Software
Atrophy Simulation Package
Project Page
This software package is used to simulate brain images with local growth/atrophy within a prescribed spherical region. Specifically, given an input image and its segmented image, the location of the center of the spherical region, and the radius of that sphere, it simulates new images that have tissue growth or shrinkage within that pre-specified brain region according to given rates (atrophy for rates less than one and growth for rates greater than one). The algorithm uses an iterative procedure that tries to achieve the given level of volumetric change for brain tissues within the region, by seeking a smooth deformation field, whose Jacobian determinants match the prescribed volume change rate within the region. Note that in the current software, the simulation of growth or atrophy for brain tissue requires that the input spherical region has to cover some CSF or background regions.
[8] B. Karacali and C. Davatzikos, "Estimating Topology Preserving and Smooth Displacement Fields," IEEE Transactions on Medical Imaging, vol. 23, pp. 868-880, 2004.
[9] B. Karacali and C. Davatzikos, "Simulation of tissue atrophy using a topology preserving transformation model," IEEE TMI, in press 2006.
[10] Zhong Xue, Dinggang Shen, Bilge Karacali, Joshua Stern, David Rottenberg, and Christos Davatzikos, "Simulating Deformations of MR Brain Images for Validation of Atlas-based Segmentation and Registration Algorithms", submitted, 2005.
BASIS: Build system And Software Implementation Standard
Project Page
This meta-project, not specifically related to the analysis of medical images, aims at reducing and standardizing our software development and maintenance efforts. Many more recent software packages distributed by us are build upon this software package. The BASIS package consists of a documentation of the standard, CMake modules including a template project implementing the build system standard, basic utility implementations in different programming languages to facilitate the compliance with the software implementation standard, and some auxiliary command-line tools such as in particular a tool to create an empty project.
BRAID: BRAin Image Database
The BRAin Image Database (BRAID) is a large-scale archive of normalized digital spatial and functional data with an analytical query mechanism. One of its many applications is the elucidation of brain structure-function relationships. BRAID stores spatially defined data from digital brain images which have been mapped into normalized Cartesian coordinates, allowing image data from large populations of patients to be combined and compared. The database also contains neurological data from each patient and a query mechanism that can perform statistical structure-function correlations.
The BRAID project is funded by a grant from the National Institutes of Health. The BRAID project is developing database technology for the manipulation and analysis of three-dimensional brain images derived from MRI, PET, CT, etc. BRAID is based on the PostgreSQL server, an object/relational DBMS, which allows a standard relational DBMS to be augmented with application-specific datatypes and operators. The BRAID project is adding operations and datatypes to support querying, manipulation and analysis of three-dimensional medical images, including:
- Image Datatypes
- BRAID supports a family of 3D image datatypes, each having an abstract type and an implementation type. Abstract types include boolean (for regions of interest), integer, float, vector (for representing morphological changes), tensor (for representing derivatives and standard deviations of vector images) and color. Implementation types at present include line-segment format and voxel array.
- Image Operators
- BRAID supports addition of images, multiplication (which is interpreted as intersection for boolean images), coercion of an image's abstract or implementation type to another value, and determination of volumes of regions of interest.
- Statistical Operators
- A chi-squared test has been added to SQL as an aggregate operator on pairs of boolean values.
- Web Interface
- A general-purpose Web gateway allows the results of queries that return computed images to be displayed.
You can download the BRAID source code 2.0. This version is developed under postgreSQL 7.3.4. The installation guide is here.
You can download BRAID LiveCD here (iso, md5). BRAID LiveCD is created based on Ubuntu 8.04. It has pre-installed all components including postgreSQL 8.2, BRAID, Apache, PHP. The best way to use this LiveCD is to install it. This LiveCD is free, but you need a password to use it. Please send your request to rong.chen@uphs.upenn.edu.
CLASSIC: Consistent Longitudinal Alignment and Segmentation for Serial Image Computing
Project Page
MR brain image segmentation is a key processing step in many brain image analysis applications, e.g. morphometry, automatic tissue labeling, tissue volume quantification, image registration, and computer integrated surgery. Analysis of a series of 3-D data of the same subject captured at different time-points, i.e. of a 4-D image, is important in many neuroimaging studies that concentrate on normal development, aging, and evolution of pathology. Consistent segmentation is particularly important in the literature of aging and Alzheimer's Disease (AD) since subtle brain changes that might be indicative of early stages of underlying pathology must be estimated from serial MR images. However, existing 3-D segmentation algorithms may not provide adequate longitudinal stability for serial brain images since they process each image at a time.
We have proposed a 4-D segmentation method that overcomes this limitation and significantly improves longitudinal stability of segmentation, referred to as CLASSIC: Consistent Longitudinal Alignment and Segmentation for Serial Image Computing. CLASSIC builds upon previous 3D methods for brain image segmentation, especially FANTASM [4], and extends them to 4D as well as to simultaneous estimation of segmentation and atrophy.
[22] Zhong Xue, Dinggang Shen, Christos Davatzikos, "CLASSIC: Consistent Longitudinal Alignment and Segmentation for Serial Image Computing", Neuroimage, 388-399, Vol. 30, No. 2, 2006
COMPARE:
Classification Of Morphological Patterns using Adaptive Regional Elements
Project Page
COMPARE is a method for classification of structural brain magnetic resonance (MR) images, which is a combination of deformation-based morphometry and machine learning methods. Before running classification, a morphological representation of the anatomy of interest is obtained from structural MR brain images using a high-dimensional mass-preserving template warping method [1, 2]. Regions that display strong correlations between tissue volumes and classification (clinical) variables learned from training samples are extracted using a watershed segmentation algorithm. To achieve robustness to outliers, the regional smoothness of the correlation map is estimated by a cross-validation strategy. A volume increment algorithm is then applied to these regions to extract regional volumetric features. To improve efficiency and generalization ability of the classification, a feature selection technique using Support Vector Machine-based criteria is used to select the most discriminative features, according to their effect on the upper bound of the leave-one-out generalization error. Finally, SVM-based classification is applied using the best set of features, and it is tested using a leave-one-out cross-validation strategy. Although the algorithm is designed for structural brain image classification, it is readily applicable for functional brain image classification with proper feature images. For simplicity, here we focus on structural brain image classification.
[16] COMPARE: Classification Of Morphological Patterns using Adaptive Regional Elements Yong Fan, Dinggang Shen, Ruben C. Gur, Raquel E. Gur, Christos Davatzikos IEEE Transactions on Medical Imaging, 93-105, Vol. 26, No. 1, 2007.
[17] C. Davatzikos, A. Genc, D. Xu, and S. M. Resnick, "Voxel-Based Morphometry Using the RAVENS Maps: Methods and Validation Using Simulated Longitudinal Atrophy", NeuroImage, vol. 14, pp. 1361-1369, 2001.
DTI-DROID:
Deformable Registration using Orientation and Intensity Descriptors
Project Page
Diffusion tensor (DT) imaging is a relatively new magnetic resonance imaging method, which has emerged during the past few years as a potentially powerful way of understanding connectivity in the brain. DT imaging is based on measurements of microscopic diffusion of water molecules, which provides insight into homogeneous white matter and indicates the direction of nerve bundles. Since brain connectivity is important in studying brain development, aging, and disease processes, DTI is bound to play an important role in these scientific areas. Spatial normalization of tensor fields pose difficulties not previously considered in deformable registration methods of scalar images. In addition to the tensor's relocation to the template space, the orientation of each tensor has to be properly adjusted, which implies that the actual measurement on each voxel is both displaced and changed by the spatial transformation. Moreover, the reorientation of a tensor relies on the shape of the tensor in relation to the deformation field direction at that location. Therefore, the same deformation field prescribes a different re-orientation for different tensors.
The DTIGUI performs spatial normalization on DTI human brain images to facilitate subsequent statistical analysis (such as voxel based analysis or group averaging).
[14] J.Yang, D. Shen, C. Misra, X. Wu, S. Resnick, C. Davatzikos, and R. Verma Spatial Normalization of Diffusion Tensor Images Based on Anisotropic Segmentation SPIE Medical Imaging 2008, San Diego, Feb 2008.
[15] J.Yang, D. Shen, C. Davatzikos, and R. Verma Diffusion tensor image registration using tensor geometry and orientation features MICCAI 2008.
GLISTR: GLioma Image SegmenTation and Registration
Project Page
Automatic segmentation and atlas normalization of brain tumor images are extremely challenging and clinically important tasks. We have developed a package for GLioma Image SegmenTation and Registration (GLISTR) for such specific goals. This method performs both spatial normalization of the brain tumor images into an originally healthy atlas, and segmentation of multi-channel MR images into six tissue types: tumor, necrosis, edema, cerebrospinal fluid, gray and white matters. The method introduces a glioma growth diffusion-reaction model into the segmentation procedure, and estimates the tissue labels, warping parameters and the diffusion-reaction model parameters in an EM framework. The tumor embedding in the atlas space is intended for subject-specific modification of the originally healthy into an atlas with tumor and edema priors. It introduces mass-effect and diffusion of (artificial) glioma cells into the healthy tissues. This greatly improves the registration performance by allowing the inferred deformation field to be smooth.
Fig.1 Sample screenshots for one patient. See project page for details.
[1] A.Gooya, K.M.Pohl, M.Billelo, G.Biros and C.Davatzikos, "Joint segmentation and deformable registration of brain scans guided by a tumor growth model", MICCAI (2011), Toronto, Canada.
[2] K.M.Pohl, J.Fisher, R.Kikinis, W.M.Wells, "A bayesian model for joint segmentation and registration", NeuroImage 31, 228–239 (2006)
[3] A.Gooya, G.Biros, C.Davatzikos, "Deformable registration of glioma images using EM algorithm and diffusion reaction modeling", IEEE TMI. 30, 375–90 (2011)
[4] C.Hogea, C.Davatzikos, G.Biros, "An image-driven parameter estimation problem for a reaction-diffusion glioma growth model with mass effects" , J.Math.Biol 56,793–825 (2008)
HAMMER:
Hierarchical Attribute Matching Mechanism for Elastic Registration
Project Page
The HAMMER package performs high-dimensional warping of brain images [1]. It can be run in two modes. In the first mode, it produces labels for a number of anatomical regions of interest (ROIs), which correspond to gyral and subconrtical brain structures, such as hippocampus, superior temporal gyrus, superior frontal gyrus, etc. Approximately 100 ROIs are generated. In the second mode, it produces tissue density maps, i.e. images whose local intensity is directly proportional to the local GM, WM or CSF volume in the MR image before warping. These tissue density maps can be used for voxel-based analysis of regional volumetrics. They are derived using mass-preserving transformations [2,3], which warranty that brain tissues are preserved during warping to a template. For example, local contraction when warping an individual image to the template results in increase in tissue density, so that the total amount of tissue is preserved. Conversely for local expansion, which spreads the tissue over a larger volume, and therefore decreases its density to maintain the total amount if tissue fixed. Standard voxel-based analysis (such as t-tests of the general linear model of SPM) can be applied to these tissue density maps, in order to examine regional volumetrics, effects of disease, or correlations with clinical measurements.
[1] D. Shen and C. Davatzikos, "HAMMER: Hierarchical attribute matching mechanism for elastic registration", IEEE Transactions on Medical Imaging, vol. 21, pp. 1421-1439, 2002.
[2] D. G. Shen and C. Davatzikos, "Very high resolution morphometry using mass-preserving deformations and HAMMER elastic registration", NeuroImage, vol. 18, pp. 28-41, 2003.
[3] C. Davatzikos, A. Genc, D. Xu, and S. M. Resnick, "Voxel-Based Morphometry Using the RAVENS Maps: Methods and Validation Using Simulated Longitudinal Atrophy", NeuroImage, vol. 14, pp. 1361-1369, 2001.
Mouse Datasets (Diffusion Tensor Images)
Project Page
These mouse datasets provide developmental patterns across subjects and gender for understanding growth and development of murine brains with respect to structural changes observed via diffusion properties. In the following table, the representative datasets from 9 development stages (days) are provided. In each stage, we include one male subject and one female subject, as well as the group averaged image in this day. Also provided are the corresponding FA images. The corresponding filenames are provided in the table.
Data format:
The provided diffusion tensor image is a voxel-wise image with each voxel containing 6 tensor components [Dxx Dyy Dzz Dxy Dxz Dyz] in float format. The FA image is in float format. All datasets are stored with little endian. They are resampled to be with size 300x300x200, and voxel resolution 0.06x0.06x0.06 mm.
The details about the dataset are:
| Days | Male | Female | Average |
| Day 2 | M_2_M_D (DTI & FA) | M_2_F_A (DTI & FA) | meanDay2 (DTI & FA) |
| Day 3 | M_3_M_A (DTI&FA) | M_3_F_B (DTI&FA) | meanDay3 (DTI&FA) |
| Day 5 | N/A | M_5_F_C (DTI&FA) | meanDay5 (DTI&FA) |
| Day 7 | M_7_M_E (DTI&FA) | M_7_F_B (DTI&FA) | meanDay7 (DTI&FA) |
| Day 10 | M_10_M_C(DTI&FA) | M_10_F_A (DTI&FA) | meanDay10(DTI&FA) |
| Day 15 | M_15_M_B(DTI&FA) | M_15_F_A(DTI&FA) | meanDay15(DTI&FA) |
| Day 20 | M_20_M_F (DTI&FA) | N/A | N/A |
| Day 30 | M_30_M_A (DTI&FA) | N/A | N/A |
| Day 40 | N/A | M_40_F_C(DTI&FA) | N/A |
| Day 80 | N/A | N/A | N/A |
Please refer to the following papers in case you use the dataset in your research: [12] [13]. Also please send an e-mail to Christos.Davatzikos [christos@rad.upenn.edu]. If you need the complete dataset details for which are provided here. Please send an email to Christos and Ragini and we will send you the data on a DVD
[12] Sajjad Baloch, Ragini Verma, Hao Huang, Parmeshwar Khurd, Sarah Clark, Paul Yarowsky, Ted Abel, Susumu Mori, Christos Davatzikos, "Quantification of brain maturation and growth patterns in C57BL/6J mice via computational neuroanatomy of diffusion tensor images", Cerebral Cortex, in press, 2008.
[13] Ragini Verma, Susumu Mori, Dinggang Shen, Paul Yarowsky, Christos Davatzikos, "Spatio-temporal maturation patterns of murine brain quantified via diffusion tensor MRI and deformation-based morphometry", Proceedings of the National Academy of Sciences 102(19): 6978-6983, May 2005.
ODVBA: Optimally-Discriminative Voxel-Based Analysis
Project Page
This software is the implementation of a framework termed Optimally-Discriminative Voxel-Based Analysis (ODVBA) [1] which is developed for determining the optimal spatial smoothing of structural and potentially functional images, prior to applying voxel-based group analysis.
The method introduced Nonnegative Discriminative Projection (NDP) to find the optimal discriminative direction in each learning set constructed by the neighborhood centered at the given voxel. Subsequently, each voxel’s statistic is determined by a composition of all the smoothing directions which are associated with the given voxel. Finally, permutation tests are used to obtain the statistical significance of the resulting ODVBA maps.
It is evaluated that ODVBA is much more powerful to detect the true atrophy hidden in the data than the traditional methods, e.g., SPM [2] and SnPM [3]. Moreover, ODVBA possesses the spatial adaptivity to the shape and spatial extent of the region of interest, while SPM and SnPM do not (see Fig.2).
Fig.2 ODVBA precisely describes the "U"-like shape of the atrophy. See project page for details.
[1] T. Zhang, C. Davatzikos, "Optimally-Discriminative Voxel-Based Analysis", Proceeding of International Conference on Medical Image Computing and Computer-Assisted Intervention, vol. 13, no.2, pp: 257-265 (2010)
[2] J. Ashburner, K.J. Friston, "Voxel-based morphometry - The methods", Neuroimage, 11(6) 805–821 (2000)
[3] T.E. Nichols, A.P. Holmes, "Nonparametric permutation tests for functional neuroimaging: a primer with examples", Human Brain Mapping, vol. 15, no. 1, pp: 1-25 (2002)
Optimized Prostate Cancer Detection using a Statistical Atlas of Cancer Distribution
Project Page
Prostate cancer is one of the leading causes of death in men. Early diagnosis of cancer, and especially of clinically significant cancer, is critical for effective treatment. Current imaging methods are insufficient for accurate detection of prostate cancer; therefore definitive diagnosis is typically achieved via biopsy. Current biopsy approaches place 6 or more needles into the prostate and extract tissue samples, and are known to miss a large number of cancers due to simply sampling error. We have investigated systematic (targeted) biopsy methods that obtain tissue samples from locations that, together, maximize the probability of detecting cancer [11]. We have utilized a database of 158 radical prostatectomy specimens processed at the Center for Prostate Disease Research, and developed a statistical atlas of the spatial distribution of prostate cancer; this process required the application of elastic-type deformable registration techniques, as explained in [11] and the documentation. We subsequently applied optimization methods and determined the optimal needle placement that maximized probability of cancer detection for unconstrained, transrectal and transperineal biopsies. We are currently working on adding multi-parametric MRI segmentation and pattern classification methods, in order to combine population-based data with patient-specific information towards a patient-optimized biopsy system.
[11] Y. Zhan, D. Shen, J. Zeng, L. Sun, G. Fichtinger, J. Moul, and C. Davatzikos, "Targeted Prostate Biopsy Using Statistical Image Analysis", IEEE Transactions on Medical Imaging, vol. 26, pp. 779-788, June 2007.
Statistically-based Simulation of Deformations: Brain Anatomy Simulator Using Statistical Shape Modeling
This package estimates the statistical properties of high-dimensional deformation fields[6], which are produced by deformable registration packages like HAMMER above, and then uses the estimates statistics to simulate brain images with very high degree of realism. The statistical properties of a family of deformations are estimated via a combination of wavelet packet decomposition on the deformations and their Jacobians, and PCA. This package can be used to better represent the statistical priors of deformation fields in deformable registration methods that use priors, such as active shape models. Moreover, SSD can be used to generate very realistic and rich deformations, which can be used for generation of gold standard against which different registration methods can be compared. Finally, a method for generation of simulated tissue atrophy is supplied as a separate package [7], which further enables simulation and validation studies.
[6] Z. Xue, D. Shen, B. Karacali, and C. Davatzikos, "Statistical Representation and Simulation of High-Dimensional Deformations: Application to Synthesizing Brain Deformations," presented at Medical Image Computing and Computer Assisted Intervention (MICCAI 2005), Palm Springs, California, USA, 2005.
[7] B. Karacali and C. Davatzikos, "Estimating Topology Preserving and Smooth Displacement Fields," IEEE Transactions on Medical Imaging, vol. 23, pp. 868-880, 2004.
TetSplit: Mesh Generation for Finite Element Models
The TetSplit package is a general purpose program for generating finite element tetrahedral meshes from 3D labeled images [4,5]. The resulting tetrahedral have both good quality and they conform to the geometry specified by the labels. Moreover, their density can be spatially variable and specified by the user. It has been used primarily with medical images.
[4] A. Mohamed and C. Davatzikos, "Finite Element Mesh Generation and Remeshing from Segmented Medical Images", presented at 2004 IEEE International Symposium on Biomedical Imaging: From Nano to Macro, Arlington, Va., 2004.
[5] A. Mohamed, "Combining Statistical and Biomechanical Models for Estimation of Anatomical Deformations, Ph.D. thesis", in Computer Science. Baltimore: Johns Hopkins University, 2005.
WMLS: White Matter Lesion Segmentation
Project Page
White matter lesions (WML) are brain abnormalities that appear in different brain diseases, such as multiple sclerosis (MS), head injury, vascular disease particularly related to hypertension possibly diabetes, and some forms of dementia. Their incidence also increases with normal aging. MRI is routinely used as surrogate in the study of WMLs, as MRI signal changes reflect certain aspects of the underlying brain pathology. Out of the many available MRI acquisition protocols, T1-w and T2-w, PD, and FLAIR are among the most commonly used to evaluate white matter lesion load in the brain. Computer analysis methods have started to complement expert-readings of these images, as they may improve throughput and consistency, in addition to providing more accurate quantitative measures of WML. Computer analysis is even more critical in longitudinal studies that involve relatively small changes in WML, thereby rendering it advantageous, if not necessary, to use unbiased computer-assisted segmentation methods to detect WMLs and assess their longitudinal change.
[18] E.I. Zacharaki, S. Kanterakis, R.N. Bryan, C. Davatzikos, Measuring brain lesion progression with a supervised tissue classification system MICCAI 2008, September 6 -10, 2008, New York, USA.
[19] Z. Lao, D. Shen, D. Liu, A. Jawad, E.R. Melhem, L.J. Launer, R.N. Bryan, C. Davatzikos Computer-Assisted Segmentation of White Matter Lesions in 3D MR images, Using Support Vector Machine Academic Radiology, 300-313, Vol. 15, No. 3, 2008.
[20] G. Moonis, Z. Lao, D. Liu, C. Davatzikos, R.N. Bryan, A. Jawad Validation of an automated segmentation technique against at established visual rating system for measures of brain atrophy and white matter lesion load ASNR 44th annual meeting, San Diego, California, April 29-May 5, 2006.
[21] Z.Q. Lao, D. Shen, A. Jawad, B. Karacali, D. Liu, E.R. Melhem, N.R. Bryan, C. Davatzikos Automated Segmentation of White Matter Lesions in 3D Brain MR Images, Using Multivariate Pattern Classification Third IEEE International Symposium on