Left Ventricular Thrombus in Ischemic Heart Failure: Machine-learning-based Prediction of Six-month Persistence and...

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning in Medicine

Background:

• Left ventricular thrombus (LVT) in ischemic heart failure poses a significant embolic risk.
• Current tools for predicting LVT persistence are limited, impacting clinical management.
• Guideline-concordant anticoagulation and serial echocardiography are standard care for LVT.

Purpose of the Study:

• To identify predictors of LVT non-regression and 1-year Major Adverse Cardiovascular Events (MACE).
• To evaluate the utility of machine learning models in predicting LVT outcomes.
• To inform clinical decision-making regarding imaging follow-up and anticoagulation intensity.

Main Methods:

• Studied 190 patients with imaging-confirmed LVT receiving anticoagulation.
• Employed serial echocardiography to assess LVT regression over 6 months.
• Combined classical statistics with explainable machine learning (CatBoost, SHAP) for outcome prediction.

Main Results:

• CatBoost model achieved good discrimination for LVT non-regression (CV-AUC 0.76, test accuracy 0.79).
• Key predictors identified by SHAP included left atrial diameter, pulmonary artery pressure, platelet count, and LV end-diastolic diameter.
• Thrombus size and CHA2DS2-VA score independently predicted 1-year MACE (logistic AUC 0.71).

Conclusions:

• Interpretable machine learning models can effectively stratify early risk of LVT persistence.
• Baseline cardiac remodeling and coagulability markers are crucial for risk assessment.
• These findings complement clinical decision-making for LVT management and follow-up strategies.