Abstract / Summary
Radiofrequency ablation is an established therapy for atrial fibrillation (AF), but outcomes are heterogeneous in patients with concomitant heart failure (HF). Therefore, we aimed to develop an explainable machine learning (ML) model based on cardiac CT angiography (CTA) for predicting functional improvement after ablation in patients with AF combined with HF. Patients at our institution were randomly divided into training and validation sets in the ratio of 7:3 and two other organizations were included as testing sets. Morphological features of the left atrium (LA) and pulmonary vein (PV) and radiomics features of the left atrial wall (LAW) were extracted, and Shape and Wall models were constructed using the ML algorithm, respectively. Shape_score and Wall_ score were calculated and integrated with important clinical factors to construct a combined predictive model (COMB). The predictive performance of the model was assessed by area under the receiver operating characteristic curve (AUC). Shapley additive explanations (SHAP) were used to interpret the contributions of individual features. 240, 101, and 75 patients were included in the training, validation, and testing sets, respectively. Gender, hyperlipidemia, urea, and AF type were retained in the final clinical model. Two shapes and three LAW radiomics features were screened to build Shape and Wall models, respectively. The developed COMB model presented good performance in predicting functional improvement after ablation in the three sets (AUC = 0.866, 0.803, and 0.845). SHAP analysis revealed the features with the greatest impact on the prediction outcomes in the Shape and Wall models. The model integrating cardiac CTA-derived morphological and radiomics features with clinical factors showed potential for predicting functional improvement after ablation in patients with AF and HF. Further prospective evaluation is required before clinical implementation.