And the winners of the Imaging Open Challenge are… 

The BHF Data Science Centre recently ran an Imaging Open Challenge, engaging a global community of researchers and data scientists. The aims of this challenge were to see whether artificial intelligence could be used to identify and classify heart rhythm disorders from electrocardiogram (ECG) images, offer a space for collaboration, and provide a valuable learning opportunity for entrants, many of whom were early-career scientists. 

The challenge ran for 12 weeks, from October to December 2024, and was hosted through the platform Kaggle. This provided competitors with a forum to chat through their ideas and ask questions, and a leaderboard showing results in real time. 152 entrants from across the world, split into 11 teams, got involved.  The top 7 teams, whittled down from 208 submissions, made it through to round two. Following a review from the Open Challenging Steering Group, we are very pleased to announce the winner of this Imaging Open Challenge is Xiaoyu Wang (University of Leeds), and the runners up were Oğuzhan Büyüksolak, Toygar Tanyel and İlkay Öksüz (Istanbul Technical University). 
 
The winners will have the opportunity to discuss their teams’ findings and approaches to the Challenge at this year’s British Cardiovascular Society Annual Conference in Manchester.   

Winner, Xiaoyu Wang, said “As a PhD student at the University of Leeds, I was excited to hear about the BHF Open Challenge. The challenge, to interpret Electrocardiogram (ECG) images, aligned with both my technical interests and the clinical topic of my PhD. I was particularly drawn to the clinically-relevant problem of interpreting ECG images, which mimics the everyday reality for many doctors. 

“Under the mentorship of my supervisor, Dr David Wong, I was pleased to have developed a deep learning model that could successfully interpret ECG images. This is a key improvement over historical research that required raw ECG signal data. 

“Moving forward, I plan to improve my deep learning model to better capture the fine details in images. I will also reuse this model in my PhD as I develop new ways to detect early signs of arrhythmias” 

Runner up, İlkay Öksüz, said “We decided to take part in the BHF Data Science Centre ECG Challenge because it closely aligned with our ongoing project on ‘Acute Coronary Syndrome Diagnosis from ECG using Explainable Deep Learning.’. In our project we face the lack of digitally stored ECG data.  

“The challenge offered a rich, realistic dataset and the opportunity to benchmark our methods against other top teams, making it a perfect fit. Being part of the competition was a truly rewarding experience — the dataset’s complexity and realism pushed us to refine our approaches and think creatively. We learned a great deal about ECG signal segmentation, digitization, and handling real-world artefacts, all of which are critical skills for advancing ECG-based diagnostics.

“Going forward, we are excited to transfer these insights back into our project to enhance our models and make them more robust in real-world clinical settings. We are very grateful for the opportunity to participate and look forward to future challenges!” 

What happened? 

An ECG is a test which records the electrical activity of the heart, including rate and rhythm.  

Using a synthetic imaging dataset (artificially-generated data rather than real-world data) made up of approximately 22,000 simulated ECG images, competitors worked to develop artificial intelligence (AI) algorithms to make clinical diagnoses based on information in these images.  

Competitors were asked to train a model which could identify five conditions from the ECG images.

The public voice in the Challenge 

The Open Challenge had input from three members of the BHF Data Science Centre Public Advisory Group at all stages, from helping to design the Challenge to judging competitors’ submissions. They valued how unfamiliar terms and platforms were explained, and appreciated the use of synthetic data, which is both readily available and anonymous. 

Phil Blakelock, a public contributor, described his involvement in the process: 
 
“As a member of the public and former patient, I want to see intelligent improvements being made so that the NHS can continue to deliver better healthcare to our ever-growing and ever-ageing population. The delivery of healthcare in our changing society must also change – prevention, rather than treatment, has to become a reality. 
 
“The Open Challenge sets the standard and the scene in how A.I. and data science can be applied to the large-scale analysis of ECG traces. In the future, this has the potential for heart disease and many other conditions to be diagnosed and treated, before they become more serious. 

Having the opportunity to participate and contribute to this global Open Challenge has opened my eyes to how the newest and most innovative technologies are now ready to enable the transition to prevention, not just treatment.” 

Opening up the future 

Public contributors also highlighted the need for Open Challenges based on subjects that are “inspirational, understandable, tangible, novel, and hard-hitting”, as well as those which would result in real benefit to patients and the public. This vital feedback and reflections will provide a strong foundation for planning future challenges.  

We look forward to welcoming Xiaoyu, from the University of Leeds, and İlkay, from Istanbul Technical University, at the BCS Conference on 4th June, where they will present their findings.