Ultrasound in medicine & biology · 2026
An Explainable Machine Learning Model Integrating Clinical Parameters for Identifying Left Atrial Appendage Thrombus in Non-valvular Atrial Fibrillation.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Zhou Jianhui, Zeng Linhui, Chen Lin, Zhang Youhong
Аннотация
OBJECTIVE: This study aimed to develop a predictive model for left atrial appendage thrombus (LAAT) in patients with non-valvular atrial fibrillation by integrating clinical and echocardiographic features using machine learning algorithms. The SHapley Additive exPlanations (SHAP) framework was applied to interpret the model and quantify the contribution of each predictor, thereby enhancing the accuracy and clinical interpretability of LAAT risk prediction.
METHODS: In this retrospective study, 325 patients with non-valvular atrial fibrillation who underwent transesophageal echocardiography at our institution between September 2022 and October 2024 were included. Least absolute shrinkage and selection operator regression with fivefold cross-validation was performed exclusively within the training cohort to select candidate features, followed by multivariable logistic regression to quantify independent associations. Six machine learning algorithms-logistic regression, support vector classifier, random forest, GradientBoosting, Extreme Gradient Boosting and Light Gradient Boosting Machine-were then trained and evaluated for model construction. The SHAP analysis was subsequently used to visualize and interpret the impact of each feature on model prediction.
RESULTS: Multivariable logistic regression identified lower LAA blood flow velocity (odds ratio, 0.805; p < 0.001) and persistent atrial fibrillation (odds ratio, 4.344; p = 0.005) as independent predictors of LAAT. Among the six models, logistic regression demonstrated the best predictive performance, with an area under the curve of 0.958 in the training cohort and 0.913 in the test cohort. SHAP interpretation revealed that LAA blood flow velocity, atrial fibrillation type, left atrial diameter, CHA₂DS₂-VASc score (congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, stroke/transient ischemic attack, vascular disease, age 65-74 years, sex category [female]), left ventricular ejection fraction and fibrinogen were the most influential features contributing to thrombus risk prediction.
CONCLUSION: By combining machine learning algorithms with the SHAP interpretability framework, this study successfully developed an efficient and explainable model for predicting LAAT in patients with non-valvular atrial fibrillation. The proposed model integrates readily available clinical and echocardiographic variables, providing a practical tool to assist in individualized anticoagulation management and clinical decision-making.
METHODS: In this retrospective study, 325 patients with non-valvular atrial fibrillation who underwent transesophageal echocardiography at our institution between September 2022 and October 2024 were included. Least absolute shrinkage and selection operator regression with fivefold cross-validation was performed exclusively within the training cohort to select candidate features, followed by multivariable logistic regression to quantify independent associations. Six machine learning algorithms-logistic regression, support vector classifier, random forest, GradientBoosting, Extreme Gradient Boosting and Light Gradient Boosting Machine-were then trained and evaluated for model construction. The SHAP analysis was subsequently used to visualize and interpret the impact of each feature on model prediction.
RESULTS: Multivariable logistic regression identified lower LAA blood flow velocity (odds ratio, 0.805; p < 0.001) and persistent atrial fibrillation (odds ratio, 4.344; p = 0.005) as independent predictors of LAAT. Among the six models, logistic regression demonstrated the best predictive performance, with an area under the curve of 0.958 in the training cohort and 0.913 in the test cohort. SHAP interpretation revealed that LAA blood flow velocity, atrial fibrillation type, left atrial diameter, CHA₂DS₂-VASc score (congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, stroke/transient ischemic attack, vascular disease, age 65-74 years, sex category [female]), left ventricular ejection fraction and fibrinogen were the most influential features contributing to thrombus risk prediction.
CONCLUSION: By combining machine learning algorithms with the SHAP interpretability framework, this study successfully developed an efficient and explainable model for predicting LAAT in patients with non-valvular atrial fibrillation. The proposed model integrates readily available clinical and echocardiographic variables, providing a practical tool to assist in individualized anticoagulation management and clinical decision-making.