Snippet Policy Network V2: Knee-Guided Neuroevolution for Multi-Lead ECG Early Classification

Area of Science:

• Artificial Intelligence
• Machine Learning
• Biomedical Signal Processing

Background:

• Early time series classification is crucial for time-critical applications, such as monitoring patients with arrhythmias in intensive care units.
• Timely predictions can significantly improve patient outcomes and potentially save lives by enabling prompt treatment.
• Existing methods may not optimally balance classification accuracy with the need for early decision-making.

Purpose of the Study:

• To propose a novel deep reinforcement learning framework, snippet policy network V2 (SPN-V2), for early classification of long and varied-length multi-lead electrocardiogram (ECG) data.
• To develop a method that effectively learns the trade-off between classification accuracy and decision earliness.
• To optimize the framework using a novel knee-guided neuroevolution algorithm (KGNA).

Main Methods:

• The proposed SPN-V2 framework integrates snippet representation learning (SRL) and early classification timing learning (ECTL).
• SRL encodes spatial and temporal correlations within ECG snippets.
• ECTL employs a decision agent for accurate and early time series classification.
• KGNA is utilized to optimize SPN-V2 for the accuracy-earliness trade-off.

Main Results:

• Experiments on two real-world ECG datasets demonstrate the effectiveness of the proposed SPN-V2 framework.
• The KGNA optimization algorithm successfully balances accuracy and earliness in classification.
• The proposed algorithm shows superior performance compared to existing state-of-the-art methods.

Conclusions:

• The SPN-V2 framework offers a promising approach for early time series classification in critical applications.
• The KGNA algorithm provides an effective means to optimize deep reinforcement learning models for accuracy-earliness trade-offs.
• This research advances the field of early ECG classification, with potential benefits for patient monitoring and diagnosis.