Loss functions
The loss functions (also called Cost functions) quantify the difference between the expected output and the actual one obtained from the training phase. The all process consists to calculate all the parameters in order to minimize the loss function. The loss functions can differ based on the type of learning task we are dealing with such as regression or classification tasks. The most common ones are the Mean square error, quadratic loss, hinge loss, or entropy loss.
Deep Learning
The main characteristic of deep learning methods is their ability to learn features while performing a task. In the medical field, the most common method is convolutional neural networks (CNNs). CNN is a powerful method to learn useful representations of images and structured data. Indeed, this particular ANNs is built in order to preserve the spatial relationship in the data with very few connections between the layers. A CNN has multiple layers of convolutions and activations, often interspersed with pooling layer, and is trained using backpropagation and gradient descent, and finally fully connected layers at the end to compute the output layer.