User Tools

Site Tools


miccai

!!!Evaluation of Appearance Models of the Brain

!!Abstract

Appearance models are an applicable approach to the analysis of anatomical variability. They are able to distinguish between groups, e.g. normal and diseased, as a model encapsulates the properties of a group from which it was derived. The construction of such models is closely-related to the task of registration and they require the correspondence, which registration is able to obtain.

We developed a framework which evaluates appearance models, based on the statistics of large sets of images. The framework is capable of distinguishing between good models of the brain and worse ones. Furthermore, it provides a method of validating models. It does so by measuring how well a model and its data fit together.

Two measures are defined which reflect on the quality of a model. The first of these – specificity – approximates the level to which data generated by the model fits data from which the model was constructed. The complementary measure – generalisation – is able to quantify 'distance' between data from which the model was constructed and model-generated data.

Results show that as models degrade in quality, their specificity and generalisation ability rise, as expected. The algorithms are used to compare models of the brains, which were built automatically by independent approaches of registration. This greatly helps in identifying better model construction algorithms, which are analogous to registration algorithms.

!!1 Introduction

A powerful method for the modelling of anatomy was introduced by Edwards et al. [Edwards] and it is known as appearance models – a natural successor to shape models [Cootes]. This method requires a large enough set of data, which is representative of a population and ideally spans its full variability. Appearance models are able to learn what characterises inter-subject changes and determine the prominence of the main characteristics. Hence, it is able to identify changes and derive a model that encapsulates change – all in a data-driven manner.

Non-rigid image registration is ubiquitously used as the basis for analysis of medical images. The results of registration can be used for structural analysis, atlas matching, and analysis of change. Methods for obtaining registration are are well-established and quite uniform. The goal is achieved by warping pairs of images so that they appear more similar. The similarity leads to overlap, which allows corresponding structures to be identified. This problem is complementary to that of modelling groups of images. A statistical model of a group of images needs dense correspondence to be defined across the group; non-rigid registration provides exactly that.

Since the emergence of appearance models, attempts have been made to reproduce and improve it. To name a few such efforts, Stegmann [Stegmann] built 4-dimensional cardiac models and Reuckert et al. [Rueckert] derived statistical deformation models from several registrations of the brain. Models have been built in a variety of ways, but what is yet lacked is the ability to compare them. It becomes clear from experience that attempts to distinguish between them by eyesight is hopeless. More recently, appearance models were built automatically using piece-wise affine registration [IPMI - YET TO ADD]. Evaluation of models in this particular case enables evaluation of registration algorithms.

The idea of evaluating models is not unexampled. Davies et al. [Davies] explored the evaluation of shape models and ultimately developed a robust framework. This paper outlines a principled approach to the evaluation of appearance models, which is a challenging task since their complexity is very high. The approach is shown to be reliable in evaluation of brain models (FOOTER: Examples from non-medical domains are beyond the remit of this paper, but they have been very successful.) and it is then used to learn about registration algorithms, from which appearance models have been derived.

miccai.txt · Last modified: 2014/05/31 17:33 by admin