The cardiogram will be decoded by a computer
Artificial intelligence analyzes ECG better than cardiologists
Julia Korowski, XX2 century, based on Stanford News Service: Stanford computer scientists develop an algorithm that diagnoses heart arrhythmias with cardiologist-level accuracy
Scientists from Stanford University have developed a neural network that recognizes cardiac arrhythmias more accurately than cardiologists. To do this, the algorithm was fed ECG data from almost 30,000 patients – this sample is 500 times larger than in the largest studies known at the moment. A preprint of the work is published in the arXiv repository (Rajpurkar et al., Cardiologist-Level Arrhythmia Detection with Convolutional Neural Networks).
People suspected of having an arrhythmia are usually referred to the hospital for an ECG. But if the ECG does not reveal problems, the doctor can give the patient a wearable heart monitor and send him home to do normal things while the device monitors the heart. As a rule, the so–called "long-term ECG registration" lasts up to 7 days, but relatively recently, monitors have appeared that allow monitoring twice as long - two weeks. These include Zio Patch from the company iRhythm – it was used in the course of a new study.
For 14 days, while recording continues, the device accumulates hundreds of hours of data, which are then analyzed almost every second. This allows you to identify dangerous arrhythmias, some of which are very difficult to distinguish from harmless deviations.
Stanford researchers looked at the diagnosis of arrhythmia from the point of view of a data processing problem. They decided to create an algorithm capable of analyzing ECG signals and recognizing 12 types of rhythm disturbances. The scientists agreed with iRhythm, and the company provided them with the data of 300,000 Zio users, of whom 30,000 people chose. The authors of the work collected 64,000 recordings with a duration of 30 seconds – each piece was analyzed by a specialist who classified ECG segments according to the type of rhythm.
Based on these data, the researchers trained a convolutional neural network. After 7 months, she independently analyzed the ECG and recognized the arrhythmia. In order to evaluate the accuracy of machine diagnostics, the researchers used 336 cardiograms that were not included in the training set. First, they were reviewed by a group of three cardiologists: doctors had to come to a common opinion about what kind of arrhythmia they were dealing with – these assessments became the "gold standard" with which the rest of the results were compared. Then the same thing was assigned to 6 doctors who performed the task independently of each other. It remained to test the neural network and see who came closer to the standard – artificial intelligence or single cardiologists.
"The difference in heartbeat signals can be very subtle, but it has a huge impact on how you treat the patient," explains one of the authors of the work Pranav Rajpurkar. "For example, two forms of arrhythmia, known as atrioventricular block (AVB) of the first and second degree, look very similar, but there is no need to treat one, and the second requires immediate intervention." It turned out that in many cases the neural network gives a more accurate diagnosis than doctors, and this is most clearly manifested in the case of AVB. "One of the important features of this work, in my opinion, is that we do not just recognize anomalies – we recognize anomalies of various types with high accuracy," says Awni Hannun, a member of the scientific group. "You definitely won't find that level of accuracy anywhere else."
Artificial intelligence also sometimes made mistakes, confusing the types of arrhythmias, but in addition to high accuracy, it had other important advantages – high speed and complete absence of fatigue. The authors note that the neural network does not recognize many heart diseases, including myocardial infarction, but it can be improved in the course of further research. They hope that in the future algorithms will facilitate the work of cardiologists or even replace them altogether in remote areas where there is no doctor, and wearable devices are available.
Portal "Eternal youth" http://vechnayamolodost.ru 11.07.2017