pyPCG: a Python toolbox specialized for phonocardiography analysis.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Phonocardiography (PCG) is increasingly used for remote and low-cost monitoring, particularly for fetal heart rate.
• Existing PCG analysis methods lack standardization, requiring extensive custom implementation.
• Standardization of PCG datasets and labeling, especially for fetal recordings, is critically needed.

Purpose of the Study:

• To introduce a standardized toolbox for heart sound analysis, serving as a foundation for future frameworks.
• To provide a modular set of widely used processing steps for creating complex analysis pipelines.
• To enable individual testing and fine-tuning of functions for specific datasets.

Main Methods:

• Development of a Python toolbox (pyPCG) for phonocardiogram analysis.
• Validation of the segmentation stage using a fetal PCG dataset (50 recordings) and a PhysioNet Challenge dataset (413 records).
• Comparison of the toolbox's segmentation performance against established methods like Neurokit2 and Hidden Semi-Markov Model.

Main Results:

• The best model achieved a 96.1% F1 score and 11.7 ms mean absolute error for fetal S1 detection.
• For the PhysioNet dataset, the model attained an 81.3% F1 score and 50.5 ms mean absolute error for S1 detection.
• The developed method surpassed other tested methods on the fetal dataset and showed state-of-the-art performance on the PhysioNet dataset.

Conclusions:

• Accurate signal segmentation is crucial for reliable statistical measures and classification models in PCG analysis.
• The pyPCG toolbox provides compatible functions for feature extraction and statistical calculations.
• The toolbox facilitates fine-tuning for diverse datasets and is available for public use.