Very preterm birth coincides with a period of major development in the brain, with striking changes in volume, cortex folding and significant change at the microstructural level. Diffusion MRI is sensitive to motion of water on the scale of microns, allowing us to investigate some of these changes. Mapping of diffusion tensors is a challenging process, and existing methods fail to account for the major changes that take place between 30 and 40 weeks equivalent gestational age. In this paper we introduce the spectral matching in the context of non-linear registration of diffusion images. Spatial correspondences are defined with respect to the main spectral modes of the images, which are global descriptors of the tensor information. We apply tensor spectral matching (TSM) in two different ways: by estimation of spatial correspondences uniquely based on the spectral decomposition of diffusion tensor images, and by combination of TSM with a standard diffusion tensor registration algorithm (TSM-DTI-TK). We validate the proposed approaches on 20 adult controls, and we compare it to the state-of-art registration method. We then apply these methods to longitudinal diffusion data acquired from 6 extremely preterm-born infants scanned shortly after birth and at term equivalent age. The experimental results combining TSM with standard tensor registration outperform the state-of-art when applied to both adult and preterm data. Having a reliable anatomical correspondence in preterm infants allows us to quantify microstructural changes and to work towards developing biomarkers of neurological impairment.

Tensor Spectral Matching of Diffusion Weighted Images

The Gyrification Index (GI) quantifies the amount of folding of a cortical surface. In this paper, we show the efficiency of spectral analysis to perform such a task, and in particular we explain how the graph windowed Fourier transform can be used as a tool to define new GIs with multiscale properties. We propose two different GIs and study their effect on a set of subjects whose cortical surfaces are modeled with triangular meshes.

The graph windowed Fourier transform: a tool to quantify the gyrification of the cerebral cortex

Medical image analyses rely on diverse software packages assembled into a “pipeline”. The Java Image Science Toolkit (JIST) has served as a standalone plugin into the Medical Image Processing Analysis and Visualization (MIPAV). We addressed shortcomings that previously prevented deeper integration of JIST with other E-science platforms. First, we developed an interface for integrating externally compiled packages (similar to the interfaces in NiPy) such that the application can become a “draggable module” in the module tree. This allows for connection of inputs and outputs to other JIST modules while maintaining external processing and monitoring. Second, we develop an integration interface with the Neuroimaging Informatics Tools and Resources Clearinghouse Cloud Environment (NITRC-CE). User can launch and terminate pre-configured nodes to utilize computational resources of the Amazon cloud. Finally, we define a new external data source, which can connect to the eXtensible Neuroimaging Archive Toolkit (XNAT) to query and retrieve remote data using XNAT's REST API. Specifically, we define dataflow for files that can readily be converted into volumes and collections of volumes to interface with any JIST module that expects volumetric image data as an input. Users now have the ability to run their pipelines from a well-defined external data source and no longer are required to already have data on the disk. With these upgrades we have extended JIST’s capabilities outside of complied java source code and enhanced capabilities to seamlessly interface with E-science platforms. 

Integration of the Java Image Science Toolkit with E-Science Platform

A multi-atlas approach is proposed for the automatic segmentation of nine different structures in a set of head and neck CT images for radiotherapy. The approach takes advantage of a training dataset of 25 images to build average head and neck atlases of high-quality. By registering patient images with the atlases at the global level, structures of interest are aligned approximately in space, which allowed multi-atlas-based segmentations and correlation-based label fusion to be performed at the local level in the following steps. Qualitative and quantitative evaluations are performed on a set of 15 testing images. As shown by the results, mandible, brainstem and parotid glands are segmented accurately (mean volume DSC>0.8). The segmentation accuracy for the optic nerves is also improved over previously reported results (mean DSC above 0.61 compared with 0.52 for previous results). 

A Multi-atlas Approach for the Automatic Segmentation of Multiple Structures in Head and Neck CT Images

A new patch based label fusion method based on generative approach is proposed for segmentation of mandible, brainstem, parotid and submandibular glands, optic nerves and the optic chiasm in head and neck CT images. The proposal constructs local classifiers from a dictionary of patches and weights their contribution using a generative probabilistic criterion. Also, a gaussian slide window is used to weight the multiples estimations of neighboring voxels. The proposed method was evaluated on a set of 15 CT images (10 off-site and 5 onsite) provided by the organizers of the Head and neck Auto-Segmentation challenge(MICCAI 2015), where the obtained results are comparable to many of the other methods used in the challenge. 

Head and Neck Auto Segmentation Challenge based on Non-Local Generative Models

We present a fully automatic model based system for segmenting the mandible, parotid and submandibular glands, brainstem, optic nerves and the optic chiasm in CT images, which won the MICCAI 2015 Head and Neck Auto Segmentation Grand Challenge. The method is based on Active Appearance Models (AAM) built from manually segmented examples via a cancer imaging archive provided by the challenge organisers. High quality anatomical correspondences for the models are generated using a Minimum Description Length (MDL) Groupwise Image Registration method. A multi start optimisation scheme is used to robustly match the model to new images. The model has been cross validated on the training data to a good degree of accuracy, and successfully segmented all the test data. 


Fully Automatic Segmentation of Head and Neck Organs using Active Appearance Models

We describe a segmentation method that was used in the Head and Neck Auto Segmentation Challenge held at the MICCAI 2015 conference. The algorithm consists of two building blocks. First, we employ a multi-atlas segmentation to obtain an initial segmentation for the considered organs at risk. Secondly, we use an Active Shape Model (ASM) segmentation to refine the initial segmentation of some of the organs. Leave-one-out experiments with the training data were used to determine suitable parameters for the individual steps of the segmentation. The ASM refinement resulted in improved segmentation for the optic nerves and submandibular glands, while for the brain stem, parotid glands, chiasm, and mandibular bone, the multi-atlas segmentation was preferable. Our submission achieved the second rank in the challenge.

Multi Atlas Segmentation with Active Shape Model Refinement for Multi-Organ Segmentation in Head and Neck Cancer Radiotherapy Planning

The Insight Segmentation and Registration Toolkit (www.itk.org) has become a standard in academia and industry for medical image analysis. In recent years, the ITK developers' community has focused on providing programming interfaces to ITK from Python, Java, and Javascript and making ITK available via leading applications such as Slicer and ImageJ. In this course we present best practices for taking advantage of ITK in your imaging research and commercial products. We demonstrate how script writing and interactive GUIs can be used to access the algorithms in ITK and the multitude of ITK extensions that are freely available on the web. We also cover the opportunities and challenges with using open-source software in research and in commercial applications: from prototypes that can lead to venture capital funding to applications for first-in-human trials and ultimately for regulatory approval.

ITK in Biomedical Research and Commercial Applications

The common approach to do a fully automatic segmentation of multiple structures is an atlas or multi-atlas based solution.
These already have proven to be suitable for the segmentation of structures in the head and neck area and provide very accurate segmentation results, but can struggle with challenging cases with unnatural postures, where the registration of the reference patient(s) is extremely difficult.
Therefore, we propose an coupled shape model (CoSMo) algorithm for the segmentation relevant structures in parallel. The model adaptation to a test image is done with respect to the appearance of its items and the trained articulation space. Even on very challenging data sets with unnatural postures, which occur far more often than expected, the model adaptation algorithm succeeds.
The approach is based on an articulated atlas \cite{Steger2012a}, that is trained from a set of manually labeled training samples. Furthermore, we have combined the initial solution with statistical shape models \cite{Kirschner2011} to represent structures with high shape variation. CoSMo is not tailored to specific structures or regions. It can be trained from any set of given gold standard segmentations and makes it thereby very generic.

Automatic Segmentation of Structures in CT Head and Neck Images using a Coupled Shape Model

We present a fully automatic method for segmenting mandible in CT images using anatomic landmarks and prior knowledge. The aim is to utilize spatial relationship of anatomic landmarks with image processing techniques to detect mandible robustly and efficiently. Applying prior knowledge and reliable anatomical landmarks to define an optimal Region of Interest (ROI) which contains the mandible is an effective way for fast localization and successful segmentation. This approach can be used to segment other structures such as optic nerves by defining a new set of relevant landmarks. This approach is robust to CT data with different scanner setting and does not require large training data sets.