ABSTRACT
Multi-atlas segmentation is a general-purpose segmentation paradigm whereby multiple existing segmentations (atlases) are mapped onto a query image, and then merged into a single segmentation. Early advances in this technique were driven by neuroscience researchers in the mid 2000s, with applications in radiotherapy being advanced in the late 2000s. Despite recent advances in deep learning segmentation methods, multi-atlas segmentation remains relevant due to its generality and flexibility. This chapter introduces multi-atlas segmentation and provides an overview of its major algorithmic components: database creation, atlas selection, image registration, label fusion, and post-processing. A brief summary of the chapters within this part of the book is also provided.
