Left Ventricular Myocardial Dysfunction Evaluation in Thalassemia Patients Using Echocardiographic Radiomic Features and Machine Learning Algorithms

Abstract

Heart failure caused by iron deposits in the myocardium is the primary cause of mortality in beta-thalassemia major patients. Cardiac magnetic resonance imaging (CMRI) T2* is the primary screening technique used to detect myocardial iron overload, but inherently bears some limitations. In this study, we aimed to differentiate beta-thalassemia major patients with myocardial iron overload from those without myocardial iron overload (detected by T2*CMRI) based on radiomic features extracted from echocardiography images and machine learning (ML) in patients with normal left ventricular ejection fraction (LVEF > 55%) in echocardiography. Out of 91 cases, 44 patients with thalassemia major with normal LVEF (> 55%) and T2* ≤ 20 ms and 47 people with LVEF > 55% and T2* > 20 ms as the control group were included in the study. Radiomic features were extracted for each end-systolic (ES) and end-diastolic (ED) image. Then, three feature selection (FS) methods and six different classifiers were used. The models were evaluated using various metrics, including the area under the ROC curve (AUC), accuracy (ACC), sensitivity (SEN), and specificity (SPE). Maximum relevance-minimum redundancy-eXtreme gradient boosting (MRMR-XGB) (AUC = 0.73, ACC = 0.73, SPE = 0.73, SEN = 0.73), ANOVA-MLP (AUC = 0.69, ACC = 0.69, SPE = 0.56, SEN = 0.83), and recursive feature elimination-K-nearest neighbors (RFE-KNN) (AUC = 0.65, ACC = 0.65, SPE = 0.64, SEN = 0.65) were the best models in ED, ES, and ED&ES datasets. Using radiomic features extracted from echocardiographic images and ML, it is feasible to predict cardiac problems caused by iron overload.

Keywords: Cardiac magnetic resonance imaging; Echocardiography; Machine learning; Radiomics; Thalassemia.

Fig. 1

The framework of the current study

Fig. 2

Feature selection using the ANOVA method based on scoring

Fig. 3

Selected features using the MRMR method and their related scores

Fig. 4

Distribution of radiomic features selected by three feature selection methods including ANOVA, MRMR, and RFE in three different datasets (ED, ES, ED + ES)

Fig. 5

Performance of ML models in different settings including different datasets, feature selection methods, and classifiers. ACC: accuracy, SPE: specificity, SEN: sensitivity, KNN: K-nearest neighbors, LR: logistic regression, MLP: multi-layer perceptron, RF: random forest, SVM: support vector machine, XGB: eXtreme gradient boosting, ANOVA: analysis of variance, MRMR: maximum relevance-minimum redundancy, RFE: recursive feature elimination, ED: end-diastolic, ES: end-systolic

Fig. 6

Comparison of normal and thalassemia cases using a feature map approach in four different features to visualize voxel-wise radiomic feature. ED: end-diastole, ES: end-systole

Fig. 7

Model performance is compared using the DeLong test, which is run on the models’ AUCs. In this figure, the models on column and row were evaluated against each other. Green, if the row model considerably outperformed the column model in terms of P value; red, if the row model’s P value was much lower than the column model’s. If the comparison between the row model and column model yielded a non-significant P value, light blue would be the color