ABSTRACT
Machine learning is an extremely powerful tool for identifying and extracting important information from medical images. While simple in concept, there are many important steps required to ensure that results are reliable and robust. The data must be prepared and normalized in most cases, and ensuring that not all data is used to optimize the machine learning process is critical. Traditionally, it was necessary for humans to engineer and select the best features to be used by a machine learning algorithm. Deep learning methods now can also perform the feature selection step which both reduces the amount human labor required and may help to find features that are not expected. While some may view machine learning as a ‘black box’ it is possible to extract these learned features from the system, promising new discoveries about the information present in medical images.
