ABSTRACT
Automated computer analysis of medical images to detect patterns and tissue structures and to distinguish between various diseases is a challenging task. Three main application purposes of medical image analysis include tissue detection, tissue segmentation, and medical image classification. Machine learning algorithms have been employed to accomplish this analysis via computation. The ultimate goal of applying computation to this analysis is to achieve more accurate patient outcomes than with manual analysis, especially when evaluating hundreds and even thousands of patient imaging records. However, accurately labeling training data can be time consuming and nearly impossible to achieve in the medical domain. Further, collecting a sufficient number of samples for model training is a major limitation for machine learning and is especially acute for patient medical images. These limitations prevent proposing a robust and general machine learning model for medical images. Recently, deep learning models have garnered interest compared to traditional machine learning algorithms, as computing resources such as Graphical Processing Units (GPUs) have increased in performance. Deep learning has shown high accuracy with large datasets, especially when applied to medical images. This chapter provides a brief overview of machine learning and deep learning methods, focusing on their application to medical image analysis. In Section 10.1, the basics of machine learning and deep learning algorithms are reviewed. In Section 10.2, challenges in developing these algorithms are discussed. In Section 10.3, common deep learning methods employed with medical images are discussed.
