AUTHOR=Chang Cheng , Sun Xiaoyan , Wang Gang , Yu Hong , Zhao Wenlu , Ge Yaqiong , Duan Shaofeng , Qian Xiaohua , Wang Rui , Lei Bei , Wang Lihua , Liu Liu , Ruan Maomei , Yan Hui , Liu Ciyi , Chen Jie , Xie Wenhui 

TITLE=A Machine Learning Model Based on PET/CT Radiomics and Clinical Characteristics Predicts ALK Rearrangement Status in Lung Adenocarcinoma

JOURNAL=Frontiers in Oncology

VOLUME=Volume 11 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2021.603882

DOI=10.3389/fonc.2021.603882

ISSN=2234-943X

ABSTRACT=Introduction: This study intends to non-invasively predict the ALK rearrangement status in lung adenocarcinoma by a developing machine learning model that combine PET/CT radiomic features and clinical characteristics. 
Methods: 526 patients of lung adenocarcinoma are enrolled to construct machine learning models, including 109 cases of positive and 417 cases of negative ALK rearrangements. The mRMR and LASSO logistic regression were used to select the most distinguishable features derived from PET/CT images. ROC (receiver operating characteristic) analysis were used to evaluate the performance of the models, and the performance of different models were compared by DeLong test. 
Results: 22 radiomic features were extracted from PET/CT images, and the majority of radiomic features used to develop this model were based on CT features (20 out of 22), only 2 PET features were included (PET percentile 10 and PET difference entropy). Three clinical features associated with ALK mutations (age, burr and pleural effusion) were also employed to construct the model. We found that integrated model developed by combining clinical characteristics and PET/CT radiomic features has significant advantage to predict the ALK mutation status in the training group (AUC = 0.87) and in the testing group (AUC = 0.88) compared with use clinical model alone in the training group (AUC = 0.76) and in the testing group (AUC = 0.74) respectively.
Conclusion: This study demonstrated that PET/CT radiomics-based machine learning model can be used a non-invasive diagnostic method to predict ALK mutations for lung adenocarcinoma patients in clinic.