Towards Transparent ECG Diagnostics - When AI Listens to the Heart
Dr. Temesgen Mehari defended his PhD in early 2025 with a thesis titled "Advancing Cardiac Health: Trustworthy and Practical Approaches to Deep 12-lead ECG Analysis." His research bridges the fields of AI and medicine, focusing on the development of explainable and robust deep learning models for ECG diagnostics. He defended his thesis under the official supervision of BIFOLD co-director Klaus-Robert Müller. His examination committee also included Wojciech Samek, Stefan Haufe, and Sebastian Zaunseder. Throughout the PhD, he worked closely with Nils Strodthoff, who played a key role in his academic supervision.
Please describe and explain your research focus.
Temesgen: My research focuses on utilizing artificial intelligence to enhance the analysis of electrocardiograms (ECGs) - the curves obtained when measuring the heart's electrical activity. These signals are crucial for detecting heart conditions, such as arrhythmias - abnormal heart rhythms - or even heart attacks. I developed deep learning models that can not only recognize these conditions more accurately, but also explain why these models made a certain decision, helping doctors trust and understand AI-assisted diagnoses.
What are your recent projects?
Temesgen: I am currently developing a tool that uses advanced AI models to automatically analyze ECGs and generate diagnostic summaries in natural language. The aim is to connect innovative research with practical clinical applications, particularly in underserved or low-resource settings.
What draws you to research?
Temesgen: To me, research feels like the creative side of tech. It is all about exploring new approaches or reimagining old ideas to solve interesting problems. While engineering often focuses on building products or increasing efficiency, I believe that research is more about the journey itself and the underlying concepts and ideas. I am especially drawn to the ideas behind AI.
Could you illustrate one of the “ideas behind AI”?
Temesgen: I'm fascinated by Yann LeCun for his groundbreaking contributions to the field of AI. Convolutional neural networks and his early advocacy of self-supervised learning are, in my view, two of the most impactful ideas in the field. What fascinates me most is witnessing their transformative impact firsthand. Despite his Turing Award, I believe his work is still underappreciated and will be recognized as even more foundational in the decades to come.
Which major innovations do you expect in your research field in the next ten years?
Temesgen: I expect the integration of multimodal AI, that is, models that can simultaneously interpret signals, images, and patient history, to become standard in clinical diagnostics. For cardiology, this could mean AI systems that combine ECGs with ultrasound images and laboratory results to deliver a more holistic and personalized diagnosis. I think they will likely not only support decisions but also generate full-text diagnostic reports that are on par with those written by medical professionals.
Where would one find you if you are not sitting in front of the computer?
Temesgen: I enjoy photographing everyday life on film as a slow, mindful counterbalance to working in tech. Also, you’ll most likely find me outside running or cycling, possibly training for a race.
