Image descriptors are widely used in applications such as object recognition, pattern classification and image registration. The descriptors encode the local visual content of the image to provide a compact, robust and distinctive representation of objects. If images differ in orientation, descriptors must be rotation invariant. This paper introduces a compact rotation invariant descriptor. The approach is based on the representation of the local visual content by a graph. A function living on the graph vertices is evaluated and transformed through spectral trimming. This transform is rotation invariant and reduces the dimensionality of the descriptor. The performance of the introduced descriptor is as good as the SIFT descriptor performance, while being about ten times more compact, as shown by experiments on transmission electron microscope images.

The Brain Connectome project is a multi-institution project aimed at creating a graph of all connections between brain cells to better understand the brain circuitry and to explore the causes of neurodegenerative diseases [1]–[3]. Today’s challenge for neurobiologists is to segment the volumetric datasets of the brain imaged using electron microscopy [15]– [21]. Although an expert-supervised dense segmentation of these multi-terabyte datasets is desirable, a sparse and automated segmentation representing a tractable number of anatomical shapes would provide a huge leap in terms of speed of scientific discovery. Observing the multi-scale nature of anatomical shapes, such as large nuclei and elongated dendrites [22], [23], we designed a multi-scale segmentation architecture that offers both the scalability to analyze massive datasets and the usability required in an interdisciplinary research environment. We detail the methodology that has led to our computational architecture and report our first results on our 19-Terabyte 3D dataset of the visual cortex.

Several groups in neurobiology have embarked into deciphering the brain circuitry using large-scale imaging of a mouse brain and manual tracing of the connections between neurons. Creating a graph of the brain circuitry, also called a connectome, could have a huge impact on the understanding of neurodegenerative diseases such as Alzheimer's disease. Although considerably smaller than a human brain, a mouse brain already exhibits one billion connections and manually tracing the connectome of a mouse brain can only be achieved partially. This paper proposes to scale up the tracing by using automated image segmentation and a parallel computing approach designed for domain experts. We explain the design decisions behind our parallel approach and we present our results for the segmentation of the vasculature and the cell nuclei, which have been obtained without any manual intervention.

Many neuroanatomy studies rely on brain tissue segmentation in Magnetic Resonance images (MRI). The Expectation-Maximization (EM) theory offers a popular framework for this task. We studied the EM algorithm developed at the Surgical Planning Laboratory (SPL) at Harvard Medical School and implemented in the Slicer3 software. We observed that the segmentation lacks accuracy if the image exhibits some intensity inhomogeneity. Moreover the optimum parameters are challenging to estimate. This document aims at describing our solutions within the context of statistical modeling. Our contributions range from algorithm improvements to novel representations of the statistical distribution model. First we added a bias field correction module and exposed the most significant parameters. Second we proposed a new way to select the distribution of the tissues to be segmented. Finally we designed a set of interactive tools to make the segmentation process easier and more accurate. To validate the new segmentation pipeline, we performed our experiments on MRI data and a clinical expert evaluated our results.The source code developed for this work, i.e. the code of the MRIBiasFieldCorrection and EMSegment modules, is part of Slicer3 version 3.5 and can be downloaded by this command: svn co http://svn.slicer.org/Slicer3/trunk Slicer3

Isosurface extraction from brain images often creates handles between brain folds that are anatomically separated. Although manual editing could optimally correct these anatomical errors, it is not realistic due the size of the 3D data and the convoluted geometry of the brain. We propose an algorithm to automatically repair the isosurface and we make our code available at http://www.OpenTopology.org

Virtual cystoscopy is a developing technique for bladder cancer screening. In a conventional cystoscopy, an optical probe is inserted into the bladder and an expert reviews the appearance of the bladder wall. Physical limitations of the probe place restrictions on the examination of the bladder wall. In virtual cystoscopy, a computed tomography (CT) scan of the bladder is acquired and an expert reviews the appearance of the bladder wall as shown by the CT. The task of identifying tumors in the bladder wall has often been done without extensive computational aid to the expert. We have developed an image processing algorithm that aids the expert in the detection of bladder tumors. Compared with an expert observer reading the CT, our algorithm achieves 89% sensitivity, 88% specificity, 48% positive predictive value, and 98% negative predictive value.

We propose to match a labeled mesh onto the patient brain surface in a multiresolution way for labeling the patient brain. Labeling the patient brain surface provides a map of the brain folds where the neuroradiologist and the neurosurgeon can easily track the features of interest. Due to the complexity of the cortical surface, this task usually depends on the intervention of an expert, and is time-consuming. Our multiresolution representation for the brain surface allows the automated classification of the folds based on their size. The atlas mesh is deformed from coarse to fine to robustly capture the patient brain folds from the largest to the smallest. Once the atlas mesh matches the patient mesh, the atlas labels are transferred to the patient mesh, and color coded for visualization.

Common problems in medical image analysis involve surface-based registration. The applications range from atlas matching to tracking an object's boundary in an image sequence, or segmenting anatomical structures out of images. Most proposed solutions are based on deformable surface algorithms. The main problem of such methods is that the local accuracy of the matching must often be traded off against global smoothness of the surface in order to reach global convergence of the deformation process. Our contribution is to first build a Multi-Resolution (M-R) surface from a reference segmented image, and then match this surface onto the target image in an M-R fashion using a deformable surface-like algorithm. As we proceed from lower to higher resolution, the smoothing effect of the deformable surface is more and more localized, and the surface gets closer and closer to the target boundary. We present initial results of our algorithm for atlas registration onto brain MRI showing improved convergence and accuracy over classical deformable surface methods.

We present a method based on multiresolution signal processing on meshes to create a thickness atlas. We applied this method to construct an atlas of bladder wall thickness. Bladder cancer is associated with increased bladder wall thickness. A thickness atlas helps to detect abnormal thickening in the bladder wall. Extracting inner and outer surface meshes from segmented images, we compute the thickness on the inner surface and map it to a sphere. We average the thickness at each position on the sphere to create a thickness atlas. We then compute Z-score values on the configuration of the patient's bladder to show regions of unusual thickness.