Contents lists available at ScienceDirect Lung Cancer journal homepage: www.elsevier.com/locate/lungcan Extraction of radiomic values from lung adenocarcinoma with near-pure subtypes in the International Association for the Study of Lung Cancer/the American Thoracic Society/the European Respiratory Society (IASLC/ATS/ ERS) classification Shun-Mao Yang a,b,1 , Li-Wei Chen a,1 , Hao-Jen Wang a , Leng-Rong Chen a , Kuo-Lung Lor a , Yi-Chang Chen a,d , Mong-Wei Lin c , Min-Shu Hsieh e , Jin-Shing Chen c , Yeun-Chung Chang d, ⁎⁎ , Chung-Ming Chen a, ⁎ a Institute of Biomedical Engineering, College of Medicine and College of Engineering, National Taiwan University, Taiwan b Department of Surgery, National Taiwan University Hospital, Hsin-Chu Branch, Taiwan c Department of Surgery, National Taiwan University Hospital and National Taiwan University College of Medicine, Taiwan d Department of Medical Imaging, National Taiwan University Hospital and National Taiwan University College of Medicine, Taiwan e Department of Pathology, National Taiwan University Hospital and National Taiwan University College of Medicine, Taiwan ARTICLE INFO Keywords: Lung neoplasms Computed tomography Pathological stratification ABSTRACT Introduction: Histological subtypes of lung adenocarcinomas (ADCs) classified by the International Association for the Study of Lung Cancer/American Thoracic Society/European Respiratory Society (IASLC/ATS/ERS) system have been investigated using radiomic approaches. However, the results have had limitations since < 80% of invasive lung ADCs were heterogeneous, with two or more subtypes. To reduce the influence of het- erogeneity during radiomic analysis, computed tomography (CT) images of lung ADCs with near-pure ADC subtypes were analyzed to extract representative radiomic features of different subtypes. Methods: We enrolled 95 patients who underwent complete resection for lung ADC and a pathological diagnosis of a “near-pure” (≥70%) IASLC/ATS/ERS histological subtype. Conventional histogram/morphological features and complex radiomic features (grey-level-based statistical features and component variance-based features) of thin-cut CT data of tumor regions were analyzed. A prediction model based on leave-one-out cross-validation (LOOCV) and logistic regression (LR) was used to classify all five subtypes and three pathologic grades (lepidic, acinar/papillary, micropapillary/solid) of ADCs. The validation was performed using 36 near-pure ADCs in a later cohort. Results: A total of 31 lepidic, 14 papillary, 32 acinar, 10 micropapillary, and 8 solid ADCs were analyzed. With 21 conventional and complex radiomic features, for 5 subtypes and 3 pathological grades, the prediction models achieved accuracy rates of 84.2% (80/95) and 91.6% (87/95), respectively, while accuracy was 71.6% and 85.3%, respectively, if only conventional features were used. The accuracy rate for the validation set (n = 36) was 83.3% (30/36) and 94.4% (34/36) in 5 subtypes and 3 pathological grades, respectively, using conventional and complex features, while it was 66.7% and 77.8% only using conventional features, respectively. Conclusion: Lung ADC with high purity pathological subtypes demonstrates strong stratification of radiomic values, which provide basic information for accurate pathological subtyping and image parcellation of tumor sub-regions. https://doi.org/10.1016/j.lungcan.2018.03.004 Received 11 November 2017; Received in revised form 20 February 2018; Accepted 6 March 2018 ⁎ Corresponding author at: Institute of Biomedical Engineering, College of Medicine and College of Engineering, National Taiwan University, 1, Section 1, Jen-Ai Road, Taipei 100, Taiwan. ⁎⁎ Corresponding author at: Department of Medical Imaging, National Taiwan University Hospital, 7 Chung-Shan South Road, Taipei, Taiwan. 1 These authors contributed equally to this work. E-mail addresses: ycc5566@ntu.edu.tw (Y.-C. Chang), chung@ntu.edu.tw (C.-M. Chen). Lung Cancer 119 (2018) 56–63 0169-5002/ © 2018 Elsevier B.V. All rights reserved. T