Liwen Zhang1, Di Dong1, Wenjuan Zhang2, Xiaohan Hao3, Mengjie Fang1, Shuo Wang4, Wuchao Li5, Zaiyi Liu6, Rongpin Wang7, Junlin Zhou8, Jie Tian9. 1. School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China; CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing, China. 2. Department of Radiology, Lanzhou University Second Hospital, Lanzhou, China; Key Laboratory of Medical Imaging of Gansu Province, Lanzhou, China. 3. CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing, China. 4. CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing, China; Beijing Advanced Innovation Center for Big Data-Based Precision Medicine, School of Medicine, Beihang University, Beijing, China. 5. Department of Radiology, Guizhou Provincial People's Hospital, Guiyang, China. 6. Department of Radiology, Guangdong General Hospital, Guangzhou, China. Electronic address: zyliu@163.com. 7. Department of Radiology, Guizhou Provincial People's Hospital, Guiyang, China. Electronic address: wangrongpin@126.com. 8. Department of Radiology, Lanzhou University Second Hospital, Lanzhou, China; Key Laboratory of Medical Imaging of Gansu Province, Lanzhou, China. Electronic address: ery_zhoujl@lzu.edu.cn. 9. CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing, China; Beijing Advanced Innovation Center for Big Data-Based Precision Medicine, School of Medicine, Beihang University, Beijing, China; Key Laboratory of Big Data-Based Precision Medicine (Beihang University), Ministry of Industry and Information Technology, Beijing, China; Engineering Research Center of Molecular and Neuro Imaging of Ministry of Education, School of Life Science and Technology, Xidian University, Xi'an, China. Electronic address: jie.tian@ia.ac.cn.
Abstract
BACKGROUND AND PURPOSE: Risk prediction of overall survival (OS) is crucial for gastric cancer (GC) patients to assess the treatment programs and may guide personalized medicine. A novel deep learning (DL) model was proposed to predict the risk for OS based on computed tomography (CT) images. MATERIALS AND METHODS: We retrospectively collected 640 patients from three independent centers, which were divided into a training cohort (center 1 and center 2, n = 518) and an external validation cohort (center 3, n = 122). We developed a DL model based on the architecture of residual convolutional neural network. We augmented the size of training dataset by image transformations to avoid overfitting. We also developed radiomics and clinical models for comparison. The performance of the three models were comprehensively assessed. RESULTS: Totally 518 patients were prepared by data augmentation and fed into DL model. The trained DL model significantly classified patients into high-risk and low-risk groups in training cohort (P-value <0.001, concordance index (C-index): 0.82, hazard ratio (HR): 9.79) and external validation cohort (P-value <0.001, C-index: 0.78, HR: 11.76). Radiomics model was developed with selected 24 features and clinical model was developed with three significant clinical variables (P-value <0.05). The comparison illustrated DL model had the best performance for risk prediction of OS according to the C-index (training: DL vs Clinical vs Radiomics = 0.82 vs 0.73 vs 0.66; external validation: 0.78 vs 0.71 vs 0.72). CONCLUSION: The DL model is a powerful model for risk assessment, and potentially serves as an individualized recommender for decision-making in GC patients.
BACKGROUND AND PURPOSE: Risk prediction of overall survival (OS) is crucial for gastric cancer (GC) patients to assess the treatment programs and may guide personalized medicine. A novel deep learning (DL) model was proposed to predict the risk for OS based on computed tomography (CT) images. MATERIALS AND METHODS: We retrospectively collected 640 patients from three independent centers, which were divided into a training cohort (center 1 and center 2, n = 518) and an external validation cohort (center 3, n = 122). We developed a DL model based on the architecture of residual convolutional neural network. We augmented the size of training dataset by image transformations to avoid overfitting. We also developed radiomics and clinical models for comparison. The performance of the three models were comprehensively assessed. RESULTS: Totally 518 patients were prepared by data augmentation and fed into DL model. The trained DL model significantly classified patients into high-risk and low-risk groups in training cohort (P-value <0.001, concordance index (C-index): 0.82, hazard ratio (HR): 9.79) and external validation cohort (P-value <0.001, C-index: 0.78, HR: 11.76). Radiomics model was developed with selected 24 features and clinical model was developed with three significant clinical variables (P-value <0.05). The comparison illustrated DL model had the best performance for risk prediction of OS according to the C-index (training: DL vs Clinical vs Radiomics = 0.82 vs 0.73 vs 0.66; external validation: 0.78 vs 0.71 vs 0.72). CONCLUSION: The DL model is a powerful model for risk assessment, and potentially serves as an individualized recommender for decision-making in GC patients.