Year: 2024
Type: dissertation
Abstract: This dissertation explores the application of Generative Adversarial Networks (GANs) to generate time-series data, with a particular focus on Electrocardiogram (ECG) signals used for arrhythmia detection. Data scarcity in medical fields is compounded by privacy regulations, the technical complexities of data collection, and the rarity of certain pathologies, all of which limit access to comprehensive datasets. With a foundation in the MIT-BIH Arrhythmia Database, this study leverages a Wasserstein GAN with Gradient Penalty (WGAN-GP) architecture and changes the model’s structure by adding bidirectional Long Short-Term Memory (LSTM) layers to generate realistic synthetic ECG signals. These synthetic signals aim to balance datasets for arrhythmia classification, improving classifier performance where traditional Data Augmentation (DA) methods fall short due to privacy, rarity, and complexity constraints in medical data. The GAN model’s training was evaluated using a combination of quantitative metrics such as Euclidean Distance and Dynamic Time Warping (DTW), alongside visual techniques like Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (t-SNE). Additionally, a classification model trained on augmented ECG data demonstrated potential in addressing dataset imbalances and enhancing accuracy in detecting arrhythmic events, demonstrating the GAN’s effectiveness in enhancing model performance. This work contributes to the broader field of healthcare data science. It highlights the potential of GANs to overcome significant challenges by providing privacy-preserving, diverse datasets that improve diagnostic model accuracy. Through this approach, GANs offer a tool for medical research, facilitating the development of robust predictive models while maintaining data integrity and confidentiality. The results underscore the potential for GANs to impact, where enhanced data accessibility and diversity can significantly improve patient outcomes in arrhythmia detection and beyond.
Source: Repositório do ISCTE-IUL
Author: Faria, Carlos Daniel Costa
Language: English
Topic: ECG Monitoring and Analysis
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7008688461