Enhancing Heart Failure Care: Deep Learning-Based Activity Classification in Left Ventricular Assist Device Patients

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiovascular Device Technology

Background:

• Accurate patient activity classification is vital for adaptive control of left ventricular assist devices (LVADs).
• Current methods may lack the precision needed for real-time adjustments in device operation.
• Closed-loop control systems require reliable feedback on patient status.

Purpose of the Study:

• To develop and evaluate deep neural networks (DNNs) for precise activity classification in LVAD patients.
• To compare the performance of binary and multiclass DNN classifiers for different activity states.
• To identify optimal DNN architectures and data integration strategies for enhanced classification accuracy.

Main Methods:

• Analysis of physiological data (heart rate, LVAD flow) and accelerometer data from 13 LVAD patients.
• Training of binary and multiclass DNNs, including recurrent and convolutional layers.
• Hyperparameter optimization and testing of various model architectures, specifically bidirectional long short-term memory (LSTM) layers.

Main Results:

• Integration of LVAD flow, heart rate, and accelerometer data yielded the highest classification accuracy.
• Optimal DNN architectures achieved 91% accuracy for binary classification (active/inactive) and 84% for multiclass classification.
• Bidirectional LSTM layers proved effective, with two layers for binary and three for multiclass tasks.

Conclusions:

• Deep neural networks offer a robust and accurate method for classifying patient activities in the context of LVADs.
• This technology is essential for enabling effective closed-loop control and optimizing device performance in cardiac care.
• The findings support the integration of AI-driven activity classification into future medical device control systems.