Assessing the Role of Radiomics in Classifying Pericardial Effusion: Insights from a Proof-of-Concept Study

Abstract

Introduction: Pericardial effusion presents diverse clinical challenges, ranging from asymptomatic findings to life-threatening conditions. Determining the etiology of a pericardial effusion is often done empirically but may be assisted by invasive pericardiocentesis, which is the only technique that provides direct access to pericardial fluid. Cardiac computed tomography (CCT) is useful in imaging pericardial effusion, as it offers its detailed visualization. Through radiomic analysis, qualitative CCT images can be transformed into quantitative data, enabling the extraction of features that describe tissue characteristics. Objectives: This proof-of-concept study evaluated the potential of radiomic analysis of cardiac computed tomography (CCT) images in distinguishing between post-procedural and non-post-procedural pericardial effusions. Methods: A retrospective cohort of 109 patients was analyzed, with radiomic features extracted using PyRadiomics and statistical models developed through LASSO logistic regression. Results: Of the 851 radiomic features initially extracted, six were identified as the most predictive, encompassing first-order and texture-based attributes. The model demonstrated moderate discriminatory capacity (AUC: 0.696) in differentiating effusion types. While classification performance was limited by data imbalance and the absence of invasive laboratory confirmation, the findings highlight the potential of radiomics in complementing traditional diagnostic approaches. Conclusion: This study highlights the viability of applying radiomics to analyze pericardial fluid characteristics and encourages further research focusing on certain effusion subtypes, such as exudative or malignancy-related effusions. By providing a new perspective on the assessment of pericardial effusions, these discoveries may enhance non-invasive diagnostic techniques.