Yakang Dai

eConnectome: A MATLAB Toolbox for Mapping and Imaging of Brain Functional Connectivity

We have developed a MATLAB-based toolbox, eConnectome (electrophysiological connectome), for mapping and imaging functional connectivity at both the scalp and cortical levels from the electroencephalogram (EEG), as well as from the electrocorticogram (ECoG). Graphical user interfaces were designed for interactive and intuitive use of the toolbox. Major functions of eConnectome include EEG/ECoG preprocessing, scalp spatial mapping, cortical source estimation, connectivity analysis, and visualization. Granger causality measures such as directed transfer function and adaptive directed transfer function were implemented to estimate the directional interactions of brain functional networks, over the scalp and cortical sensor spaces. Cortical current density inverse imaging was implemented using a generic realistic geometry brain-head model from scalp EEGs. Granger causality could be further estimated over the cortical source domain from the inversely reconstructed cortical source signals as derived from the scalp EEG. Users may implement other connectivity estimators in the framework of eConnectome for various applications. The toolbox package is open-source and freely available at http://econnectome.umn.edu under the GNU general public license for noncommercial and academic uses.

LABEL: Pediatric brain extraction using learning-based meta-algorithm

Magnetic resonance imaging of pediatric brain provides valuable information for early brain development studies. Automated brain extraction is challenging due to the small brain size and dynamic change of tissue contrast in the developing brains. In this paper, we propose a novel Learning Algorithm for Brain Extraction and Labeling (LABEL) specially for the pediatric MR brain images. The idea is to perform multiple complementary brain extractions on a given testing image by using a meta-algorithm, including BET and BSE, where the parameters of each run of the meta-algorithm are effectively learned from the training data. Also, the representative subjects are selected as exemplars and used to guide brain extraction of new subjects in different age groups. We further develop a level-set based fusion method to combine multiple brain extractions together with a closed smooth surface for obtaining the final extraction. The proposed method has been extensively evaluated in subjects of three representative age groups, such as neonate (less than 2 months), infant (1-2 years), and child (5-18 years). Experimental results show that, with 45 subjects for training (15 neonates, 15 infant, and 15 children), the proposed method can produce more accurate brain extraction results on 246 testing subjects (75 neonates, 126 infants, and 45 children), i.e., at average Jaccard Index of 0.953, compared to those by BET (0.918), BSE (0.902), ROBEX (0.901), GCUT (0.856), and other fusion methods such as Majority Voting (0.919) and STAPLE (0.941). Along with the largely-improved computational efficiency, the proposed method demonstrates its ability of automated brain extraction for pediatric MR images in a large age range.

A Novel Software Platform for Medical Image Processing and Analyzing

The design of software platform for medical imaging application has been increasingly prioritized as the sophisticated application of medical imaging. With this demand, we have designed and implemented a novel software platform in traditional object-oriented fashion with some common design patterns. This platform integrates the mainstream algorithms for medical image processing and analyzing within a consistent framework, including reconstruction, segmentation, registration, visualization, etc., and provides a powerful tool for both scientists and engineers. The overall framework and certain key technologies are introduced in detail. Presented experiment examples, numerous downloads, extensive uses, and practical applications commendably demonstrate the validity and flexibility of the platform.

iBEAT: A Toolbox for Infant Brain Magnetic Resonance Image Processing

It’s a great challenge to analyze infant brain MR images due to the small brain size and low contrast of the developing brain tissues. We have developed an Infant Brain Extraction and Analysis Toolbox (iBEAT) for various processing of magnetic resonance (MR) images of infant brains. Several major functions generally used in infant brain analysis are integrated in iBEAT, including image preprocessing, brain extraction, tissue segmentation, and brain labeling. The functions of brain extraction, tissue segmentation, and brain labeling are provided respectively by three state-of-the-art algorithms. First, a learning-based meta-algorithm which integrates a group of brain extraction results generated by the two existing brain extraction algorithms (BET and BSE) was implemented in iBEAT for extraction of infant brains from MR images. Second, a level-sets-based tissue segmentation algorithm that utilizes multimodality information, cortical thickness constraint, and longitudinal consistency constraint was also included in iBEAT for segmentation of infant brain tissues. Third, HAMMER (standing for Hierarchical Attribute Matching Mechanism for Elastic Registration) registration algorithm was further included in iBEAT to label regions of interest (ROIs) of infant brain images by warping the pre-labeled ROIs of a template to the infant brain image space. By integration of these state-of-the-art methods, iBEAT is able to segment and label infant brain MR images accurately. Moreover, it can process not only single-time-point images for cross-sectional studies, but also multiple-time-point images of the same infant for longitudinal studies. The performance of iBEAT has been comprehensively evaluated with hundreds of infant brain images. A Linux-based standalone package of iBEAT is freely available at http://www.nitrc.org/projects/ibeat.

Real-Time Visualized Freehand 3D Ultrasound Reconstruction Based on GPU

Visualized freehand 3-D ultrasound reconstruction offers to image incremental reconstruction during acquisition and guide users to scan interactively for high-quality volumes. We originally used the graphics processing unit (GPU) to develop a visualized reconstruction algorithm that achieves real-time level. Each newly acquired image was transferred to the memory of the GPU and inserted into the reconstruction volume on the GPU. The partially reconstructed volume was then rendered using GPU-based incremental ray casting. After visualized reconstruction, hole-filling was performed on the GPU to fill remaining empty voxels in the reconstruction volume. We examine the real-time nature of the algorithm using in vitro and in vivo datasets. The algorithm can image incremental reconstruction at speed of 26-58 frames/s and complete 3-D imaging in the acquisition time for the conventional freehand 3-D ultrasound.

Source Connectivity Analysis from MEG and its Application to Epilepsy Source Localization

We report an approach to perform source connectivity analysis from MEG, and initially evaluate this approach to interictal MEG to localize epileptogenic foci and analyze interictal discharge propagations in patients with medically intractable epilepsy. Cortical activities were reconstructed from MEG using individual realistic geometry boundary element method head models. Directional connectivity among cortical regions of interest was then estimated using directed transfer function. The MEG source connectivity analysis method was implemented in the eConnectome software, which is open-source and freely available at http://econnectome.umn.edu. As an initial evaluation, the method was applied to study MEG interictal spikes from five epilepsy patients. Estimated primary epileptiform sources were consistent with surgically resected regions, suggesting the feasibility of using cortical source connectivity analysis from interictal MEG for potential localization of epileptiform activities.

Fast Katsevich Algorithm Based on GPU for Helical Cone-Beam Computed Tomography

Katsevich reconstruction algorithm represents a breakthrough for helical cone-beam computed tomography (CT) reconstruction, because it is the first exact cone-beam reconstruction algorithm of filtered backprojection (FBP) type with 1-D shift-invariant filtering. Although FBP-type reconstruction algorithm is effective, 3-D CT reconstruction is time-consuming, and the accelerations of Katsevich algorithm on CPU or cluster have been widely studied. In this paper, Katsevich algorithm is accelerated by using graphics processing unit, including flat-detector and curved-detector geometry in the case of helical orbit. An overscan formula is derived, which helps to avoid unnecessary overscan in practical CT scanning. Based on the overscan formula, a volume-blocking method in device memory is proposed. One advantage of the blocking method is that it can reconstruct large volume with high speed.

Robust Non-Rigid Point Set Registration Using Student's-t Mixture Model

The Students-t mixture model, which is heavily tailed and more robust than the Gaussian mixture model, has recently received great attention on image processing. In this paper, we propose a robust non-rigid point set registration algorithm using the Students-t mixture model. Specifically, first, we consider the alignment of two point sets as a probability density estimation problem and treat one point set as Students-t mixture model centroids. Then, we fit the Students-t mixture model centroids to the other point set which is treated as data. Finally, we get the closed-form solutions of registration parameters, leading to a computationally efficient registration algorithm. The proposed algorithm is especially effective for addressing the non-rigid point set registration problem when significant amounts of noise and outliers are present. Moreover, less registration parameters have to be set manually for our algorithm compared to the popular coherent points drift (CPD) algorithm. We have compared our algorithm with other state-of-the-art registration algorithms on both 2D and 3D data with noise and outliers, where our non-rigid registration algorithm showed accurate results and outperformed the other algorithms.

Cerebral alterations of type 2 diabetes mellitus on MRI: A pilot study.

This study is to investigate gray matter volume, cortical thickness, and surface area of the brain in patients with type 2 diabetes mellitus (T2DM). High resolution T1-weighted MR images were obtained from eighteen T2DM and seventeen normal controls. All images were processed using our newly developed BrainLab toolbox. Declines of gray matter volume, cortical thickness, and surface area were found in T2DM patients. Significantly reduced ROIs of gray matter volume happened in subcortical gray nuclei (left caudate and right caudate), and significantly reduced ROIs of cortical thickness occurred in temporal lobe (left superior temporal gyrus), parietal lobe (left angular gyrus), and occipital lobe (right superior occipital gyrus, left middle occipital gyrus and right cuneus). Apparently reduced ROIs of surface area were mainly distributed in frontal lobe (right superior frontal gyrus (dorsal) and left paracentral lobule). The findings indicated that T2DM caused brain changes in specific regions. This work revealed neural alterations of T2DM, which had a great significance in early diagnosis of the disease.

Multi-modal Medical Image Registration Based on Adaptive Combination of Intensity and Gradient Field Mutual Information

Mutual information (MI) is an effective criterion for multi-modal image registration. However the traditional MI function only includes intensity information of images and lacks sufficient spatial information to accurately measure the degree of alignment of two images, and besides, it is apt to be influenced by intensity interpolation, therefore presents many local maxima which frequently lead to misregistration. Our paper proposes a criterion of adaptive combination of intensity and gradient field mutual information (ACMI). Unlike the intensity MI computed from two original images, the gradient field MI of two images is calculated from their gradient code maps (GCM) constructed by coding the gradient field information of corresponding original image. Because of their complementary properties, these two MI functions are combined to form ACMI by a nonlinear weight function which can be adaptively regulated according to their performances and make the better dominant in the combination. Experimental results demonstrate that the ACMI outperforms the traditional MI and furthermore the former is much less sensitive than the latter to the reduction of resolution or overlapped region of images