ABSTRACT
Interpretation of medical imaging is a key component of workflows for patient care. One challenge is that human interpretation is often qualitative and subject to variability. Radiomics addresses this by allowing for quantitative assessment of disease using features that capture imaging characteristics of pathological lesions such as intensity, shape, and texture. The predictive capacity of the numerous features extracted are synergistically combined using machine learning techniques to create a highly predictive model. Numerous studies have used radiomics to predict clinical variables such as genomics, treatment response, and survival. In the era of “precision medicine”, radiomics has the potential to aid tailoring of medical care to each individual patient. In this chapter, we review the key components of radiomic pipelines: lesion segmentation, feature extraction, feature selection, classification, and prediction. We also discuss the challenge of repeatability and reproducibility as well as standardization efforts. To facilitate the development of radiomic pipelines, we provide an overview of publicly available datasets and toolkits. Lastly, we discuss the next generation of machine learning algorithms for medical imaging with the advent of convolutional neural networks.
