Medical Time Series

Fraunhofer HHI possesses comprehensive expertise in the analysis of medical time series. Within the framework of the DAKI project (https://daki-fws.de), we have developed innovative AI-based tools for pandemic control and conducted pioneering research on predicting disease spread and progression. 

Our focus lies on developing robust and reliable forecasting systems with high spatial resolution. To this end, we have built an extensive pandemic database that intelligently links multimodal data sources – from weather and wastewater data to vaccination and mobility data, as well as Google Trends and statistical surveys from various institutions. 

Based on this foundation, we developed a state-of-the-art live forecasting system for COVID-19, whose results were made available to German health authorities as a decision-making basis. 

Experience our technology: Our interactive demonstrator for visualizing and predicting disease data is available at: https://hhi.fraunhofer.de/aml-demonstrator/health 

Furthermore, HHI explores innovative approaches for integrating historical epidemiological insights into modern data-driven methods. We use partial differential equations to systematically embed proven epidemiological principles into AI models. 

In our current research work from 2025 [1], we demonstrated a groundbreaking approach: Using epidemiological differential equations, we generated synthetic training data for our forecasting models. The results show a significant performance improvement over purely data-based methods. 

The differential equation we developed models both the temporal evolution and spatial diffusion of infected persons, thus creating a realistic foundation for more precise predictions. 

The effectiveness of our methods was also confirmed in international applications: We achieved similarly outstanding results when analyzing Mexican COVID-19 data as well as other disease datasets from different regions. 

In the course of these research activities, valuable cooperations emerged with institutions such as the Robert Koch Institute, the Brazilian Fiocruz, and the Helmholtz Centre for Environmental Research, which contribute to the continuous development and validation of our technologies.

Fraunhofer HHI has also built substantial knowledge in the field of electrocardiography (ECG) in recent years. With the PTB-XL dataset [2], HHI provides a comprehensive publicly available collection – consisting of over 21,800 clinical 12-lead ECGs with a duration of 10 seconds each and more than 70 diagnostic statements that are hierarchically structured (according to SCP-ECG standard) and validated by up to two cardiologists. The dataset also contains extensive metadata on patient demographics, infarction characteristics, diagnostic probabilities, and signal features, and offers pre-structured training and test splits to promote the comparability of machine learning models. 

Since then, PTB-XL has been repeatedly established as a benchmark for deep learning models in the ECG context. ResNet and Inception architectures regularly achieved top positions in classification tasks such as diagnostics, age and gender estimation. Recent work at HHI expanded these analytical tools: By employing structured state space models (SSMs), methodologically combined with self-supervised learning and the integration of demographic metadata, significant performance improvements on PTB-XL were achieved – both in supervised and self-supervised settings. 

In parallel, HHI engaged in the field of Responsible AI and in the certification of AI systems in the medical context.  As part of project P540 (“Use of Artificial Intelligence in Medical Diagnosis and Prognosis Systems”), a prototype development of test criteria and evaluation procedures was undertaken to ensure the reliability and safety of AI systems in medicine. This work was published by the BSI [3]. 

[1] Jost Arndt, Utku Isil, Michael Detzel, Wojciech Samek, Jackie Ma Synthetic Datasets for Machine Learning on Spatio-Temporal Graphs using PDEs, Journal of Data-centric Machine Learning Research, May 2025, Open Access  

[2] Wagner, P., Strodthoff, N., Bousseljot, R., Samek, W., & Schaeffter, T. (2022). PTB-XL, a large publicly available electrocardiography dataset (version 1.0.3). PhysioNet. RRID:SCR_007345. https://doi.org/10.13026/kfzx-aw45  

[3] Bundesamt für Sicherheit in der Informationstechnik. (n.d.). Projekt P540 – Einsatz von Künstlicher Intelligenz in medizinischen Diagnose- und Prognosesystemen. Retrieved August 20, 2025, from https://www.bsi.bund.de/SharedDocs/Downloads/DE/BSI/Publikationen/Studien/Projekt_P540/Projekt_P540.html