A Resource-Efficient Cardiac Arrhythmia Detection Using Nonlinear Dynamics in Optimized Delay State Networks.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Signal Processing

Background:

• Traditional arrhythmia detection methods struggle with subtle temporal dynamics and require significant computational resources.
• Existing techniques often lack real-time applicability and efficiency for early diagnosis.
• Handcrafted features in conventional approaches limit their generalizability and effectiveness.

Purpose of the Study:

• To develop a novel methodology for enhanced cardiac arrhythmia detection and classification.
• To improve the efficiency and real-time applicability of arrhythmia diagnostic tools.
• To create a robust, scalable solution for time-series-based diagnostics in resource-constrained environments.

Main Methods:

• Combining Reconstructed Phase Space (RPS) analysis with an optimized Delay State Network (DSN).
• Leveraging the entire Phase Space Structure (PSS) as input to the DSN, which uses a single nonlinear node with delayed feedback.
• Integrating delay and embedding optimization, with PCA and Ridge Embedding for dimensionality management, and utilizing shared memory and multiprocessing for scalability.

Main Results:

• Achieved 99.3% accuracy, 99.1% sensitivity, and 99.7% specificity on benchmark datasets.
• Demonstrated efficient edge deployment on a Raspberry Pi 5, with inference times of 1.2-4.8 seconds for ECG segments.
• Showcased low power consumption (<2.5 W) and manageable memory usage (2.57 GB for 60s segments).

Conclusions:

• The proposed RPS and optimized DSN framework offers a robust, scalable, and accurate solution for arrhythmia classification.
• The methodology significantly reduces hardware demands compared to conventional deep learning models.
• This approach is suitable for real-time, resource-constrained diagnostic applications and broader time-series analysis.