Solving Inverse Electrocardiographic Mapping Using Machine Learning and Deep Learning Frameworks.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Medical Imaging

Background:

• Electrocardiographic imaging (ECGi) reconstructs cardiac electrograms from body surface potentials, addressing the inverse problem of electrocardiography.
• Current solutions for the ECGi inverse problem require improvement for enhanced accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate machine learning and deep learning frameworks for solving the inverse problem of ECGi.
• To improve the accuracy of reconstructing electrograms at the heart's surface using body surface potential data.

Main Methods:

• Simultaneous electrocardiogram recordings from pig ventricles and body surfaces.
• Development and application of Fully Connected Neural Network (FCN), Long Short-term Memory (LSTM), and Convolutional Neural Network (CNN) models.
• Implementation of a data alignment method for inter-pig variability and leave-one-out cross-validation for evaluation.

Main Results:

• The developed neural network models demonstrated effectiveness in solving the ECGi inverse problem.
• The best model achieved a median correlation coefficient of 0.74 for the predicted ECG wave.
• Successful reconstruction of electrograms with accuracy compatible with current standard methods was achieved.

Conclusions:

• Neural networks, including deep learning architectures, can effectively solve the inverse problem of ECGi.
• This approach is viable even with relatively small datasets, offering accuracy comparable to established methods.
• Machine learning provides a promising avenue for advancing ECGi technology.