A primer on deep learning and convolutional neural networks for clinicians

Abstract

Deep learning is nowadays at the forefront of artificial intelligence. More precisely, the use of convolutional neural networks has drastically improved the learning capabilities of computer vision applications, being able to directly consider raw data without any prior feature extraction. Advanced methods in the machine learning field, such as adaptive momentum algorithms or dropout regularization, have dramatically improved the convolutional neural networks predicting ability, outperforming that of conventional fully connected neural networks. This work summarizes, in an intended didactic way, the main aspects of these cutting-edge techniques from a medical imaging perspective.

Keywords: Deep learning; Educational; Image processing; Medical imaging.

Fig. 1

Classical programming vs supervised learning approach

Fig. 2

Matrix representation of a 8-bits black and white image. The values range from 0 to 255

Fig. 3

The three color channels (red, green, blue) of a retina image for diabetic retinopathy detection

Fig. 4

Solving a two-dimensional classification problem by eye by changing the data representation

Fig. 5

Structure of a single neuron in an artificial neural network

Fig. 6

Sigmoid function

Fig. 7

Forward propagation of a single neuron

Fig. 8

Minimization of a parable. The random W value at point 1 presents a negative slope while at point 2 presents a positive slope

Fig. 9

Derivative calculation using the chain rule and update of the W and b parameters in the direction of the minimum

Fig. 10

Neural network with 5 hidden layers

Fig. 11

ReLU and leaky ReLU activation functions

Fig. 12

General convolutional neural network architecture

Fig. 13

Pixel representation of a vertical line filter (left) and its visualization (right)

Fig. 14

Extraction of features from the original image by using sliding filters

Fig. 15

Max-pooling applied to a feature map, reducing its dimensionality