Electromechanical Wave Imaging With Machine Learning for Automated Isochrone Generation.

Area of Science:

• Cardiovascular Imaging
• Biomedical Engineering
• Computational Cardiology

Background:

• Standard electromechanical wave imaging (EWI) isochrone generation requires manual selection of zero-crossing (ZC) points.
• This manual process is time-consuming and prone to inter-observer variability and operator bias, especially in large patient cohorts.

Purpose of the Study:

• To develop and optimize an automated ZC selection algorithm for faster and more robust EWI isochrone generation.
• To compare the performance of heuristic-based and machine learning approaches for automated ZC selection.

Main Methods:

• Development of an automated ZC selection algorithm utilizing heuristic-based baselines and machine learning classifiers (logistic regression, SVM, Random Forest).
• Training and validation using manually generated isochrones, previously validated against 3D intracardiac mapping.
• Evaluation of classifier performance in identifying cardiac conditions and pacing locations.

Main Results:

• The Random Forest classifier achieved the highest precision (up to 97%) using a voting approach for ZC candidate selection.
• Machine learning models successfully identified accessory pathways in Wolff-Parkinson-White patients and localized pacing electrodes in canines.
• The Random Forest classifier demonstrated consistent predictivity across different datasets, outperforming SVM.

Conclusions:

• An automated machine learning approach, specifically Random Forest, can significantly reduce user variability and processing time for EWI isochrone generation.
• The developed algorithm preserves the accuracy of cardiac activation patterns.
• This automated method offers a more efficient and reliable alternative to manual ZC selection in clinical practice.