Shape, Appearance and Spatial Relationships

• Published in: Guide to Medical Image Analysis pp. 405 - 472
• Main Author: Toennies, Klaus D
• Format: Book Chapter
• Language: English
• Published: United Kingdom Springer London, Limited 2017; Springer London
• Series: Advances in Computer Vision and Pattern Recognition
• Subjects: Medial Axis; Model Instance; Object Instance; Shape Model; Shape Variation
• ISBN: 9781447173182; 144717318X
• ISSN: 2191-6586; 2191-6594
• DOI: 10.1007/978-1-4471-7320-5_11

Object detection in medical image analysis can be modelled as a search for an object model in the image. The model describes attributes such as shape and appearance of the object. The search consists of fitting instances of the model to the data. A quality-of-fit measure determines whether one or several objects have been found. Generating the model for a structure of interest can be difficult. It has to include knowledge about acceptable variation of attributes within an object class while remaining suitably discriminative. Several techniques to generate and use object models will be presented in this chapter. Information about acceptable object variation in these models is either generated from training or is introduced via modeling. In either case, an efficient representation is needed with (relatively) few parameters yet being able to represent variation of shape and appearance between subjects. Applying shape (and appearance) models to the data may produce the object segmentation or they may be used as additional constraint in a subsequent segmentation process. This chapter closes with a discussion on how shape models can be used to augment data-driven segmentation by a shape model component.