Conference Agenda

Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).

 
 
Session Overview
Session
11f. Joint Session mit GMDS: Synthetic and augmented image and biosignal data enhancing and enhanced by AI research
Time:
Wednesday, 18/Sept/2024:
10:00am - 11:30am

Session Chair: Nicolai Spicher
Session Chair: Sebastian Zaunseder
Location: V 9.02

Session Topics:
Biosignal Analysis and Data Aggregation

Show help for 'Increase or decrease the abstract text size'
Presentations
10:00am - 10:20am
ID: 118 / Synthetic and augmented image an: 1
Abstract
Oral Session
Topics: Biosignal Analysis and Data Aggregation

Generative models for medical EEG time series data - applications, challenges, and opportunities

Alexandra Reichenbach1,2, Yasmin Hollenbenders1,2, Friedrich Carrle1,2

1Heilbronn University of Applied Sciences, Germany; 2Heidelberg University, Germany

Introduction

Electroencephalography (EEG) offers the opportunity to derive robust biomarkers for diagnosis and prediction of treatment response for psychiatric diseases such as major depressive disorder (MDD) or attention deficit disorder (ADD) [1]. Generative adversarial networks (GAN) form a class of generative machine learning models capable of producing data with desired characteristics by comparison with data processing said characteristics [2].

Methods

GANs can be utilized to create EEG time-series data from random noise to boost the training data set e.g. for a diag-nostic MDD classifier [3]. Beyond this direct usage, synthetic EEG data also offers the opportunity to create public datasets for developing clinical applications without privacy concerns. Instead of creating synthetic data from scratch, GANs can also be used for pre-processing tasks such as artefact removal [4] from noisy EEG data.

Results

Some characteristics of the EEG signal can be faithfully reproduced by GANs, some cannot yet, which becomes espe-cially apparent in the frequency domain [3].

Conclusion

In order to further the development of models for synthetic EEG data, however, suitable evaluation metrics are still needed. This work demonstrates the current state of what GANs can achieve with EEG signals exemplarily on selected use cases. It furthermore pinpoints their shortcomings when it comes to EEG data. I conclude with a discussion of the next steps to further this field.

Reichenbach-Generative models for medical EEG time series data-118_a.pdf


10:20am - 10:40am
ID: 128 / Synthetic and augmented image an: 2
Abstract
Oral Session
Topics: Biosignal Analysis and Data Aggregation

Mechanistic Modeling and Simulation of the ECG as an Enabling Technology for Machine Learning: Concepts, Datasets, and Real-World Performance

Silvia Becker, Axel Loewe

Karlsruhe Institute of Technology (KIT), Germany

Machine learning in medicine is often hindered by limited availability of large, high-quality datasets. Recently, synthetic ECGs were used to enrich sparse clinical data or even replace them completely during training of machine learning approaches leading to improved performance on real-world clinical test data.

We introduce the concept of computational modeling & simulation of organ systems in the human body using the example of cardiac electrophysiology and highlight how simulation spaces can be augmented in a biologically consistent way. This scalable and well controlled research environment allows for the generation of almost arbitrary large, high-fidelity, well-labeled datasets such as the publicly available MedalCare-XL dataset.

Applications comprise the enrichment of clinical data in so-called hybrid datasets that were used with ECG data for identifying subtypes of atrial flutter and for diagnosing left atrial enlargement. Going further, such synthetic datasets can also be used as the sole resource during training and then evaluated on clinical data eventually. We will show examples for the prediction of the acute success of pulmonary vein isolation to treat atrial fibrillation and for localizing the origin of ventricular extrasystoles from body surface recordings to inform clinical procedure planning.

In conclusion, we present how mechanistic and data-driven modeling can synergistically complement each other and how simulation-based approaches can help to overcome current challenges in machine learning for medicine.

Becker-Mechanistic Modeling and Simulation of the ECG as an Enabling Technology-128_a.pdf


10:40am - 11:00am
ID: 132 / Synthetic and augmented image an: 3
Abstract
Poster Session
Topics: Biosignal Analysis and Data Aggregation

Diffusion-based conditional ECG generation with structured state space models

Juan Miguel Lopez Alcaraz, Nils Strodthoff

Carl von Ossietzky Universität Oldenburg, Germany

Introduction

Generating synthetic data is a promising solution for addressing privacy concerns that arise when distributing sensitive health data. In recent years, diffusion models have become the new standard for generating various types of data, while structured state space models have emerged as a powerful approach for capturing long-term dependencies in time series.

Methods

Our proposed solution, SSSD-ECG, combines these two technologies to generate synthetic 12-lead electrocardiograms (ECGs) based on over 70 ECG statements. As reliable baselines are lacking, we also propose conditional variants of two state-of-the-art unconditional generative models. We conducted a thorough evaluation of the quality of the generated samples by assessing pre-trained classifiers on the generated data and by measuring the performance of a classifier trained only on synthetic data.

Results

SSSD-ECG outperformed its GAN-based competitors. Our approach was further validated through experiments that included conditional class interpolation and a clinical Turing test, which demonstrated the high quality of SSSD-ECG samples across a wide range of conditions.

Conclusion

Through rigorous evaluation, SSSD-ECG outperformed existing GAN-based approaches, showcasing its effectiveness in producing high-quality ECG samples across a range of conditions. This approach holds promise for addressing privacy concerns in health data distribution while preserving data utility.

Lopez Alcaraz-Diffusion-based conditional ECG generation with structured state space models-132_a.pdf


11:00am - 11:20am
ID: 370 / Synthetic and augmented image an: 4
Abstract
Oral Session
Topics: Biosignal Analysis and Data Aggregation

Creating Synthetic Long Term Coherent ECGs with Latent Diffusion Models

Dominik D. Kranz1,2, Jan F. Krämer1, Oruc Kahriman1, Alexander Nelde2, Maximilian Schöls2, Christian Meisel2, Niels Wessel1,3

1Humboldt-Universität zu Berlin, Germany; 2Charité - Universitätsmedizin Berlin; 3Medical School Berlin

Following the introduction of products like DALLE and ChatGPT, generative AI models have demonstrated tremendous capabilities in the areas of Natural Language and Image Processing. As access to high quality biosignal data remains one of the limiting factors for the development of new methods, especially in the context of datasets for machine learning approaches, we investigated whether generative AI architectures could be employed to generate or augment such datasets.

Kranz-Creating Synthetic Long Term Coherent ECGs with Latent Diffusion Models-370_a.pdf


 
Contact and Legal Notice · Contact Address:
Privacy Statement · Conference: BMT 2024
Conference Software: ConfTool Pro 2.8.104+TC
© 2001–2025 by Dr. H. Weinreich, Hamburg, Germany