Supervised learning (machine learning)

A convolution neural network is a form of machine learning that exclusively processes array data such as medical images ³.They follow the traditional supervised learning hierarchy whereby analysis and retrospective adjustment of significant sources of training data is its primary purpose ⁴.

The algorithm is 'multilayered' otherwise known as a 'deep neural network' ^5-7

1. convolution layer
   • the algorithm is provided in input (region of interest)
   • input undergoes feature-based, convolution filters (kernels) creating a feature map
2. pooling layer
   • feature maps are downsized to a smaller matrix
   • preserves important information by maintaining a maximal pixel value from each filtered set
3. rectified linear unit
   • during convolution and pooling processes. Some pixels in the matrix may possess negative values
   • the rectified being a unit ensures all negative values as at a zero
4. fully connected layer
   • each layer will feed into a fully connected layer that presents each separate filtered image as a vote
   • each 'vote' possesses a weight in determining the category of the region of interest

As a region of interest is any known a convolution neural network will undergo backpropagation to adjust the weights of each vote each time a classification is incorrect, otherwise known as the training phase of supervised learning ⁴. 

Convolution neural networks will then be subjected to a 'validated training set' to determine if the training phase was successful.

​