Domain Agnostic Post-Processing for QRS Detection Using Recurrent Neural Network.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Deep-learning models for QRS detection require post-processing for R-peak localization.
• Existing methods use domain-specific filters and thresholds, leading to dataset bias and performance variability.
• The relative contributions of deep learning and post-processing are often not well-defined.

Purpose of the Study:

• To investigate the effectiveness of domain-specific post-processing in deep-learning-based QRS detection.
• To develop and evaluate a novel domain-agnostic automated post-processing method.
• To compare the performance of the proposed method against traditional techniques.

Main Methods:

• Identified three key steps in domain-specific post-processing for QRS detection.
• Developed a recurrent neural network (RNN)-based model for automated, domain-agnostic post-processing.
• Evaluated the RNN post-processor against domain-specific methods on various datasets, including TWADB.

Main Results:

• Domain-specific post-processing, while effective, introduces bias and limits generalizability.
• The RNN-based post-processor showed superior performance in most cases compared to domain-specific methods.
• Performance differences were minimal (≤ 2%) even in challenging scenarios, highlighting the RNN's consistency.

Conclusions:

• A domain-agnostic RNN-based post-processor offers a stable and generalizable alternative to traditional methods for QRS detection.
• Automated post-processing using RNNs can achieve high accuracy and consistency in R-peak localization.
• This approach advances the development of robust and reliable deep-learning cardiovascular signal analysis tools.