Cardiovascular diseases: 1st cause of death in the world
A cardiac arrhythmia is a problem due to an abnormal functioning of the cardiac electricity. The heart beats normally, between 50 and 80 times a minute, at rest.
An arrhythmia can slow down the heartbeat (bradycardia, less than 50 beats per minute), speed it up (tachycardia, more than 100 beats per minute) or cause an irregular rhythm (atrial fibrillation, extrasystole).
The most common causes of non-congenital heart arrhythmias are: heart attack (acute or past), heart failure, thyroid disease, myocarditis, high blood pressure or cardiomyopathy. Certain types of arrhythmias increase the risk of stroke.
Cardiovascular disease remains the leading cause of death worldwide. In 2011, coronary heart disease and stroke killed the most people: 13.2 million people, 24% of the 54.6 million deaths in the year.
Neural networks as efficient as cardiologists
To detect cardiac arrhythmias, scientists have traditionally used algorithms that detect peaks and intervals between peaks based on ECG recordings (signal processing).
The Stanford Machine Learning Group and iRhythm Technologies developed a new approach using a 34-layer Convolutional Neural Network (CNN) trained on data captured by a Zio Monitor device.
The results of the model developed show that a deep learning approach allows the classification of a wide range of distinct arrhythmias from ECGs with high diagnostic performance, similar to those of cardiologists.
1-D Convolutional Networks
The neural network built by Stanford researchers is based on a series of 1-D Convolutional Networks (CNN 1D).
The 1D CNN have superior performance for analyzing highly fluctuating signals such as a patient’s ECG.
The 1D CNN has achieved peak performance levels in detecting and identifying anomalies in power electronics and detecting engine failures.
Performing an ECG with an Apple Watch
From 2017 to 2019, Stanford University conducted the Apple Heart Study. More than 419,000 participants voluntarily provided their cardiac data to help study arrhythmia. This study is an extension of the research undertaken with the Zio Monitor by Stanford researchers.
As a result of this study, Apple obtained authorization to market an application that allows an ECG to be performed with the Apple watch. This is the first application, intended for the general public, of artificial intelligence to carry out medical diagnoses.
The collaboration between Stanford and Apple does not stop there, Stanford has launched a study to determine whether Apple Watch can be used to detect COVID-19 before the first symptoms appear. This project demonstrates the potential of Artificial Intelligence coupled with data-collecting connected clothing, the Wearables.
This article was written from these resources…
MIT 6.S897 Machine Learning for Healthcare, Spring 2020- Lesson 6 Phylosiogical Time-series
