A primer in artificial intelligence in cardiovascular medicine

Abstract

Driven by recent developments in computational power, algorithms and web-based storage resources, machine learning (ML)-based artificial intelligence (AI) has quickly gained ground as the solution for many technological and societal challenges. AI education has become very popular and is oversubscribed at Dutch universities. Major investments were made in 2018 to develop and build the first AI-driven hospitals to improve patient care and reduce healthcare costs. AI has the potential to greatly enhance traditional statistical analyses in many domains and has been demonstrated to allow the discovery of 'hidden' information in highly complex datasets. As such, AI can also be of significant value in the diagnosis and treatment of cardiovascular disease, and the first applications of AI in the cardiovascular field are promising. However, many professionals in the cardiovascular field involved in patient care, education or science are unaware of the basics behind AI and the existing and expected applications in their field. In this review, we aim to introduce the broad cardiovascular community to the basics of modern ML-based AI and explain several of the commonly used algorithms. We also summarise their initial and future applications relevant to the cardiovascular field.

Keywords: Artificial intelligence; Artificial neural networks; Cardiovascular disease; Deep learning; Machine learning.

Fig. 1

a–d Different approaches to data separation in classification problems. For this illustrative purpose, the presented data exist in 2D space. Dimensionality in real datasets and images is much higher, but the same principles apply. a Example of two classes of samples that can be separated linearly. b Two non-linearly separable data classes. c Data separation in a support vector machine. The optimal support vector (green) keeps a balanced optimal distance to all data in both classes to prevent misclassification in added data points. d Example of multiple data clusters. The samples in each cluster belong together, based on their relatively close distances

Fig. 2

Various influences on the development of deep learning. Methods: Starting with Legendre, mathematicians have developed various techniques, some of them specifically aimed at artificial intelligence since the Dartmouth conference. LeCun developed the convolutional neural network (1989), used by Hinton to win the 2012 ImageNet competition. Storage: Maturing over a century, grown significantly after the introduction of the internet, leading to current cloud systems. Computational power: After the introduction of computers during World War II, an exponential growth of computing power was recognised by Moore, continuing until today. Driven by the gaming industry, graphical processing units have given computers the final boost to implement deep learning

Fig. 3

a, b Detection and segmentation of objects. a Detected objects are marked by a coloured rectangle and the identified object type. Segmentations are displayed by a semi-transparent colour overlay. A trained model may be more or less certain about its predictions, as indicated by the percentages alongside the object classifications. b Segmentation and object detection applied to an ECG signal. A trained model can detect separate segments (e.g. ST: magenta, T‑top: yellow) of the ECG and detect abnormal beats, such as a ventricular extrasystole (VES, red)

Fig. 4

a Feedforward network. Input flows from left to right predict output values. b Artificial neuron: each input x_i is multiplied by its own weight w_i. To introduce an intercept, input b (with value 1) is introduced, with its own weight w_b. The sum of all weighed inputs x and intercept b is used as input for the activation function φ that yields an output above a certain threshold. c Examples of activation functions. Tanh is the hyperbolic tangent, with outputs quickly changing from −1 to 1 around input x = 0. The sigmoid resembles tanh, with outputs shifting from 0 to 1 around x = 0. Relu is the rectified linear unit, with output = 0 for inputs smaller than zero and outputs equal to input for inputs greater than or equal to zero. Softplus is a continuous function that only slightly differs from relu for inputs around zero and is described by ln(1 + e^x) d Recurrent neural network, which adds a time component to the feedforward network. In addition to input x_t at a given moment t, the previous output y_t-1 is passed through the model to predict y_t. e Long short-term memory, which adds a persisted model state s in addition to the previous model output y_t–1, to better assess the effect of input changes, over a longer period, on model output

Fig. 5

Schematic presentation of a convolutional neural network. Each convolutional layer develops several filters to detect shapes or objects in the previous layer (red), from basic shapes in the first convolutional layer to highly complex objects in the last layer. The thickness of a layer resembles its filter count. Following each convolutional layer, the data are down-sampled in a pooling layer (blue), which results in a higher level of complexity for the next convolution. After the last pooling layer, all filters are flattened to a single layer that passes some fully connected layers (like a basic feedforward network, Fig. 4a) to finally come to a predicted output, where the predicted output class acronym DCM stands for dilated cardiomyopathy, HCM for hypertrophic cardiomyopathy and RCM for restrictive cardiomyopathy