AI-Based Detection of Atrial Fibrillation from ECG Data for e-Health Applications

In this project, students will develop an AI model for detecting Atrial Fibrillation (AF) from electrocardiogram (ECG) recordings. The project emphasizes the development of machine learning (ML) models for use in e-Health and Clinical Decision Support (CDS) systems. Using publicly available or simulated ECG datasets, students will train and evaluate deep learning models (e.g., CNN, RNN, LSTM) to classify ECG signals as normal or AF-affected.

This project is derived from the EU-funded SEARCH initiative (https://ihi-search.eu).

In this project, students will develop an AI model for detecting Atrial Fibrillation (AF) from electrocardiogram (ECG) recordings. The project emphasizes the development of machine learning (ML) models for use in e-Health and Clinical Decision Support (CDS) systems. Using publicly available or simulated ECG datasets, students will train and evaluate deep learning models (e.g., CNN, RNN, LSTM) to classify ECG signals as normal or AF-affected.

Additionally, the project will explore explainable AI (XAI) methods to interpret the model’s predictions and ensure transparency, which is crucial for medical applications.

Goals / learning outcomes

Students will:

• Use open ECG datasets (e.g., MIT-BIH AFDB) or simulate ECGs with AF and normal rhythms.
• Train baseline deep learning models (e.g., CNN, LSTM) for binary classification (AF vs. normal).
• Evaluate model performance using clinical metrics (e.g., sensitivity, specificity, AUROC).
• Investigate interpretability using XAI tools (e.g., Grad-CAM, saliency maps).
• Optionally, explore multi-label classification scenarios involving co-existing conditions like ischemic heart disease (IHD).

Atrial Fibrillation is a leading risk factor for stroke and heart failure. Detecting it early using automated, AI-powered tools can significantly improve patient outcomes. This project provides hands-on experience with biomedical signal processing, deep learning, and medical AI applications, contributing to digital health innovations.

Qualifications

• Basic understanding of biomedical signals and time-series data
• Python programming and ML foundations
• Knowledge of deep learning (CNNs, LSTMs) is required

All datasets will be publicly available or simulated, ensuring no privacy risks. The work contributes directly to practical goals in the SEARCH project and aligns with modern trends in interpretable and ethical AI in medicine.

Supervisors

• Vajira Thambawita
• Molly Maleckar
• Pål Halvorsen

Collaboration partners

This project is part of SEARCH (Synthetic hEalthcare dAta goveRnanCe Hub), a multi-disciplinary initiative focused on creating synthetic healthcare data and facilitating secure data sharing across the biomedical ecosystem. Read more about SEARCH here.