Ultrasonic texture features for assessing abnormal interventricular septum motion following cardiac surgery

Background: Cardiac surgery alters cardiomyocyte geometry and architecture, leading to changes in the acoustic properties of the myocardium. This study investigated ultrasomics—a novel cardiac ultrasound-based radiomics technique that extracts high-throughput pixel-level information from images—to identify ultrasonic texture and morphological changes associated with abnormal motion of the interventricular septum (IVS) following cardiac surgery. Methods: A total of 650 patients from two centers were included in the study. The 460 patients from Center A were randomly allocated to the training (n=322) and internal validation (n=138) sets at a 7:3 ratio, while the 190 patients from Center B constituted the independent external validation set. All patients underwent transesophageal echocardiography (TEE) at two time points: 10 minutes after anesthesia induction (preoperative) and 10 minutes after chest closure (postoperative). Standard left ventricular long-axis and four-chamber views were acquired to visualize the IVS. The workflow consisted of IVS segmentation, radiomics feature extraction, and feature selection using the least absolute shrinkage and selection operator (LASSO) algorithm. Three machine learning methods were compared: logistic regression (LR), random forest (RF), and support vector machine (SVM). Model interpretability was evaluated using SHapley Additive exPlanations (SHAP) analysis. Results: Twenty-six significant radiomics features, including shape, first-order, and texture features, were identified. The incidence of postoperative abnormal IVS motion was 50.93%, 46.38%, and 40.00% in the training, internal validation and external validation sets, respectively, with no significant intergroup difference (P>0.05). The SVM model demonstrated the best diagnostic performance, with an area under the curve (AUC) of 0.943 [95% confidence interval (CI): 0.925–0.961] in the training set, 0.948 (95% CI: 0.921–0.974) in the internal validation set, and 0.803 (95% CI: 0.772–0.834) in the external validation set. Postoperative analysis revealed significant changes in the core IVS radiomics features. Specifically, increased gray-scale heterogeneity and decreased texture correlation in the IVS were observed in patients with abnormal septal motion after cardiac surgery (all relevant P<0.001). Conclusions: This study demonstrates the potential of cardiac ultrasomics to differentiate between normal and abnormal IVS motion in patients after cardiac surgery. A highly interpretable SVM ultrasomics model was developed based on texture and morphological features, providing quantitative characterization of microstructural changes in the IVS associated with abnormal motion following cardiac surgery and potentially supporting more precise perioperative clinical decision-making.