CT-based radiomics signature for differentiation between cardiac tumors and thrombi: a retrospective, multicenter study

This study aimed to validate the effectiveness of CT radiomics analysis in differentiating between cardiac tumors and thrombi. Using a retrospective dataset of 192 patients with cardiac masses, three models were developed: a radiomics model, a clinical model (including clinical and conventional CT variables), and a combined model. Feature extraction was performed using AVIEW software (Coreline Soft) based on the Pyradiomics library. The radiomics and combined models showed significantly better differentiation performance than the clinical model in the training dataset (AUC 0.973 and 0.983 vs. 0.870, p < 0.001). In an external validation dataset of 63 patients, the combined model also outperformed the clinical model (AUC 0.911 vs. 0.802, p = 0.037). The study concluded that CT radiomics analysis, particularly when combined with clinical and conventional CT features, is highly effective in distinguishing between cardiac tumors and thrombi.