Ejection Fraction Estimation from Echocardiograms Using Optimal Left Ventricle Feature Extraction Based on Clinical...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Echocardiography is crucial for assessing heart function, with Left Ventricle Ejection Fraction (LVEF) being a key metric.
• Quantifying LVEF involves significant inter-observer and intra-observer variability, impacting diagnostic consistency.
• Machine learning (ML) offers potential to analyze complex echocardiogram data and improve diagnostic accuracy.

Purpose of the Study:

• To develop and evaluate ML algorithms for automated LVEF estimation from echocardiogram data.
• To compare the accuracy of ML-based LVEF quantification against traditional methods like Simpson's method.
• To identify optimal ML models for robust LV segmentation and LVEF regression.

Main Methods:

• Left Ventricle (LV) segmentation was performed on echocardiogram data using DeepLab, a convolutional neural network.
• Clinical features were extracted from the segmented LV.
• Extracted features were analyzed using neural networks and traditional ML algorithms, including Long Short-Term Memory Networks (LSTM), for LVEF regression.

Main Results:

• ML techniques demonstrated higher accuracy in LVEF estimation compared to Simpson's method.
• The combination of DeepLab for segmentation and LSTM for regression achieved a Dice Similarity Coefficient of 0.92.
• This combined approach yielded a mean absolute error of 5.736% for LVEF estimation.

Conclusions:

• ML-based approaches significantly enhance the accuracy and reduce variability in LVEF quantification.
• DeepLab and LSTM networks provide a powerful framework for computer-assisted cardiac diagnostics using echocardiography.
• This study highlights the potential of AI in improving the reliability of echocardiogram-based cardiac assessments.