 长短径比值 1.29 ± 0.25 1.30 ± 0.26 0.314 0.754 分叶征 168(80.00) 71(78.89) 0.048 0.827 毛刺征 144(68.57) 62(68.89) 0.003 0.957 平直边 48(22.86) 19(21.11) 0.111 0.739 空泡征 26(12.38) 11(12.22) 0.001 0.969 含GGO成分 158(75.24) 72(80.00) 0.799 0.372 纤维条索 41(19.52) 18(20.00) 0.009 0.924 胸膜牵拉 63(30.00) 26(28.89) 0.037 0.847 钙化 14(6.67) 7(7.78) 0.119 0.730 边界 0.003 0.953 清晰 158(75.24) 68(75.56) 模糊 52(24.76) 22(24.44) 密度 0.003 0.955 非实性结节 151(71.90) 65(72.22) 实性结节 59(28.10) 25(27.78) GLCM参数-
对比度1109.48 ± 160.521110.06 ± 159.510.029 0.977 GLCM参数-
相关性(×10−3)0.31 ± 0.05 0.30 ± 0.06 1.492 0.137 GLCM参数-
角二阶矩1.42 ± 0.76 1.43 ± 0.71 0.106 0.915 GLCM参数-熵 6.76 ± 0.48 6.72 ± 0.50 0.653 0.514 GLCM参数-
逆差距(×10−3)49.12 ± 6.23 49.19 ± 6.18 0.089 0.929 注:HSP90α:热休克蛋白90α;GGO:磨玻璃影;GLCM:灰度共生矩阵。 表 2 建模队列良性组与恶性组的临床资料及影像特征比较[($ \bar{x} $ ± s)/n(%)]
Table 2. Comparison of clinical data and imaging features between benign group and malignant group in modeling cohort[($ \bar{x} $ ± s)/n(%)]
项目 恶性组
(n=161)良性组
(n=49)χ2/t P 临床资料 性别 0.027 0.870 男 80(49.69) 25(51.02) 女 81(50.31) 24(48.98) 年龄(岁) 56.48 ± 5.12 57.62 ± 4.56 1.398 0.163 体质量指数
(kg/m2)23.45 ± 2.13 23.52 ± 2.87 0.185 0.854 吸烟史 16(9.94) 5(10.20) 0.047 0.828 饮酒史 15(9.32) 7(14.29) 0.989 0.320 糖尿病史 20(12.42) 6(12.24) 0.001 0.974 高血压史 25(15.53) 13(26.53) 3.068 0.080 收缩压(mmHg) 130.05 ± 8.06 128.72 ± 7.54 1.026 0.306 舒张压(mmHg) 79.86 ± 6.23 79.01 ± 6.95 0.814 0.417 CA125(U/mL) 15.86 ± 3.78 15.80 ± 3.51 0.099 0.921 CA724(U/mL) 5.60 ± 0.83 5.35 ± 0.89 1.815 0.071 HSP90α
(pg/mL)225.71 ± 39.72 199.40 ± 34.58 4.679 <0.001 影像特征 结节位置 0.013 0.911 上叶 100(62.11) 30(61.22) 其他 61(37.89) 19(38.78) 结节最大径(cm) 1.81 ± 0.68 1.67 ± 0.52 1.327 0.186 长短径比值 1.25 ± 0.27 1.33 ± 0.24 1.862 0.064 分叶征 136(84.47) 32(65.31) 8.625 0.003 毛刺征 117(72.67) 27(55.10) 5.380 0.020 平直边 28(17.39) 20(40.82) 11.691 0.001 空泡征 19(11.80) 6(12.24) 0.007 0.933 含GGO成分 129(80.12) 29(59.18) 8.842 0.003 纤维条索 30(18.63) 11(22.45) 0.348 0.555 胸膜牵拉 49(30.43) 14(28.57) 0.062 0.803 钙化 10(6.21) 4(8.16) 0.023 0.879 边界 0.184 0.668 清晰 120(74.53) 38(77.55) 模糊 41(25.47) 11(22.45) 密度 26.694 <0.001 非实性结节 130(80.75) 21(42.86) 实性结节 31(19.25) 28(57.14) GLCM参数-
对比度1084.86 ± 165.051073.06 ± 159.480.442 0.659 GLCM参数-
相关性(×10−3)0.31 ± 0.06 0.30 ± 0.05 1.060 0.291 GLCM参数-
角二阶矩1.11 ± 0.49 1.72 ± 1.09 5.519 <0.001 GLCM参数-熵 6.91 ± 0.44 6.57 ± 0.51 4.559 <0.001 GLCM参数-
逆差距(×10−3)49.56 ± 5.54 50.14 ± 6.97 0.602 0.548 注:HSP90α:热休克蛋白90α;GGO:磨玻璃影;GLCM:灰度共生矩阵。 表 3 影响孤立性肺结节良恶性的Logistic回归分析
Table 3. Logistic regression analysis of benign and malignant solitary pulmonary nodules
因素 赋值 β S.E Waldχ2 OR(95%CI) P HSP90α 实测值 0.02 0.01 17.70 1.02(1.01~1.03) <0.001* 平直边 1=是,0=否 −1.19 0.36 11.02 0.31(0.15~0.61) <0.001* 含GGO成分 1=是,0=否 1.02 0.35 8.47 2.78(1.40~5.54) 0.004* 密度 1=非实性结节,0=实性结节 1.72 0.35 24.03 5.59(2.81~11.13) <0.001* GLCM参数-角二阶矩 实测值 −1.21 0.28 18.35 0.30(0.17~0.51) <0.001* GLCM参数-熵 实测值 1.77 0.41 16.76 5.89(2.62~13.25) <0.001* *P<0.05。 -
[1] 齐庆彬, 苏海涛, 胡基刚, 等. 血清SHOX2、RASSF1A、PTGER4甲基化水平对孤立性肺结节良恶性及病理分型的诊断价值[J]. 诊断病理学杂志, 2025, 32(8): 1010-1015. [2] 马波, 陈均, 朱进, 等. 采用MR扩散加权成像鉴别良、恶性孤立性肺结节病变的可靠性研究[J]. 中国CT和MRI杂志, 2023, 21(1): 70-71. [3] Hou X, Wu M, Chen J, et al. Establishment and verification of a prediction model based on clinical characteristics and computed tomography radiomics parameters for distinguishing benign and malignant pulmonary nodules[J]. J Thorac Dis, 2024, 16(3): 1984-1995. [4] Shen J, Du H, Wang Y, et al. A novel nomogram model combining CT texture features and urine energy metabolism to differentiate single benign from malignant pulmonary nodule[J]. Front Oncol, 2022, 12: 1035307. doi: 10.3389/fonc.2022.1035307 [5] 马赵, 夏春华, 王峻奇, 等. 孤立性肺结节CT征像在鉴别良恶性结节中的价值[J]. 医学影像学杂志, 2021, 31(4): 582-585. [6] 张静, 武志峰, 鄂林宁, 等. 肺结节(≤2 cm)及其周围组织的影像组学特征在其良恶性鉴别中的价值[J]. 中国临床医学影像杂志, 2020, 31(7): 478-481+485. [7] 赵波, 刘永利, 李京京. 血浆外泌体miR-4306表达水平联合CT征象对单发非实性肺结节良恶性鉴别的价值研究[J]. 现代检验医学杂志, 2023, 38(4): 46-50+77. [8] Chen B, Li Q, Hao Q, et al. Malignancy risk stratification for solitary pulmonary nodule: A clinical practice guideline[J]. J Evid Based Med, 2022, 15(2): 142-151. [9] 王德清, 赵振华. 基于CT灌注成像影像组学在良恶性孤立性肺结节鉴别诊断中的应用[J]. 实用放射学杂志, 2023, 39(1): 37-40. [10] 周倩倩, 曲歌平, 张晓军, 等. 218例老年患者孤立性肺结节的影像组学分析[J]. 中华保健医学杂志, 2021, 23(3): 250-254. [11] 中华医学会糖尿病学分会. 中国2型糖尿病防治指南(2013年版)[J]. 中华糖尿病杂志, 2014, 6(7): 447-498. [12] 国家心血管病中心国家基本公共卫生服务项目基层高血压管理办公室, 国家基层高血压管理专家委员会, 李静. 国家基层高血压防治管理指南2025版[J]. 中华心血管病杂志, 2025, 53(9): 977-991. [13] Liang G, Yu W, Liu S Q, et al. The value of radiomics based on dual-energy CT for differentiating benign from malignant solitary pulmonary nodules[J]. BMC Med Imag, 2022, 22(1): 95. [14] Chen X, Zhu X, Yan W, et al. Serum lncRNA THRIL predicts benign and malignant pulmonary nodules and promotes the progression of pulmonary malignancies[J]. BMC Cancer, 2023, 23(1): 755. [15] 王凯跃, 金欢乐, 金欢迪, 等. 舌象表征与肺结节影像组学特征相关性分析[J]. 中国医学计算机成像杂志, 2025, 31(4): 472-476. [16] 王丽红, 马小花, 高晶晶, 等. 血清PCNA、HSP90α水平对孤立性肺结节良恶性的鉴别诊断价值[J]. 中国实验诊断学, 2025, 29(6): 693-697. [17] 李红英, 胡鑫, 宋瑞祥. 高分辨CT影像学特征对孤立性肺结节良恶性的鉴别诊断效能[J]. 海南医学, 2022, 33(19): 2540-2543. [18] Feng B, Chen X, Chen Y, et al. Radiomics nomogram for preoperative differentiation of lung tuberculoma from adenocarcinoma in solitary pulmonary solid nodule[J]. Eur J Radiol, 2020, 128: 109022. [19] 晏睿滢, 李华秀, 丁莹莹, 等. 孤立性肺结节良恶性预测模型的建立及验证[J]. 放射学实践, 2020, 35(3): 352-359. [20] 陈金钗, 胡小飞, 梁李娟, 等. 多模态融合建立孤立性肺结节良恶性预测模型[J]. 临床肺科杂志, 2025, 30(9): 1341-1347. [21] 余翔, 张敏, 刘建光. 孤立性肺结节影像学特征的logistic回归分析[J]. 蚌埠医学院学报, 2019, 44(7): 936-939. [22] 吴宇强, 秦涛, 马晓臣, 等. 实性孤立性肺结节CT影像组学参数测量的可重复性研究[J]. 放射学实践, 2020, 35(9): 1106-1111. [23] Shen J, Du H, Wang Y, et al. A novel nomogram model combining CT texture features and urine energy metabolism to differentiate single benign from malignant pulmonary nodule[J]. Front Oncol, 2022, 12: 1035307. [24] Zhang H, Zhang H, Liu K, et al. Predictive value of artificial intelligence-based quantitative CT feature analysis for diagnosing the pathological types of pulmonary nodules[J]. Eur Radiol, 2026, 36(4): 3119-3130. [25] Chen Q, Sun H, Jiang Q, et al. Preoperative CT-based artificial intelligence-derived quantitative parameters and imaging features for predicting the invasiveness of histologically confirmed subcentimeter adenocarcinomatous nodules: a two-center study[J]. Quant Imaging Med Surg, 2025, 15(12): 12593-12606. [26] Jiang L, Jiang L, Zhou Y, et al. Artificial intelligence-assisted quantitative CT parameters in predicting the degree of risk of solitary pulmonary nodules[J]. Ann Med, 2024, 56(1): 2405075. [27] Ding Y, Gao Z, Kuang K, et al. Improving the efficiency of identifying malignant pulmonary nodules before surgery via a combination of artificial intelligence CT image recognition and serum autoantibodies[J]. Eur Radiol, 2023, 33(5): 3092-3102. -
下载: