Article Text
Statistics from Altmetric.com
Learning objectives
Understand the paradigm of machine learning.
Understand the differences between supervised and unsupervised learning.
Understand the design principles behind convolutional neural networks, and why they excel at medical image analysis.
Introduction
Machine learning (ML) is a revolution in computer science and is set to change the face of cardiology practice. In ML, humans no longer need to convert an understanding of a problem into a stepwise algorithmic solution; instead, the computer learns to solve a task for itself.
While ML can seem intimidating, the underlying principles build on familiar and established techniques. The recent revolution that made ML so effective, however, was the recognition that numerous sequential layers of simple arithmetic, termed neural networks, become surprisingly effective at solving difficult problems. This ‘deep learning’ has been startlingly effective across a variety of problems, and a particular type, the convolutional neural network (CNN) has revolutionised image analysis.
CNNs are inspired by the human visual cortex and have been used successfully in cardiology to process data that are one-dimensional (1D) (ECGs, pressure waveforms), two-dimensional (2D) (X-rays, MRIs) and three-dimensional (3D) (echocardiography videos, cardiac magnetic resonance cine videos and CT volumes). We are now entering the stage where these CNNs’ performances are starting to equal that of cardiologists in some domains.1 2
In this review, we will cover the basics of ML, before explaining the workings of neural networks, and particularly CNNs. ML will play an increasing role in medical practice and, as with any diagnostic test or piece of medical equipment, an understanding of these systems will better equip medical staff to interpret these systems’ results.
ML at its most simple
The first chapter in an ML textbook is often made up of topics that a decade ago would have been called ‘statistics’. A simple example that we commonly encounter in cardiology is the formula for predicting …
Footnotes
Twitter @DrJHoward, @profdfrancis
Contributors JPH and DPF conceived, drafted and revised the work and have given final approval for the version to the published.
Funding JPH is supported by the Wellcome Trust (212183/Z/18/Z).
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Commissioned; externally peer reviewed.
Author note References which include a * are considered to be key references.
Request Permissions
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.