Precision diagnostics in cardiac tumours: Integrating echocardiography and pathology with advanced machine learning on limited data

Abstract

This study pioneers the integration of echocardiography and pathology data with advanced machine learning (ML) techniques to significantly enhance the diagnostic accuracy of cardiac tumours, a critical yet challenging aspect of cardiology. Despite advancements in diagnostic methods, cardiac tumours' nuanced complexity and rarity necessitate more precise, non-invasive, and efficient diagnostic solutions. Our research aims to bridge this gap by developing and validating ML models—Support Vector Machines (SVM), Random Forest (RF), and Gradient Boosting Machines (GBM)—optimized for limited datasets prevalent in specialized medical fields. Utilizing a dataset comprising clinical features from 399 patients at the Heart Hospital, our study meticulously evaluated the performance of these models against traditional diagnostic metrics. The RF model emerged superior, achieving a groundbreaking accuracy of 96.25 % and a perfect ROC AUC score of 0.99, significantly outperforming existing diagnostic approaches. Key predictors identified include age, echo malignancy, and echo position, underscoring the value of integrating diverse data types. Clinical validation conducted at the Heart Hospital further confirmed the models' applicability and reliability, with the RF model demonstrating a diagnostic accuracy of 94 % in a real-world setting. These findings advocate for the potential of ML in revolutionizing cardiac tumour diagnostics, offering pathways to more accurate, non-invasive, and patient-centric diagnostic processes. This research not only highlights the capabilities of ML to enhance diagnostic precision in the realm of cardiac tumours but also sets a foundation for future explorations into its broader applicability across various domains of medical diagnostics, emphasizing the need for expanded datasets and external validation.