Abstract
OBJECTIVE: This study aimed to develop and validate a deep learning radiomics (DLR) nomogram for individualized CHD risk assessment in the COPD population. METHODS: This retrospective study included 543 COPD patients from two different centers. Comprehensive clinical and imaging data were collected for all participants. In Center 1, 398 patients were randomly allocated into a training set and an internal validation set at a 7:3 ratio. An external test set was established using 145 patients from Center 2. Radiomics features were extracted from computed tomography (CT) images, and deep learning features were generated using ResNet50. By integrating traditional clinical data, radiomics features, and three-dimensional (3D) deep learning features, a combined predictive model was developed to estimate the risk of CHD in COPD patients. RESULTS: Validation cohort AUCs revealed the nomogram's optimal predictive performance (Internal: 0.800; External: 0.761) compared to clinical (0.759, 0.661), radiomics (0.752, 0.666), and DLR (0.767, 0.732) models. This integrative approach demonstrated a 9.1% and 13.4% relative AUC improvement over clinical and radiomics models in external validation. DCA corroborated these findings, showing the nomogram provides the highest net benefit for clinical decision-making across probability thresholds in COPD patients at risk for CHD. CONCLUSION: The nomogram model, which integrates clinical, radiomics, and deep learning features, exhibits promising performance in predicting CHD risk among COPD patients. It may offer valuable insights for early intervention and management strategies for CHD.