PALMS

PALMS facilitates creation and refinement of labeled reference datasets for electrocardiogram (ECG) and photoplethysmogram (PPG) medical time series to support development and validation of signal-processing and machine learning algorithms.

Key Features:

• Manual annotation types: Support for fiducials such as R-peaks, events with adjustable durations (e.g., arrhythmic episodes), and signal quality assessments including identification of segments corrupted by motion artifacts, with simultaneous annotation of multiple aspects within the same signal.
• Flexible configuration: Adaptable configuration system that accommodates various data types and annotation requirements.
• Automated algorithm integration: Integration of existing algorithms for automated signal processing, including automatic R-peak detection, with provision for manual correction of algorithm outputs.
• Built-in validated algorithms: Included ECG and PPG algorithms with reported performance metrics: ECG algorithm 99% success rate on the MIT/BIH arrhythmia database and PPG algorithm F1-score above 98% across two public databases.
• Implementation: Developed in Python.

Scientific Applications:

• Reference dataset generation: Creation of diverse, high-quality labeled ECG and PPG datasets for training and testing signal-processing and machine learning methods.
• Algorithm development and validation: Evaluation and refinement of automated detection algorithms, including R-peak detectors, using combined automated outputs and manual annotations.
• Clinical research support: Production of validated annotations to improve accuracy and reliability of ECG and PPG interpretation in diagnostic and therapeutic studies.

Methodology:

Combination of manual annotation and integration of automated algorithms for initial signal processing, with manual refinement of algorithmic outputs to produce labeled datasets for training and validating machine learning models.