News

We are delighted to announce that our research group achieved a Top-10 ranking at the George B. Moody PhysioNet/Computing in Cardiology Challenge 2025.

This year’s challenge focused on developing advanced machine learning methods for the detection of Chagas disease from ECG data, bringing together more than 40 international teams from universities, hospitals, and industry. Our submission, entitled:

“Sequential Deep Learning for Chagas Disease Screening: A CNN-BiLSTM Approach with an Attention Mechanism” 

was developed by Saber Jelodari Mamaghani from the Ambient Assisted Living & Medical Assistance Systems Group, Department of Computer Science, University of Bayreuth. 

Our open-source system combines convolutional neural networks (CNNs), bidirectional long short-term memory (BiLSTM) layers, and an attention mechanism to robustly identify subtle waveform changes in ECGs that are characteristic of Chagas disease. To handle class imbalance and data variability, we implemented focal loss, ECG-specific data augmentation, mixup, and weighted sampling.

✨ Result: Our team secured a 10th place finish out of 40 teams worldwide, ahead of many leading international groups. On our local holdout test set, our best model achieved a Challenge score of 238.  

The Challenge winners included teams from KU Leuven, Medical University of Innsbruck, National Taiwan University, ETH Zürich, University of Basel, and others, reflecting the global competitiveness of this year’s event. 

This success highlights the growing impact of our group’s research in AI for healthcare!