In this paper, we present a set of techniques for the evaluation of brain tissue classifiers on a large data set of MR images of the head. Due to the difficulty of establishing a gold standard for this type of data, we focus our attention on methods which do not require a ground truth, but instead rely on a common agreement principle. Three different techniques are presented: the Williams’ index, a measure of common agreement; STAPLE, an Expectation Maximization algorithm which simultaneously estimates performance parameters and constructs an estimated reference standard; and Multidimensional Scaling, a visualization technique to explore similarity data. We apply these different evaluation methodologies to a set of eleven different segmentation algorithms on forty MR images. We then validate our evaluation pipeline by building a ground truth based on human expert tracings. The evaluations with and without a ground truth are compared. Our findings show that comparing classifiers without a gold standard can provide a lot of interesting information. In particular, outliers can be easily detected, strongly consistent or highly variable techniques can be readily discriminated, and the overall similarity between different techniques can be assessed. On the other hand, we also find that some information present in the expert segmentations is not captured by the automatic classifiers, suggesting that common agreement alone may not be sufficient for a precise performance evaluation of brain tissue classifiers.

Regional investigations of newborn MRI are important to understand the appearance and consequences of early brain injury. Previously, regionalization in neonates has been achieved with a Talairach parcellation, using internal landmarks of the brain. Non-synostotic dolichocephaly defines a bi-temporal narrowing of the preterm infant’s head caused by pressure on the immature skull. The impact of dolichocephaly on brain shape and regional brain shift, which may compromise the validity of the parcellation scheme, has not yet been investigated. Twenty-four preterm and 20 fullterm infants were scanned at term equivalent. Skull shapes were investigated by cephalometric measurements and population registration. Brain tissue volumes were calculated to rule out brain injury underlying skull shape differences. The position of Talairach landmarks was evaluated. Cortical structures were segmented to determine a positional shift between both groups. The preterm group displayed dolichocephalic head shapes and had similar brain volumes compared to the mesocephalic fullterm group. In preterm infants, Talairach landmarks were consistently positioned relative to each other and to the skull base, but were displaced with regard to the calvarium. The frontal and superior region was enlarged; central and temporal gyri and sulci were shifted comparing preterm and fullterm infants. We found that, in healthy preterm infants, dolichocephaly led to a shift of cortical structures, but did not influence deep brain structures. We concluded that the validity of a Talairach parcellation scheme is compromised and may lead to a miscalculation of regional brain volumes and inconsistent parcel contents when comparing infant populations with divergent head shapes.

We describe a new approach for estimating the posterior probability of tissue labels. Conventional likelihood models are combined with a curve length prior on boundaries, and an approximate posterior distribution on labels is sought via the Mean Field approach. Optimizing the resulting estimator by gradient descent leads to a level set style algorithm where the level set functions are the logarithm-of-odds encoding of the posterior label probabilities in an unconstrained linear vector space. Applications with more than two labels are easily accommodated. The label assignment is accomplished by the Maximum A Posteriori rule, so there are no problems of "overlap" or "vacuum". We test the method on synthetic images with additive noise. In addition, we segment a magnetic resonance scan into the major brain compartments and subcortical structures.

Since the introduction of diffusion weighted imaging (DWI) as a method for examining neural connectivity, its accuracy has not been formally evaluated. In this study, we directly compared connections that were visualized using injected neural tract tracers (WGA-HRP) with those obtained using in-vivo diffusion tensor imaging (DTI) tractography. First, we injected the tracer at multiple sites in the brain of a macaque monkey; second, we reconstructed the histological sections of the labeled fiber tracts in 3D; third, we segmented and registered the fibers (somatosensory and motor tracts) with the anatomical in-vivo MRI from the same animal; and last, we conducted fiber tracing along the same pathways on the DTI data using a classical diffusion tracing technique with the injection sites as seeds. To evaluate the performance of DTI fiber tracing, we compared the fibers derived from the DTI tractography with those segmented from the histology. We also studied the influence of the parameters controlling the tractography by comparing Dice superimposition coefficients between histology and DTI segmentations. While there was generally good visual agreement between the two methods, our quantitative comparisons reveal certain limitations of DTI tractography, particularly for regions at remote locations from seeds. We have thus demonstrated the importance of appropriate settings for realistic tractography results.

We present a novel method of statistical surface-based morphometry based on the use of non-parametric permutation tests and a spherical wavelet (SWC) shape representation. As an application, we analyze two brain structures, the caudate nucleus and the hippocampus, and compare the results obtained to shape analysis using a sampled point representation. Our results show that the SWC representation indicates new areas of significance preserved under the FDR correction for both the left caudate nucleus and left hippocampus. Additionally, the spherical wavelet representation provides a natural way to interpret the significance results in terms of scale in addition to knowing the spatial location of the regions.