基于CT影像加权组学评分构建非小细胞肺癌的免疫治疗疗效预测模型

摘要/Abstract

摘要：

目的基于CT影像加权组学评分构建非小细胞肺癌（NSCLC）的免疫治疗疗效预测模型。方法选择2021年1月—2023年12月徐州医科大学附属沭阳医院收治的185例接受免疫治疗的NSCLC患者为研究对象，所有患者接受纳武利尤单抗（240 mg）连续治疗3个疗程，并根据实体瘤临床疗效评价标准1.1版评估疗效，根据疗效将患者分为治疗有效组和治疗无效组。基于治疗前动脉期CT图像，采用PyRadiomics工具对所有容积感兴趣区域进行重采样，并进行图像预处理（包括小波变换和拉普拉斯滤波器等方法），提取两组患者的多病灶影像组学特征，并通过Dr.Wise科研平台上对提取的特征进行标准化处理。同时，采用基于注意力机制的多示例学习（a-MIL）算法、最小绝对收缩和选择算子（LASSO）回归、logistic回归构建CT影像加权组学评分模型，并计算每例患者的CT影像加权组学评分。基于R软件（R4.3.3）和rms程序包绘制列线图，同时通过一致性指数（C-index）评估模型的一致性，采用受试者操作特征（ROC）曲线评估列线图模型对NSCLC的免疫治疗疗效预测性能，校准曲线用于评估模型预测概率与实际结果的一致性，决策曲线用于评估模型在不同阈值下的净收益。结果治疗有效组（66例）与治疗无效组（119例）患者的性别（χ²=3.86，P=0.049）、病理类型（χ²=8.41，P=0.015）、吸烟史（χ²=5.70，P=0.017）、治疗前肺内转移（χ²=5.88，P=0.015）比较，差异均有统计学意义。采用a-MIL算法对多病灶的原始组学特征在病例层面加权求和，共提取342个影像组学特征，其中162个特征一致性较好（组内相关系数>0.80）；经方差分析后剔除134个特征，剩余28个特征，经LASSO进行降维后，最终获得7个影像组学特征，分别为GLCMEnergy_angle45_offset、ShortRunEmphasis_angle90_offset1、最大灰度值、Spiculation、GLCMEnergy_angle45_offset7、Sphericity、Vessel。基于上述影像组学特征构建加权组学评分模型为：Radscore=0.624+0.022×GLCMEnergy_angle45_offset-0.227×ShortRunEmphasis_angle90_offset1+0.395×最大灰度值-8.687×Spiculation+0.384×GLCMEnergy_angle45_offset7-0.012×Sphericity-0.284×Vessel。治疗有效组NSCLC患者的CT影像加权组学评分（0.75±0.10）明显高于治疗无效组（0.43±0.14），差异有统计学意义（t=18.00，P<0.001）。ROC曲线分析显示，CT影像加权组学评分预测NSCLC免疫治疗疗效的曲线下面积（AUC）为0.96（95%CI为0.92~0.98），最佳截断值为0.62。多因素分析显示，CT影像加权组学评分≥0.62分（OR=14.77，95%CI为3.25~22.35，P<0.001）、病理类型（鳞状细胞癌）（OR=1.74，95%CI为1.35~3.52，P=0.035）、有吸烟史（OR=4.01，95%CI为1.05~15.30，P=0.042）和治疗前肺内转移（OR=1.20，95%CI为1.01~1.38，P=0.010）均为NSCLC免疫治疗有效的独立预测因素。基于上述4个变量构建预测NSCLC免疫治疗疗效的列线图模型，其模型验证结果显示，C-index为0.96（95%CI为0.93~0.99）。校准曲线分析显示，预测概率与实际概率一致性较好，趋近于理想曲线。ROC曲线分析显示，AUC为0.97（95%CI为0.94~0.99）。决策曲线分析显示，在2%~100%预测范围内，模型净获益。结论基于CT影像加权组学评分构建预测NSCLC患者免疫治疗疗效的列线图模型，具有较好的预测效能。

关键词:癌,非小细胞肺,免疫疗法,预测,影像组学

Abstract:

ObjectiveTo construct a predictive model for the efficacy of immunotherapy in non-small cell lung cancer （NSCLC） based on CT image-weighted radiomics score.MethodsA total of 185 patients with NSCLC who received immunotherapy in Affiliated Shuyang Hospital of Xuzhou Medical University from January 2021 to December 2023 were selected as the study objects. All patients underwent 3 consecutive cycles of nivolumab （240 mg） treatment，and therapeutic efficacy was evaluated using the Response Evaluation Criteria in Solid Tumors version 1.1. The patients were divided into treatment-effective group and treatment-ineffective group based on therapeutic outcomes. Based on pre-treatment arterial phase CT images，all volume regions of interest were resampled using PyRadiomics tool，and image preprocessing was performed （including methods such as Wavelet transform and Laplacian filtering）. The multi-focal imaging radiomics features of the two groups of patients were extracted，and standardized processing of the extracted features was carried out on the Dr. Wise research platform. At the same time，a CT image-weighted radiomics score model was constructed using an attention-based multi-instance learning （a-MIL） algorithm，the least absolute shrinkage and selection operator （LASSO） regression，and logistic regression，and the CT image-weighted radiomics score of each patient was calculated. The nomogram was plotted using R software （version R4.3.3） and the rms package. The concordance index （C-index） was used to evaluate the concordance of the model. Receiver operator characteristic （ROC） curve was used to evaluate the performance of the nomogram model in predicting immunotherapy efficacy in NSCLC. The calibration curve was used to evaluate the consistency of the predicted probabilities with the actual outcomes，while the decision curve was used to evaluate net benefit of the model across different thresholds.ResultsThere were statistically significant differences in sex （χ²=3.86，P=0.049），pathological type （χ²=8.41，P=0.015），smoking history （χ²=5.70，P=0.017），and pre-treatment pulmonary metastasis （χ²=5.88，P=0.015） between the treatment-effective group （n=66） and treatment-ineffective group （n=119）. The original multi-focal radiomics features were weighted and summated by a-MIL algorithm at case level，and a total of 342 imaging radiomics features were extracted，162 features among which had good consistency （intra-class correlation coefficient >0.80）. After variance analysis，134 features were eliminated，28 features remained. After dimensionality reduction by LASSO regression，7 imaging radiomics features were obtained. They were GLCMEnergy_angle45_offset，ShortRunEmphasis_angle90_offset1，maximum gray value，Spiculation，GLCMEnergy_angle45_offset7，Sphericity，and Vessel. Based on the above imaging radiomics features，the weighted radiomics score model was constructed as follows：Radscore=0.624+0.022×GLCMEnergy_angle45_offset-0.227×ShortRunEmphasis_angle90_offset1+0.395×maximum gray value-8.687×Spiculation+0.384×GLCMEnergy_angle45_offset7-0.012×Sphericity-0.284×Vessel. The CT image-weighted radiomics score in the treatment-effective group （0.75±0.10） was significantly higher than that in the treatment-ineffective group （0.43±0.14），with a statistically significant difference （t=18.00，P<0.001）. ROC curve analysis showed that the area under the curve （AUC） of CT image-weighted radiomics score for predicting immunotherapy efficacy of NSCLC was 0.96 （95%CI：0.92-0.98），and the optimal cutoff value was 0.62. Multivariate analysis showed CT image-weighted radiomics score ≥0.62 （OR=14.77，95%CI：3.25-22.35，P<0.001），pathological type （squamous cell carcinoma）（OR=1.74，95%CI：1.35-3.52，P=0.035），smoking history （OR=4.01，95%CI：1.05-15.30，P=0.042），and pre-treatment pulmonary metastasis （OR=1.20，95%CI：1.01-1.38，P=0.010） were all independent predictors of immunotherapy effectiveness in NSCLC. Based on the above 4 variables，a nomogram model was constructed to predict the immunotherapy efficacy of NSCLC，and the model validation results showed that the C-index was 0.96 （95%CI：0.93-0.99）. Calibration curve analysis showed good consistency of the predicted probabilities with the actual probabilities，closely aligning with the ideal curve. ROC curve analysis showed that AUC was 0.97 （95%CI：0.94-0.99）. Decision curve analysis showed that the model had a net benefit within the prediction range of 2% to 100%.ConclusionThe nomogram model based on CT image-weighted radiomics score is effective in predicting immunotherapy efficacy of NSCLC patients.

Key words:Carcinoma,non-small-cell lung,Immunotherapy,Forecasting,Radiomics

刘海燕, 张超. 基于CT影像加权组学评分构建非小细胞肺癌的免疫治疗疗效预测模型[J]. 国际肿瘤学杂志, 2025, 52(4): 202-208.

Liu Haiyan, Zhang Chao. A predictive model for immunotherapy efficacy in non-small cell lung cancer constructed based on CT image-weighted radiomics score[J]. Journal of International Oncology, 2025, 52(4): 202-208.

图/表7

表1

两组免疫治疗NSCLC患者临床资料分析（$\bar{x}±s$/例）"

临床资料	治疗有效组（n=66）	治疗无效组（n=119）	t/χ²值	P值
年龄（岁）	63.45±9.14	64.48±8.78	0.75	0.452
性别
男性	57	88	3.86	0.049
女性	9	31	3.86	0.049
病理类型
鳞状细胞癌	40	46
腺癌	24	65	8.41	0.015
其他	2	8
吸烟史
有	31	35	5.70	0.017
无	35	84	5.70	0.017
治疗前脑转移
有	5	11	0.15	0.699
无	61	108	0.15	0.699
治疗前肺内转移
有	19	56	5.88	0.015
无	47	63	5.88	0.015
T分期
T₁	6	12	0.83	0.843
T₂	11	25
T₃	31	55
T₄	18	27
N分期
N₀	10	17	3.12	0.374
N₁	16	17
N₂	10	21
N₃	30	64
M分期
M₀	23	30	1.93	0.165
M₁	43	89	1.93	0.165
ROS1基因突变
有	1	7	1.96	0.162
无	65	112	1.96	0.162
KRAS基因突变
有	5	18	2.22	0.136
无	61	101	2.22	0.136
EGFR基因突变
有	10	22	0.33	0.566
无	56	97	0.33	0.566
PD-L1表达
高表达	16	23
低表达	25	42	1.12	0.570
阴性	25	54

表1

图1

表2

影响185例NSCLC患者免疫治疗疗效的多因素分析"

变量	β值	SE值	Waldχ²值	OR值	95%CI	P值
CT影像加权组学评分（≥0.62分/<0.62分）	3.844	4.209	32.09	14.77	3.25~22.35	<0.001
病理类型（以其他类型为对照）
鳞状细胞癌	1.319	0.625	4.45	1.74	1.35~3.52	0.035
腺癌	0.728	0.617	1.39	0.48	0.14~1.62	0.078
性别（男/女）	1.019	0.838	1.48	2.77	0.54~14.31	0.224
吸烟史（有/无）	-1.389	0.683	4.14	4.01	1.05~15.30	0.042
治疗前肺内转移（有/无）	1.095	0.665	2.71	1.20	1.01~1.38	0.010
常量	14.071	2.551	30.42			<0.001

表2

图2

图3

图4

图5

参考文献22

[1]	张碧霞, 丁江华. EGFR突变型非小细胞肺癌EGFR-TKI获得性耐药后免疫治疗现状[J].国际肿瘤学杂志,2023,50(2): 97-101. DOI:10.3760/cma.j.cn371439-20220719-00020.
[2]	Chen P, Liu Y, Wen Y, et al. Non-small cell lung cancer in China[J].Cancer Commun (Lond),2022,42(10): 937-970. DOI:10.1002/cac2.12359.
[3]	Memmott RM, Wolfe AR, Carbone DP, et al. Predictors of response, progression-free survival, and overall survival in patients with lung cancer treated with immune checkpoint inhibitors[J].J Thorac Oncol,2021,16(7): 1086-1098. DOI:10.1016/j.jtho.2021.03.017. pmid:33845212
[4]	Sakurai E, Ishizawa H, Kiriyama Y, et al. γH2AX, a DNA double-strand break marker, correlates with PD-L1 expression in smoking-related lung adenocarcinoma[J].Int J Mol Sci,2022,23(12): 6679. DOI:10.3390/ijms23126679.
[5]	Cheng X, Wang L, Zhang Z. Prognostic significance of PD-L1 expression and CD8⁺TILs density for disease-free survival in surgically resected lung squamous cell carcinoma: a retrospective study[J].J Thorac Dis,2022,14(6): 2224-2234. DOI:10.21037/jtd-22-630.
[6]	Jeong H, Park HB, Hong J, et al. Identifying coronary artery calcification using chest x-ray radiographs and machine learning: the role of the radiomics score[J].J Thorac Imaging,2024,39(2): 119-126. DOI:10.1097/RTI.0000000000000757.
[7]	Guo Y, Xie X, Tang W, et al. Noninvasive identification of HER2-low-positive status by MRI-based deep learning radiomics predicts the disease-free survival of patients with breast cancer[J].Eur Radiol,2024,34(2): 899-913. DOI:10.1007/s00330-023-09990-6.
[8]	Liu Y, Fu Q, Peng X, et al. Attention-based deep multiple-instance learning for classifying circular RNA and other long non-coding RNA[J].Genes (Basel),2021,12(12): 2018. DOI:10.3390/genes12122018.
[9]	Moranguinho J, Pereira T, Ramos B, et al. Attention based deep multiple instance learning approach for lung cancer prediction using histopathological images[J].Annu Int Conf IEEE Eng Med Biol Soc,2021,2021: 2852-2855. DOI:10.1109/EMBC46164.2021.9631000. pmid:34891842
[10]	中华医学会, 中华医学会肿瘤学分会, 中华医学会杂志社. 中华医学会肺癌临床诊疗指南(2019版)[J].中华肿瘤杂志,2020,42(4): 257-287. DOI:10.3760/cma.j.cn112152-20200120-00049.
[11]	Winter KS, Hofmann FO, Thierfelder KM, et al. Towards volumetric thresholds in RECIST 1.1: therapeutic response assessment in hepatic metastases[J].Eur Radiol,2018,28(11): 4839-4848. DOI:10.1007/s00330-018-5424-0. pmid:29736851
[12]	Zhu Z, Chen M, Hu G, et al. A pre-treatment CT-based weighted radiomic approach combined with clinical characteristics to predict durable clinical benefits of immunotherapy in advanced lung cancer[J].Eur Radiol,2023,33(6): 3918-3930. DOI:10.1007/s00330-022-09337-7.
[13]	Long Y, Xiong Q, Song Q, et al. Immunotherapy plus chemotherapy showed superior clinical benefit to chemotherapy alone in advanced NSCLC patients after progression on osimertinib[J].Thorac Cancer,2022,13(3): 394-403. DOI:10.1111/1759-7714.14271.
[14]	杜希剑, 章凯敏, 陈斌, 等. CT影像组学对非小细胞肺癌免疫治疗疗效的预测价值[J].实用放射学杂志,2023,39(4): 548-551, 599. DOI:10.3969/j.issn.1002-1671.2023.04.008.
[15]	Bracci S, Dolciami M, Trobiani C, et al. Quantitative CT texture analysis in predicting PD-L1 expression in locally advanced or metastatic NSCLC patients[J].Radiol Med,2021,126(11): 1425-1433. DOI:10.1007/s11547-021-01399-9. pmid:34373989
[16]	Sun R, Limkin EJ, Vakalopoulou M, et al. A radiomics approach to assess tumour-infiltrating CD8 cells and response to anti-PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective multicohort study[J].Lancet Oncol,2018,19(9): 1180-1191. DOI:10.1016/S1470-2045(18)30413-3. pmid:30120041
[17]	Topalian SL, Hodi FS, Brahmer JR, et al. Safety, activity, and immune correlates of anti-PD-1 antibody in cancer[J].N Engl J Med,2012,366(26): 2443-2454. DOI:10.1056/NEJMoa1200690.
[18]	余镔.吸烟史影响免疫检查点抑制剂在晚期非小细胞肺癌中疗效的Meta分析[D]. 南昌: 南昌大学,2021. DOI:10.27232/d.cnki.gnchu.2021.000258.
[19]	Kinoshita T, Kudo-Saito C, Muramatsu R, et al. Determination of poor prognostic immune features of tumour microenvironment in non-smoking patients with lung adenocarcinoma[J].Eur J Cancer,2017,86: 15-27. DOI:10.1016/j.ejca.2017.08.026. pmid:28950145
[20]	Dall'Olio FG, Marabelle A, Caramella C. Tumour burden and efficacy of immune-checkpoint inhibitors[J].Nat Rev Clin Oncol,2022,19(2): 75-90. DOI:10.1038/s41571-021-00564-3.
[21]	Shen L, Fu H, Tao G, et al. Pre-immunotherapy contrast-enhanced CT texture-based classification: a useful approach to non-small cell lung cancer immunotherapy efficacy prediction[J].Front Oncol,2021,11: 591106. DOI:10.3389/fonc.2021.591106.
[22]	Ligero M, Garcia-Ruiz A, Viaplana C, et al. A CT-based radiomics signature is associated with response to immune checkpoint inhibitors in advanced solid tumors[J].Radiology,2021,299(1): 109-119. DOI:10.1148/radiol.2021200928. pmid:33497314