Abstract
These preliminary findings call for a longitudinal study to determine whether LV sphericity index and circumferential strain rate, or other combination of parameters, accurately predict post-surgical LVEF in PMR.
Introduction
Primary Mitral Regurgitation (PMR) occurs when the heart's mitral valve fails to close properly, leading to backward blood flow. To maintain cardiac output, the heart undergoes compensatory changes that often mask underlying myocardial damage. Current clinical guidelines rely heavily on Left Ventricular Ejection Fraction (LVEF) measured via echocardiography; patients with an LVEF >60% are typically considered low-risk for surgery.
However, a significant clinical challenge exists: approximately 20% of these "low-risk" patients experience a decline in LVEF to below 50% within six months post-surgery, indicating heart failure. This suggests that pre-surgical LVEF can be an inflated metric, failing to capture latent muscle dysfunction. This study aims to identify more sensitive markers using Cardiac MRI (CMR) and machine learning to predict high-risk patients.
Methodology
The study utilized high-resolution CMR with tissue tagging to measure "strain"—the physical deformation of the heart muscle—providing a more granular view of cardiac function than standard imaging.
The researchers compared traditional statistical approaches with advanced machine learning models to determine the most effective predictive framework. The modeling process involved:
- Statistical Methods: Logistic Regression with LASSO regularization.
- Machine Learning Methods: Random Forest (RF), Support Vector Machines (SVM), and Extreme Gradient Boosting (XGBoost).
- Feature Selection: Recursive Feature Elimination (RFE) was used to reduce 37 initial features to an optimal set.
- Interpretability: SHAP (Shapley Additive exPlanations) values were employed to explain the contribution of specific features to the model's risk predictions.
Datasets
The analysis was conducted on a single-center dataset including:
- 51 Pre-surgery PMR patients: Used for training and validation, with known six-month post-surgical outcomes.
- 49 Asymptomatic PMR patients: A group not yet meeting surgical criteria, used to test the model's predictive capabilities in early-stage disease.
- Healthy Controls: Age-matched individuals used to establish baseline CMR values.
Key Insights and Results
The "Big Four" Predictors
The most effective predictive model identified four key parameters: Baseline LVEF, LV End-Systolic Dimension (LVESD), LV Sphericity Index, and LV mid-systolic circumferential strain rate.
Model Performance
The Random Forest model demonstrated superior performance, achieving an accuracy of 86.17% and an AUROC of 0.846. This outperformed traditional Logistic Regression (77.92% accuracy), highlighting the importance of capturing non-linear interactions between cardiac shape and function.
Clinical Implications
- Sphericity and Strain: The LV Sphericity Index emerged as a potent predictor, confirming that the heart's transition from an elliptical to a spherical shape precedes functional decline. Furthermore, circumferential strain rate captured subtle contractile weaknesses missed by volume-based metrics.
- Asymptomatic Risk: When applied to the asymptomatic cohort, the model predicted that 28.6% of these patients would likely suffer post-surgical LVEF decline if they underwent surgery immediately. This suggests a need for earlier intervention strategies for specific subgroups.
Limitations and Future Directions
The study's primary limitation is its small, single-center sample size (n=51). While the results serve as a strong proof of concept, validation in larger, multi-center longitudinal trials is necessary. Future research should investigate whether intervening based on these "Big Four" markers improves long-term outcomes and explore the integration of CMR markers with circulating biomarkers for a comprehensive multi-omic approach.
Summary updated