China Oncology ›› 2019, Vol. 29 ›› Issue (10): 815-823.doi: 10.19401/j.cnki.1007-3639.2019.10.009

Previous Articles     Next Articles

Application of radiomics signature captured from CT to predicting brain metastases in stage Ⅲ/Ⅳ anaplastic lymphoma kinase-positive non-small cell lung cancer patients

XU Xinyan, CHEN Jiayan, HUANG Lü, WANG Jiazhou, LIU Di , WEN Junmiao, CAO Jianzhao, FAN Min   

  1. Department of Radiation Oncology, Fudan University Shanghai Cancer Center; Department of Oncology, Shanghai Medical College, Fudan University, Shanghai 200032, China
  • Online:2019-10-30 Published:2019-11-01
  • Contact: FAN Min E-mail: fanming@fudan.edu.cn

Abstract: Background and purpose: The probability of brain metastasis (BM) in anaplastic lymphoma kinase (ALK)-positive non-small cell lung cancer (NSCLC) patients is 20%-35%. Brain is also one of the main metastatic sites in patients with acquired resistance to targeted therapy. Therefore, early prediction of brain metastases in such patients is important for the assessment of overall prognosis and efficacy of targeted therapy. Radiomics demonstrates the idea of high-throughput extraction of quantitative features from images to macroscopically describe tumor phenotype and heterogeneity. The purpose of this study was to develop a radiomics approach to predict pre-treatment brain metastasis for stage Ⅲ/Ⅳ ALK-positive NSCLC patients. Methods: Patients with pathologically confirmed ALK-positive Ⅲ/Ⅳ NSCLC from Apr. 2014 to Sep. 2017 in Fudan University Shanghai Cancer Center were enrolled retrospectively. Their pretreatment thoracic CT images were collected, and the gross tumor volume (GTV) was defined by two experienced radiation oncologists. Patients were divided into training set and test set (4∶1). A test-retest in RIDER NSCLC dataset was performed to identify stable radiomics features. The least absolute shrinkage and selection operator (LASSO) COX regression and a leave-one-out cross-validation were conducted to identify optimal features for the logistic regression analyses to evaluate the predictive value of radiomics and clinical features for pretreatment BM. Results: In total, 132 patients were included, among which 27 patients had pretreatment BM. In the training set, one radiomics feature (W_GLCM_LH_Correlation) was significantly correlated with BM [P value=0.014, area under curve (AUC)=0.687]. It also exhibited moderate performance in the test set (AUC=0.642). Combining radiomics feature with N stage would further enhance predictive power (train set: AUC=0.697; test set: AUC=0.675). Conclusion: We identified one radiomics feature derived from pretreatment thoracic CT that was predictive of pretreatment BM in stage Ⅲ/Ⅳ ALK-positive NSCLC patients, which could be beneficial to risk stratification for such patients. Further investigation is necessary.