Supervised learning (machine learning)

Supervised learning is the most common type of machine learning algorithm used in medical imaging research. It involves training an algorithm from a set of images or data where the output labels are already known ¹.

Supervised learning is broken into two subcategories, classification and regression ². Classification refers to the prediction of whether an image falls into one or more categories while regression aims to predict a continuous label. Medical imaging machine learning deals mainly with the classification of images.

Fundamentally, the optimisation of supervised machine learning consists of two phases that are looped through continuously ³:

• forward propagation: for a given input, calculates a predicted outcome and compares this with the expected outcome to give an overall cost (error)
• backward propagation: from the cost, works backwards through the network updating the parameters in an attempt to minimise the overall cost

Radiology exampleSupervised learning in radiology 

• simplified, a given input, (such as a lung nodule ROI) calculates the predicted outcome over the expected outcome once the cost is estimated the network will update the individual weighting of the output parameters (think of it as a vote)
• the output parameters that correctly identified the nodule to the expected outcome are awarded a heavier 'vote'
• the output parameters that did not correctly identified the nodule to the expected outcome are adjusted to minimise cost