Pixel-Based Machine Learning in Medical Imaging

Table 2

Classification of ML algorithms by their input, output, and teacher (desired output).

ML algorithms	Input	Output	Teacher
Neural filters	Pixel values in a subregion (local window) in a given image	Single pixel value (image is formed by collecting pixels)	Desired pixel value
MTANNs	Pixel values in a subregion (local window) in a given image	Single pixel value (image is formed by collecting pixels; likelihood score for the given image is obtained by use of the scoring method)	Likelihood of being a specific pattern at each pixel
Convolution NNs	Pixel values in a given image	Class to which the given image belongs	Nominal class label for the given image
Shift-invariant NNs	Pixel values in a given image	Class to which each pixel belongs	Nominal class label for each pixel
Multilayer perceptron for character recognition	Pixel values in a given binary image (character)	Class to which the given image belongs	Nominal class label for the given image
Classifiers (e.g., linear discriminant analysis, NNs, support vector machines)	Features extracted from a segmented object in a given image	Class to which the segmented object belongs	Nominal class label for the segmented object