Related Experiment Video
Updated: Dec 23, 2025
09:30
Pre-Implantation Genetic Testing for Aneuploidy on a Semiconductor Based Next-Generation Sequencing Platform
Published on: August 17, 2022
3.4K
PlethAugment: GAN-Based PPG Augmentation for Medical Diagnosis in Low-Resource Settings
IEEE Journal of Biomedical and Health Informatics
|April 29, 2020
Summary
This study introduces PlethAugment, a novel method using conditional generative adversarial networks (CGANs) to create synthetic photoplethysmogram (PPG) signals. This approach enhances machine learning model performance in low-resource clinical settings, addressing data scarcity and class imbalance.
Area of Science:
- Biomedical Engineering
- Machine Learning in Healthcare
- Physiological Signal Processing
Background:
- Limited physiological time-series data from low-resource settings hinders machine learning performance.
- Class imbalance in medical datasets, where certain diagnoses are rare, further degrades model accuracy.
- Existing data augmentation methods like time-warping can negatively impact physiological data distributions, especially those with frequency components.
Purpose of the Study:
- To propose PlethAugment, a novel data augmentation technique for generating pathological photoplethysmogram (PPG) signals.
- To address challenges of data scarcity, class imbalance, and privacy preservation in low-resource clinical environments.
- To improve the performance of machine learning models for medical diagnosis using synthetic physiological data.
Main Methods:
- Development of three conditional generative adversarial networks (CGANs) with an adapted diversity term for PPG signal generation.
- Introduction of a novel metric-agnostic evaluation method, the synthetic generalization curve.
- Validation on two proprietary and two public datasets covering diverse medical conditions.
Main Results:
- Training on PlethAugment-generated data significantly improved classification performance compared to non-augmented, class-balanced datasets.
- An improvement of up to 29% in Area Under the Receiver Operating Characteristic curve (AUROC) was observed using cross-validation.
- The proposed CGANs demonstrate substantial potential for enhancing medical classification accuracy.
Conclusions:
- PlethAugment effectively generates realistic pathological PPG signals, overcoming limitations of traditional augmentation.
- The synthetic generalization curve provides a robust method for evaluating generative models in this domain.
- This work highlights the utility of advanced generative models for improving AI-driven diagnostics in resource-constrained healthcare settings.
