Left Ventricular Myocardial Dysfunction Evaluation in Thalassemia Patients Using Echocardiographic Radiomic Features...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Heart failure from myocardial iron deposits is a major cause of mortality in beta-thalassemia major.
• Cardiac magnetic resonance imaging T2* is standard for detecting myocardial iron overload but has limitations.
• Early detection of cardiac involvement is crucial for managing beta-thalassemia major.

Purpose of the Study:

• To differentiate beta-thalassemia major patients with and without myocardial iron overload using echocardiography-derived radiomic features and machine learning.
• To assess the feasibility of using machine learning models for predicting cardiac iron overload in patients with normal left ventricular ejection fraction (LVEF).

Main Methods:

• Radiomic features were extracted from end-systolic (ES) and end-diastolic (ED) echocardiography images of 44 patients with iron overload (T2* ≤ 20 ms) and 47 controls (T2* > 20 ms), all with LVEF > 55%.
• Three feature selection methods (MRMR-XGB, ANOVA-MLP, RFE-KNN) and six classifiers were employed.
• Model performance was evaluated using AUC, ACC, SEN, and SPE.

Main Results:

• The MRMR-XGB model achieved the highest performance on the ED dataset with AUC=0.73, ACC=0.73, SPE=0.73, and SEN=0.73.
• ANOVA-MLP and RFE-KNN also demonstrated promising results on ES and combined ED&ES datasets, respectively.
• These findings indicate the potential of radiomic analysis in identifying myocardial iron overload.

Conclusions:

• Radiomic features extracted from echocardiography, combined with machine learning, provide a feasible method for predicting cardiac complications related to iron overload in beta-thalassemia major patients.
• This non-invasive approach could complement or serve as an alternative to traditional imaging techniques for monitoring myocardial iron levels.