Just to let you know, we’re an affiliate for Amazon, Bluehost, CJ and Rakuten Marketing and some of the links below are affiliate links, meaning that, at no additional cost to you, I may earn a commission if you click through and make a purchase. Thank you if you use our links, we really appreciate it!
Deep learning and AI technology is taking the world by storm and many fear that AI will one day take their jobs. And now it seems that not even Cardiologists are safe!
Researchers at Stanford University have developed a deep learning AI algorithm that can sift through hours of heart rhythm data (generated by wearable monitors that produce electrocardiogram (ECG) signals) and can find potentially life-threatening irregular heartbeats knows as “arrhythmias”.
The team at the Stanford research group was led by Andrew Ng and they managed to develop an algorithm, detailed in this arXiv paper, that is capable of detecting 14 different types of arrhythmia from ECG signals.
Not only was that algorithm able to detect these irregular heartbeats, it was able to match the performance of board certified cardiologists and outperform them in most cases!
Developing the algorithm
To develop the algorithm, researchers collected data from iRhythm’s wearable ECG monitor. Over 30,000 patients wore the monitor for two weeks and were asked to carry out their normal day-to-day activities whilst the device recorded heartbeats for analysis.
Each patient that was selected had abnormal heart rhythms to make the dataset more even and served to increase the likelihood of observing unusual heartbeat activity.
To prepare the machine learning algorithm for training, each ECG record in the training set was manually annotated by a clinical ECG expert. This involved the expert marking each segment of the signal as belonging to one of 14 different rhythm types or classes.
Once the dataset was prepared and complete, the 34-layer convolutional neural network was trained using this data and the TensorFlow deep learning framework. Once trained, the output of the neural network was then able to map sequences of ECG samples it had never seen before into a sequence of rhythm classes it had been trained to recognise.
Putting the algorithm to the test
In order to test the accuracy of the algorithm, the algorithm went head to head with the expert cardiologists in reading and interpreting 300 undiagnosed ECG clips. It turned out that the algorithm was just as likely to agree with a consensus opinion of cardiologists as it was with individual cardiologists. And in most cases, it was more likely to agree with the consensus opinion of the expert cardiologists and was able to outperform cardiologists on most arrhythmias.
Not only is the algorithm competitive with expert cardiologists, algorithms like these have an advantage over their human counterparts in that they don’t get tired and can detect arrhythmia ECG signals instantly and continuously without breaks.
Well in the long term, the researchers hope to use this technology in parts of the developing world and also in rural areas where many don’t have access to cardiologists.
The algorithm could even be built into a device that people who are most at risk can wear continuously. The device would then alert emergency services in real-time when potentially deadly heartbeat patterns are detected.
Check out the video below, it’s basically an interview with the two Stanford researchers and gives a pretty good insight to the incredible work that they’re doing to detect abnormalities in a persons heart. Incredible stuff!