Development and validation of a novel echocardiography-based nomogram for the streamlined classification of cardiac tumors in cancer patients.

Abstract

Background: Differentiating cardiac tumors is crucial for treatment planning, but the specificity of echocardiography as a first-line screening tool is limited. This study aimed to develop a streamlined classification model for cardiac tumors in cancer patients using echocardiographic data.

Methods: A total of 215 echocardiographic clips representing cardiac tumors from 121 patients with extracardiac malignancies were selected and divided into training and testing cohorts. The cardiac neoplasms were classified as benign or malignant based on substantial evidence. Radiomics features were extracted utilizing PyRadiomics, and a radiomics score (Rad-score) was subsequently computed through an optimized machine learning (ML) framework tailored for tumor classification. Non-experience-dependent indicators (NDIs) derived from baseline and echocardiographic assessments were ascertained and integrated with the Rad-score to construct a classification model. A composite nomogram was developed, and its predictive accuracy was benchmarked against that of junior and senior physicians using receiver operating characteristic (ROC) curves and decision curve analysis (DCA).

Results: Significant differences in the Rad-scores and four NDIs [age, tumor location, and long and short diameters (SDs)] (all P<0.05) distinguished benign from malignant tumors. Patients with malignant cardiac tumors were more likely to be younger, for the tumor to be in the right cardiac circulatory system, be larger in size, and have a lower Rad-score. Among these indicators, the Rad-score, tumor location, and SD were shown to be independent predictors of malignancy. The integrated model demonstrated strong classification capability [area under the curve (AUC): 0.873; 95% confidence interval (CI): 0.820-0.914], which was substantiated in the test cohort (AUC: 0.861; 95% CI: 0.807-0.904). The classification performance of the generated nomogram was comparable to that of the senior doctor (AUC: 0.867 vs. 0.873, DeLong P=0.928) and surpassed that of the junior doctor (AUC: 0.867 vs. 0.669, DeLong P=0.029). DCA indicated that the nomogram was superior to the junior physician for classification tasks.

Conclusions: This study developed a nomogram that involved radiomics and objective indicators based on echocardiography to effectively distinguish between malignant and benign cardiac tumors, thereby improving classification practices and decision-making in diverse clinical settings.