Interpretable Deep Learning with Multi-Scale CT for Predicting Occult Lymph Node Metastasis in Early-Stage NSCLC: A Multicenter Study.
Authors
Affiliations (11)
Affiliations (11)
- College of Medical Informatics, Chongqing Medical University, Chongqing, 400016, China.
- Department of Radiology, The Third Affiliated Hospital of Chongqing Medical University, Chongqing, China.
- The Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital & Shenzhen Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Shenzhen, 518116, China.
- Affiliated Cancer Hospital, School of Medicine, Southern University of Science and Technology, Shenzhen, 518055, China.
- Department of Radiology, The First Affiliated Hospital of Chongqing Medical University, Chongqing, China.
- Department of Oncology, The First Affiliated Hospital of Chongqing Medical University, Chongqing, China.
- Center for Big Data and Intelligent Medicine, The First Affiliated Hospital of Chongqing Medical University, Chongqing, China. [email protected].
- Key Laboratory of Digital Health and Intelligent Medicine, Chongqing Municipal Health Commission, Chongqing, China. [email protected].
- Chongqing Translational Medicine Center, Chongqing, China. [email protected].
- College of Medical Informatics, Chongqing Medical University, Chongqing, 400016, China. [email protected].
- Department of Oncology, The First Affiliated Hospital of Chongqing Medical University, Chongqing, China. [email protected].
Abstract
Accurate preoperative prediction of occult lymph node metastasis (OLNM) in early-stage non-small cell lung cancer (NSCLC) is crucial for treatment planning. This study aimed to develop and validate a CT-based three-dimensional (3D) deep learning model to predict OLNM. In this retrospective, multicenter study, 900 patients from two hospitals were included and divided into a primary cohort (n = 500) for model development and an external test cohort (n = 400) for independent validation. We proposed a 3D EfficientNet model, and its diagnostic performance was compared against other benchmark deep learning architectures and four experienced radiologists. The proposed model achieved area under the receiver operating characteristic curve (AUC) of 0.8907 (95% CI 0.7878-0.9691) in the internal test set and 0.8721 (95% CI 0.8200-0.9170) in the external test cohort. This performance was superior to that of the other convolutional neural networks and the radiologists (all P < 0.05). Furthermore, interpretability analysis using Grad-CAM indicated that the model's predictions were based on distinct attention patterns. In conclusion, our 3D EfficientNet model demonstrates significant potential as a non-invasive and accurate tool for predicting OLNM in early-stage NSCLC. It can effectively support clinicians in making more precise staging and treatment decisions.