心律不齐 z-bo 2020-11-02 375

　　Recently, scientists at the Skoltech Institute have devised a new machine-learning-based method for detecting "atrial fibrillation drivers," small plaques in the heart muscle that are thought to cause the most common type of arrhythmia. According to the American Heart Association, this approach could lead to more effective targeted medical interventions to treat diseases that are estimated to affect 33 million people worldwide.

　　The underlying mechanism of atrial fibrillation (AF) is not yet clear. AF is an abnormal heart rhythm, which is associated with an increased risk of heart failure and stroke. Studies have shown that it may be driven by reentrant atrial fibrillation and cause repeated attacks of atrial fibrillation, which is a highly local cause of repetitive arrhythmia. At present, this symptom can be treated by surgery to alleviate the condition and even restore the normal function of the heart.

　　In order to find these reentrant AF drives for subsequent processing, doctors used multi-electrode mapping technology, which allows them to record multiple electrograms (done through a catheter) inside the heart and establish an electrical activity map in the atrium. However, the clinical application of this technology usually produces many false negatives when no existing AF driver is found, and many false positives when it is detected that there is no driver.

　　Recently, researchers have used machine learning algorithms to interpret ECG to look for atrial fibrillation. However, these algorithms require marked data with the true position of the driver, and the accuracy of multi-electrode mapping is insufficient. The authors of this new study were led by Dmitry Dylov of the Skoltech Center for Computing and Data Intensive Science and Engineering (CDISE) and Vadim Fedorov of Ohio State University. They used high-resolution near-infrared optical mapping (NIOM) to find automatic Focus the drive and stick to it as a training reference.

　　"NIOM is based on the penetrating infrared light signal, so it can record the electrical activity in the myocardium, while the traditional clinical electrode can only measure the signal on the surface. In addition, this feature also has excellent optical resolution and optical mapping if you I want to visualize and understand the propagation of electrical signals in the heart tissue." Dmitry Dylov said.

　　The team tested their method on eleven hearts that were implanted in humans, all of which were donated after death for research purposes. The researchers also performed optical and multi-electrode mapping of AF attacks induced in the heart. The ML model can indeed effectively interpret the electrogram from the multi-electrode mapping to locate the AF driver with an accuracy of up to 81%. They believe that the larger training data set validated by NIOM can improve the algorithm based on machine learning enough to make it a complementary tool in clinical practice.