FDG-PET/CT data enhances machine learning prediction of occult lymph-node metastasis in lung adenocarcinoma.
Authors
Affiliations (3)
Affiliations (3)
- Department of Thoracic Surgery, Hokkaido University Graduate School of Medicine, Sapporo, Japan.
- Department of Diagnostic Imaging, Hokkaido University Graduate School of Medicine, Sapporo, Japan.
- Faculty of Information Science and Technology, Hokkaido University, Sapporo, Japan.
Abstract
Occult lymph-node metastasis (ONM) occurs in 10-20% of primary lung cancers and can influence surgical and treatment strategies. Improving preoperative prediction of ONM may optimize lymph-node dissection and multidisciplinary planning. We aimed to develop machine learning (ML) models to predict ONM using clinical data and to assess whether adding fluorodeoxyglucose positron emission tomography (FDG-PET)/computed tomography (CT) data, including individual lymph-node maximum standardized uptake value (SUVmax), could enhance predictive performance. We retrospectively analyzed 77 patients who underwent curative lobectomy for primary lung adenocarcinoma at our institution between 2016 and 2018. Four ML models-Random Forest (RF), Support Vector Machine (SVM), Gradient Boosting Machine (GBM), and eXtreme Gradient Boosting (XGB)-were developed using clinical data, PET/CT data, and their combination. Feature selection was performed using the Minimum Redundancy Maximum Relevance (mRMR) algorithm to reduce feature redundancy. The dataset was divided into training and testing sets in an 8:2 ratio with no overlapping. Hyperparameters were tuned using five-fold cross-validation. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and average precision (AP), with confidence intervals estimated via bootstrap resampling. Feature importance was assessed using SHapley Additive exPlanations (SHAP). Using clinical data alone, AUCs were 0.88 (RF), 0.69 (SVM), 0.73 (GBM), and 0.82 (XGB). Incorporation of PET/CT features improved predictive performance, yielding AUCs of 0.91 (SVM), 0.87 (GBM), and 0.91 (XGB). SHAP analysis demonstrated that while primary tumor PET features were important, the SUVmax of individual lymph nodes emerged as additional key predictors, highlighting the critical contribution of nodal PET data to ONM prediction. Incorporating PET/CT data, including individual lymph node SUVmax, improved ONM prediction in ML models. These findings highlight the critical contribution of nodal PET data and support the integration of imaging data into ML tools to guide personalized surgical and treatment decisions in lung adenocarcinoma.