Learning objectives
To review convolutional neural networks (CNNs) as an example of machine learning.
To review CNNs applications in imaging
Background
Machine learning is a branch of artificial intelligence (AI) that is able to extrapolate meaningful patterns from datasets to train algorithms such as convolution neural networks (CNNs).1 The objective of machine learning is to generate algorithms with the ability to learn without being explicitly programmed. Convolutional neural networks are a subcategory in the hierarchic terminology that includes artificial intelligence, machine learning, and deep learning.2 A single layer artificial neural network is termed a ‘perceptron’ and a multi-layer perceptron is termed ‘neural network’. The artificial neuron...
Imaging findings OR Procedure details
Convolutional neural networks are modelled off the biology of neurons (nodes) and synapses (connections).7, 8 They are derived from artificial neural networks, but are differentiated by its definitive assumption that the inputs are images. There are three main applications of CNNs which include segmentation, classification and detection. There are typically four main types of layers that make up a CNN: convolution (CONV), pooling (pool), nonlinearity (rectified linear unit (ReLU)) and a fully connected (FC) layer. During segmentation, for example, the CNN architecture comprises of multiple...
Conclusion
The CNN is a machine learning algorithm based on the biology of the human visual cortex. The main research applications of these machine learning models are in classification, segmentation and detection. Machine learning models such as CNNs will be an important means of augmenting the clinical diagnosis for radiologists.
References
Erickson B, Korfiatis P, Akkus Z, Kline T. Machine Learning for Medical Imaging. Radiographics 2017;37(2):505-15.
Soffer, A. Ben-Cohen, O. Shimon, MM Amitai, et al. Convolutional Neural Networks for Radiologic Images: A Radiologist’s Guide. Radiology. 2019;290(3):590-606. 10.1148/radiol.2018180547.
Castro W, Oblitas J, Santa-Cruz R, Avila-George H. Multilayer perceptron architecture optimization using parallel computing techniques. PLoS One. 2017;12(12):e0189369.
Anwar SM, Majid M, Qayyum A, Awais M, Alnowami M, Khan MK. Medical Image Analysis using Convolutional Neural Networks: A Review. J Med Syst. 2018;42(11):226.
Kuzovkin I, Vicente R, Petton...