Abstract
This study aims to predict bone metastasis in lung cancer patients using radiomics and deep learning. Early prediction of bone metastasis is crucial for timely intervention and personalized treatment plans. This can improve patient outcomes and quality of life. By integrating advanced imaging techniques with artificial intelligence, this study seeks to enhance predictive accuracy and clinical decision-making. METHODS: We included 189 lung cancer patients, comprising 89 with non-bone metastasis and 100 with confirmed bone metastasis. Radiomic features were extracted from CT images, and feature selection was performed using Minimum Redundancy Maximum Relevance (mRMR) and Least Absolute Shrinkage and Selection Operator (LASSO). We developed and validated a radiomics model and a deep learning model using DenseNet-264. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. Statistical comparisons were made using the DeLong test. RESULTS: The radiomics model achieved an AUC of 0.815 on the training set and 0.778 on the validation set. The DenseNet-264 model demonstrated superior performance with an AUC of 0.990 on the training set and 0.971 on the validation set. The DeLong test confirmed that the AUC of the DenseNet-264 model was significantly higher than that of the radiomics model (p < 0.05). CONCLUSIONS: The DenseNet-264 model significantly outperforms the radiomics model in predicting bone metastasis in lung cancer patients. The early and accurate prediction provided by the deep learning model can facilitate timely interventions and personalized treatment planning, potentially improving patient outcomes. Future studies should focus on validating these findings in larger, multi-center cohorts and integrating clinical data to further enhance predictive accuracy.