A Combined Manifold Learning Analysis of Shape and Appearance to Characterize Neonatal Brain Development

Abstract

Large medical image datasets form a rich source of anatomical descriptions for research into pathology and clinical biomarkers. Many features may be extracted from data such as MR images to provide, through manifold learning methods, new representations of the population's anatomy. However, the ability of any individual feature to fully capture all aspects morphology is limited. We propose a framework for deriving a representation from multiple features or measures which can be chosen to suit the application and are processed using separate manifold-learning steps. The results are then combined to give a single set of embedding coordinates for the data. We illustrate the framework in a population study of neonatal brain MR images and show how consistent representations, correlating well with clinical data, are given by measures of shape and of appearance. These particular measures were chosen as the developing neonatal brain undergoes rapid changes in shape and MR appearance and were derived from extracted cortical surfaces, non-rigid deformations and image similarities. Combined single embeddings show improved correlations demonstrating their benefit for further studies such as identifying patterns in the trajectories of brain development. The results also suggest a lasting effect of age at birth on brain morphology, coinciding with previous clinical studies.

Learning clinically useful information from images: Past, present and future

“…An alternative, powerful approach is to model the shape and appearance of the anatomy via manifold learning. We have shown that such manifold modelling can characterise the typical trajectories of brain development in neonates [Aljabar et al (2011)] as well as capture useful information for modelling disease progression [Wolz et al (2012)]. This also provides good examples of the more recent trends that aim for a close integration of machine learning into the design of image analysis algorithms which will be discussed in more detail in the next section.…”

Hierarchical Manifold Learning

“…This method requires calculating pairwise similarities between all images for each patch at each level -but at each level this has the same number of operations as a single level manifold embedding of the whole image. The linear equation [2] is solved for every patch with the same complexity. The algorithm can therefore be applied at fine scales of large sets of 3D images.…”

“…In recent years, the use of manifold learning has become increasingly widespread in medical imaging, being used to uncover underlying structure both within a subject [14] [17], and across populations [2] [9][11] [15]. Manifold learning techniques aim to discover the intrinsic dimensionality of data: a low-dimensional embedding which retains local structure of the data.…”

Learning and Combining Image Similarities for Neonatal Brain Population Studies

“…The manifold structure of brain images has been estimated in [1] based on pairwise non-rigid transformations, whereas in [2] similarities were derived from overlaps of their structural segmentations. In [3], shape and appearance information was combined in a joint embedding for an improved characterization of brain development and in [4] clinical information was incorporated into the embedding construction. However, in general it is not clear how to combine and weight multiple features.…”