Automatic Detection of Aortic Valve Events Using Deep Neural Networks on Cardiac Signals From Epicardially Placed...

Area of Science:

• Biomedical Engineering
• Cardiovascular Physiology
• Artificial Intelligence in Medicine

Background:

• Miniaturized accelerometers on pacing leads monitor cardiac function via myocardial acceleration signals.
• Accurate extraction of functional indices requires precise detection of aortic valve opening (AVO) and aortic valve closure (AVC).
• Automated detection of AVO and AVC is crucial for advancing cardiac monitoring.

Purpose of the Study:

• To evaluate the efficacy of deep learning models in detecting AVO and AVC events.
• To assess the accuracy of deep learning algorithms using signals from epicardially attached accelerometers.
• To establish ground truth for valve events using high-fidelity pressure measurements.

Main Methods:

• A deep neural network comprising CNN, RNN, and multi-head attention was developed.
• The model was trained and tested on 130 canine and 159 porcine recordings.
• Nested cross-validation was employed to evaluate model accuracy due to limited data.

Main Results:

• High detection rates for AVO (98.9% canine, 98.2% porcine) and AVC (97.1% canine, 96.7% porcine) were achieved within 40 ms of ground truth.
• Low incorrect detection rates were observed: 0.7% AVO and 2.3% AVC in canines, 1.1% AVO and 2.3% AVC in porcines.
• Mean absolute errors were low, ranging from 7.2 ms to 10.1 ms for AVO and AVC events.

Conclusions:

• Deep neural networks provide a robust and accurate method for detecting aortic valve events.
• Epicardially attached accelerometers coupled with deep learning can reliably monitor AVO and AVC.
• This approach holds significant potential for improving cardiac function monitoring and analysis.