Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Generative Models for Physiological Signals: A Systematic Literature Review

Nour Neifar, Afef Mdhaffar, Achraf Ben-Hamadou, Mohamed Jmaiel

TL;DR

This paper addresses the gap in systematic understanding of deep generative models for physiological time-series signals (ECG, EEG, PPG, EMG). It surveys 71 studies across GANs, VAEs, and diffusion models, detailing how these architectures are used for data augmentation, denoising, imputation, forecasting, modality transfer, and anomaly detection, while also cataloging evaluation protocols and datasets. The review highlights the dominance of GANs, the growing interest in diffusion models for stable training, and the role of conditional variants and priors tailored to physiological dynamics. Key datasets (e.g., MIT-BIH, PTB-XL, SEED/DEAP) and metrics (RMSE, MMD, FID, DTW) are cataloged to support benchmarking and reproducibility. The work provides guidance for researchers on model selection, evaluation, and future directions, including multimodal integration and standardized benchmarking.

Abstract

In this paper, we present a systematic literature review on deep generative models for physiological signals, particularly electrocardiogram (ECG), electroencephalogram (EEG), photoplethysmogram (PPG) and electromyogram (EMG). Compared to the existing review papers, we present the first review that summarizes the recent state-of-the-art deep generative models. By analyzing the state-of-the-art research related to deep generative models along with their main applications and challenges, this review contributes to the overall understanding of these models applied to physiological signals. Additionally, by highlighting the employed evaluation protocol and the most used physiological databases, this review facilitates the assessment and benchmarking of deep generative models.

Deep Generative Models for Physiological Signals: A Systematic Literature Review

TL;DR

This paper addresses the gap in systematic understanding of deep generative models for physiological time-series signals (ECG, EEG, PPG, EMG). It surveys 71 studies across GANs, VAEs, and diffusion models, detailing how these architectures are used for data augmentation, denoising, imputation, forecasting, modality transfer, and anomaly detection, while also cataloging evaluation protocols and datasets. The review highlights the dominance of GANs, the growing interest in diffusion models for stable training, and the role of conditional variants and priors tailored to physiological dynamics. Key datasets (e.g., MIT-BIH, PTB-XL, SEED/DEAP) and metrics (RMSE, MMD, FID, DTW) are cataloged to support benchmarking and reproducibility. The work provides guidance for researchers on model selection, evaluation, and future directions, including multimodal integration and standardized benchmarking.

Abstract

In this paper, we present a systematic literature review on deep generative models for physiological signals, particularly electrocardiogram (ECG), electroencephalogram (EEG), photoplethysmogram (PPG) and electromyogram (EMG). Compared to the existing review papers, we present the first review that summarizes the recent state-of-the-art deep generative models. By analyzing the state-of-the-art research related to deep generative models along with their main applications and challenges, this review contributes to the overall understanding of these models applied to physiological signals. Additionally, by highlighting the employed evaluation protocol and the most used physiological databases, this review facilitates the assessment and benchmarking of deep generative models.
Paper Structure (24 sections, 2 equations, 10 figures, 4 tables)

This paper contains 24 sections, 2 equations, 10 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Research questions
  4. Query
  5. Inclusion and exclusion criteria
  6. Data collection
  7. Results and findings
  8. RQ1: classes of deep generative models for ECG, PPG, EEG, and EMG signals
  9. Generative adversarial networks
  10. Variational autoencoders:
  11. Diffusion models:
  12. RQ2: application of deep generative models
  13. Data augmentation
  14. Denoising
  15. Imputation
  16. ...and 9 more sections

Figures (10)

  • Figure 1: Samples of physiological signals from clinical datasets: ECG, EEG, PPG, and EMG.
  • Figure 2: Search methodology used for paper selection in our systematic review. From a total of 588 studies, we selected 55 studies through three exclusion steps, and we also included an additional 16 studies recommended by the expert.
  • Figure 3: Number of papers for each class of deep generative models according to the different considered physiological signals.
  • Figure 4: Principle of Generative Adversarial Networks.
  • Figure 5: Principle of Variational Autoencoders.
  • ...and 5 more figures