FACE
AI-supported analysis of long-term ECGs - The analysis of long-term ECG data is time-consuming and error-prone in medical practice, although it is crucial for the diagnosis of cardiac arrhythmias such as atrial fibrillation. The FACE project is developing AI-based methods for the automated analysis of large volumes of ECG data in order to detect abnormalities more reliably and improve diagnostics. Various AI models are being trained, compared and optimized for practical use. A particular focus is on data protection-friendly processing directly on site (edge computing) and testing the technology in real medical application scenarios.
Project description
AI-supported ECG analysis for cardiac medicine (FACE) - The manual evaluation of long-term ECG recordings takes up only a limited amount of time in medical practice, but involves enormous amounts of data: 24 hours of recording correspond to over 100,000 heartbeats that need to be assessed. Abnormalities such as atrial fibrillation, arrhythmia or other irregularities can provide vital information. Despite this importance, the average error rate for conventional evaluation is around 25 percent.
The "FACE" project addresses this challenge by using artificial intelligence to analyze heart data automatically and reliably. The focus is on developing, evaluating and optimizing various AI models that are trained on real ECG data sets in order to detect noise, classify arrhythmias and deliver reliable analysis results in a fraction of the time. In addition to performance, particular attention is paid to a data protection-friendly edge architecture in which sensitive patient data is processed directly on site and without cloud transmission. Through collaboration with practice partners, clinics and device manufacturers, the integration of AI analysis into everyday medical practice is to be tested and a feasibility study carried out in practice.
By using structured data, comparative AI evaluations and practical tests, the project improves the reliability of long-term ECG diagnostics, relieves the burden on medical professionals and at the same time lays the foundation for future cloud-based and edge-optimized assistance solutions in cardiac medicine.
Focal points of the project
- AI-supported ECG evaluation: Development and training of AI models for the automated detection and classification of cardiac arrhythmias in long-term ECG data.
- Reliability and quality assurance: Systematic comparison and optimization of different AI methods to reduce misinterpretations and improve diagnostic accuracy.
- Privacy-friendly edge architecture: Processing of sensitive patient data directly on site without cloud transmission to meet data protection requirements and enable practical integration into everyday clinical and practice work.
Everything at a glance
-
Duration
01.01.2024 - 31.12.2026 (Ongoing) -
Research area
AI and medicine -
Funding
Former Federal Ministry of Economics and Climate Protection (BMWK) funding: €430,000
Research methods & procedure
FRACTAL adopts a disruptive, cross‑layer methodology. The project uses cutting‑edge microelectronics, high‑performance processors and AI techniques to build a modular, open reference architecture for cognitive edge computing. A modular open reference architecture allows components to be selected and configured based on application needs and is visualised through the FRACTAL System configurator. Hardware advances include a Hierarchical Adaptive Time‑Triggered Multicore Architecture (HATMA) that enables adaptation while balancing safety, security, power and environmental scenarios; it leverages an Adaptive Time‑Triggered Network on Chip (ATTNoC) for scheduling adaptation. An AI‑based metascheduler supports dependable, time‑critical scheduling and run‑time adaptation. The software stack uses microservices and provides data ingestion, storage, orchestration and containerised ML model servers; this ensures the same open architecture across low‑, mid‑ and high‑end edge nodes. FRACTAL validated its methodology through eight real‑world use cases, demonstrating the modular architecture and edge AI capabilities across domains
Development of specialized AI models
We are developing modern AI models that are trained to recognize complex patterns in the electrical activity of the heart. We focus on five central topics: the evaluation of signal quality, precise heartbeat detection and the detailed analysis of rhythm, morphology and specific cardiological events.
Implementation of a cloud edge architecture
To make the analysis highly efficient, we use a combined computing strategy. This involves pre-processing the data directly on the mobile device (edge), while computationally intensive analyses are carried out in the cloud. This enables fast and in-depth monitoring of cardiac arrhythmias in real time.
Validation on the real system
The algorithms developed are continuously tested with state-of-the-art hardware. The aim is to significantly improve the diagnostic quality of long-term ECGs (24-72 hours) and to reduce the workload of medical staff through automated pre-sorting.
