Radiomic virtual biopsy for phenotyping of abnormal cardiac masses based on computed tomography: proof-of-concept study.

INTRODUCTION

Cardiac masses can present a diagnostic challenge on cardiac imaging. Differentiating between tumors, such as myxomas, and non‑neoplastic lesions, such as thrombi, is essential for appropriate management and treatment planning.

OBJECTIVES

The aim of this study was to explore the feasibility of radiomics‑based "virtual biopsy" in distinguishing between different types of cardiac masses on computed tomography (CT).

PATIENTS AND

METHODS

A total of 89 patients were included in the analysis, of which 42 (47.2%) were diagnosed with myxoma, 37 (41.6%) with thrombus, and 10 (11.2%) had other diagnoses (fibroelastoma in 6 cases [6.7%], sarcoma in 2 cases [2.2%], and angioma and metastasis in 1 case [1.1%] each). Region of interest (ROI) selection and calculation of radiomic features were performed within the syngo.via Frontier Radiomics platform. Three‑dimensional ROI of the cardiac masses for CT examination was manually delineated. A multivariate least absolute shrinkage and selection operator-based model was developed for both binary (thrombus vs tumor) and multiclass (thrombus vs myxoma vs other) classifications.

RESULTS

The exploratory multiclass radiomics model demonstrated good preliminary discriminatory performance, with validation areas under the curve ranging from 0.771 to 0.933; however, class‑specific estimates, particularly for the heterogeneous "other" category, should be interpreted with caution due to the small validation cohort. These findings suggest that radiomic features can capture subtle differences between distinct cardiac mass types, enabling noninvasive tissue characterization.

CONCLUSIONS

This proof‑of‑concept study provides the first evidence that a multiclass CT‑based radiomics model can differentiate between myxomas, thrombi, and other cardiac masses. Radiomics‑based virtual biopsy may potentially complement conventional imaging; however, these exploratory findings require validation in larger, multicenter cohorts before clinical application.