Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Usable GAN-Based Tool for Synthetic ECG Generation in Cardiac Amyloidosis Research

Francesco Speziale, Ugo Lomoio, Fabiola Boccuto, Pierangelo Veltri, Pietro Hiram Guzzi

TL;DR

The paper addresses the challenge of scarce and imbalanced CA datasets by introducing a GAN-based tool that generates labeled 1D ECG beats with class-specific generators. A bidirectional-LSTM GAN framework and a Streamlit GUI enable reproducible training and interactive generation of synthetic CA beats, preserving minority-class morphology for downstream clustering and AI-ECG tasks. The proposed evaluation framework combines morphologic, physiologic, and clustering metrics (e.g., DTW, Fréchet, KS, MMD, silhouette, ARI) to ensure synthetic data are realistic, diverse, and medically meaningful. This tool offers a practical, deployable solution to balance CA cohorts, facilitate early diagnosis, and support patient stratification in CA research while maintaining data provenance and usability across research teams.

Abstract

Cardiac amyloidosis (CA) is a rare and underdiagnosed infiltrative cardiomyopathy, and available datasets for machine-learning models are typically small, imbalanced and heterogeneous. This paper presents a Generative Adversarial Network (GAN) and a graphical command-line interface for generating realistic synthetic electrocardiogram (ECG) beats to support early diagnosis and patient stratification in CA. The tool is designed for usability, allowing clinical researchers to train class-specific generators once and then interactively produce large volumes of labelled synthetic beats that preserve the distribution of minority classes.

A Usable GAN-Based Tool for Synthetic ECG Generation in Cardiac Amyloidosis Research

TL;DR

The paper addresses the challenge of scarce and imbalanced CA datasets by introducing a GAN-based tool that generates labeled 1D ECG beats with class-specific generators. A bidirectional-LSTM GAN framework and a Streamlit GUI enable reproducible training and interactive generation of synthetic CA beats, preserving minority-class morphology for downstream clustering and AI-ECG tasks. The proposed evaluation framework combines morphologic, physiologic, and clustering metrics (e.g., DTW, Fréchet, KS, MMD, silhouette, ARI) to ensure synthetic data are realistic, diverse, and medically meaningful. This tool offers a practical, deployable solution to balance CA cohorts, facilitate early diagnosis, and support patient stratification in CA research while maintaining data provenance and usability across research teams.

Abstract

Cardiac amyloidosis (CA) is a rare and underdiagnosed infiltrative cardiomyopathy, and available datasets for machine-learning models are typically small, imbalanced and heterogeneous. This paper presents a Generative Adversarial Network (GAN) and a graphical command-line interface for generating realistic synthetic electrocardiogram (ECG) beats to support early diagnosis and patient stratification in CA. The tool is designed for usability, allowing clinical researchers to train class-specific generators once and then interactively produce large volumes of labelled synthetic beats that preserve the distribution of minority classes.
Paper Structure (18 sections, 1 figure)

This paper contains 18 sections, 1 figure.

Table of Contents

  1. Introduction
  2. ECG Characteristics of Cardiac Amyloidosis
  3. Low QRS voltage and voltage–mass discrepancy
  4. Pseudoinfarction patterns and repolarisation abnormalities
  5. Conduction disturbances and rhythm disorders
  6. Subtype‑specific and dynamic changes (AL vs ATTR)
  7. Clustering for Early Diagnosis and Patient Grouping
  8. Methods
  9. Data source and preprocessing
  10. GAN architecture
  11. Training strategy
  12. Synthetic signal generation and export
  13. Graphical User Interface
  14. Layout and interaction flow
  15. Usability and deployment
  16. ...and 3 more sections

Figures (1)

  • Figure 1: Streamlit-based graphical user interface for configuring GAN training. The panel illustrates dataset upload (Step 1) and hyper-parameter selection for epochs, batch size, learning rate, random seed and device (Step 3).