Multimodal brain image analysis : third International Workshop, MBIA 2013, held in conjunction with MICCAI 2013, Nagoya, Japan, September 22, 2013 : proceedings. MBIA (Workshop) (3rd : 2013 : Nagoya-shi, Japan)最新文献

Analysis of 4D medical images presenting pathology (i.e., lesions) is significantly challenging due to the presence of complex changes over time. Image analysis methods for 4D images with lesions need to account for changes in brain structures due to deformation, as well as the formation and deletion of new structures (e.g., edema, bleeding) due to the physiological processes associated with damage, intervention, and recovery. We propose a novel framework that models 4D changes in pathological anatomy across time, and provides explicit mapping from a healthy template to subjects with pathology. Moreover, our framework uses transfer learning to leverage rich information from a known source domain, where we have a collection of completely segmented images, to yield effective appearance models for the input target domain. The automatic 4D segmentation method uses a novel domain adaptation technique for generative kernel density models to transfer information between different domains, resulting in a fully automatic method that requires no user interaction. We demonstrate the effectiveness of our novel approach with the analysis of 4D images of traumatic brain injury (TBI), using a synthetic tumor database as the source domain.

Ventricular volume (VV) is a powerful global indicator of brain tissue loss on MRI in normal aging and dementia. VV is used by radiologists in clinical practice and has one of the highest obtainable effect sizes for tracking brain change in clinical trials, but it is crucial to relate VV to structural alterations underlying clinical symptoms. Here we identify patterns of thinner cortical gray matter (GM) associated with dynamic changes in lateral VV at 1-year (N=677) and 2-year (N=536) intervals, in the ADNI cohort. People with faster VV loss had thinner baseline cortical GM in temporal, inferior frontal, inferior parietal, and occipital regions (controlling for age, sex, diagnosis). These findings show the patterns of relative cortical atrophy that predict later ventricular enlargement, further validating the use of ventricular segmentations as biomarkers. We may also infer specific patterns of regional cortical degeneration (and perhaps functional changes) that relate to VV expansion.

Alzheimer's disease (AD) is characterized by gradual neurodegeneration and loss of brain function, especially for memory during early stages. Regression analysis has been widely applied to AD research to relate clinical and biomarker data such as predicting cognitive outcomes from MRI measures. In particular, sparse models have been proposed to identify the optimal imaging markers with high prediction power. However, the complex relationship among imaging markers are often overlooked or simplified in the existing methods. To address this issue, we present a new sparse learning method by introducing a novel network term to more flexibly model the relationship among imaging markers. The proposed algorithm is applied to the ADNI study for predicting cognitive outcomes using MRI scans. The effectiveness of our method is demonstrated by its improved prediction performance over several state-of-the-art competing methods and accurate identification of cognition-relevant imaging markers that are biologically meaningful.

Alzheimer's disease (AD) is the most common cause of dementia in older adults. By the time an individual has been diagnosed with AD, it may be too late for potential disease modifying therapy to strongly influence outcome. Therefore, it is critical to develop better diagnostic tools that can recognize AD at early symptomatic and especially pre-symptomatic stages. Mild cognitive impairment (MCI), introduced to describe a prodromal stage of AD, is presently classified into early and late stages (E-MCI, L-MCI) based on severity. Using a graph-based semi-supervised learning (SSL) method to integrate multimodal brain imaging data and select valid imaging-based predictors for optimizing prediction accuracy, we developed a model to differentiate E-MCI from healthy controls (HC) for early detection of AD. Multimodal brain imaging scans (MRI and PET) of 174 E-MCI and 98 HC participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort were used in this analysis. Mean targeted region-of-interest (ROI) values extracted from structural MRI (voxel-based morphometry (VBM) and FreeSurfer V5) and PET (FDG and Florbetapir) scans were used as features. Our results show that the graph-based SSL classifiers outperformed support vector machines for this task and the best performance was obtained with 66.8% cross-validated AUC (area under the ROC curve) when FDG and FreeSurfer datasets were integrated. Valid imaging-based phenotypes selected from our approach included ROI values extracted from temporal lobe, hippocampus, and amygdala. Employing a graph-based SSL approach with multimodal brain imaging data appears to have substantial potential for detecting E-MCI for early detection of prodromal AD warranting further investigation.

In image-based medical research, atlases are widely used in many tasks, for example, spatial normalization and segmentation. If atlases are regarded as representative patterns for a population of images, then multiple atlases are required for a heterogeneous population. In conventional atlas construction methods, the "unit" of representative patterns is images. Every input image is associated with its most similar atlas. As the number of subjects increases, the heterogeneity increases accordingly, and a big number of atlases may be needed. In this paper, we explore using region-wise, instead of image-wise, patterns to represent a population. Different parts of an input image is fuzzily associated with different atlases according to voxel-level association weights. In this way, regional structure patterns from different atlases can be combined together. Based on this model, we design a variational framework for multi-atlas construction. In the application to two T1-weighted MRI data sets, the method shows promising performance, in comparison with a conventional unbiased atlas construction method.

Increasing evidence suggests that inflammation is one pathophysio-logical mechanism in Alzheimer's disease (AD). Recent studies have identifiedan association between the poly (ADP-ribose) polymerase 1 (PARP1) gene and AD. This gene encodes a protein that is involved in many biological functions, including DNA repair and chromatin remodeling, and is a mediator of inflammation. Therefore, we performed a targeted genetic association analysis to investigate the relationship between the PARP1 polymorphisms and brain micro-glial activity as indexed by [¹¹C]PBR28 positron emission tomography (PET). Participants were 26 non-Hispanic Caucasians in the Indiana Memory and Aging Study (IMAS). PET data were intensity-normalized by injected dose/total body weight. Average PBR standardized uptake values (SUV) from 6 bilateral regions of interest (thalamus, frontal, parietal, temporal, and cingulate cortices, and whole brain gray matter) were used as endophenotypes. Single nucleotide polymorphisms (SNPs) with 20% minor allele frequency that were within +/- 20 kb of the PARP1 gene were included in the analyses. Gene-level association analyses were performed using a dominant genetic model with translocator protein (18-kDa) (TSPO) genotype, age at PET scan, and gender as covariates. Analyses were performed with and without APOE ε4 status as a covariate. Associations with PBR SUVs from thalamus and cingulate were significant at corrected p<0.014 and <0.065, respectively. Subsequent multi-marker analysis with cingulate PBR SUV showed that individuals with the "C" allele at rs6677172 and "A" allele at rs61835377 had higher PBR SUV than individuals without these alleles (corrected P<0.03), and individuals with the "G" allele at rs6677172 and "G" allele at rs61835377 displayed the opposite trend (corrected P<0.065). A previous study with the same cohort showed an inverse relationship between PBR SUV and brain atrophy at a follow-up visit, suggesting possible protective effect of microglial activity against cortical atrophy. Interestingly, all 6 AD and 2 of 3 LMCI participants in the current analysis had one or more copies of the "GG" allele combination, associated with lower cingulate PBR SUV, suggesting that this gene variant warrants further investigation.

We present an analysis framework for large studies of multimodal clinical quality brain image collections. Processing and analysis of such datasets is challenging due to low resolution, poor contrast, mis-aligned images, and restricted field of view. We adapt existing registration and segmentation methods and build a computational pipeline for spatial normalization and feature extraction. The resulting aligned dataset enables clinically meaningful analysis of spatial distributions of relevant anatomical features and of their evolution with age and disease progression. We demonstrate the approach on a neuroimaging study of stroke with more than 800 patients. We show that by combining data from several modalities, we can automatically segment important biomarkers such as white matter hyperintensity and characterize pathology evolution in this heterogeneous cohort. Specifically, we examine two sub-populations with different dynamics of white matter hyperintensity changes as a function of patients' age. Pipeline and analysis code is available at http://groups.csail.mit.edu/vision/medical-vision/stroke/.

We present a method for studying brain connectivity by simulating a dynamical evolution of the nodes of the network. The nodes are treated as particles, and evolved under a simulated force analogous to gravitational acceleration in the well-known N -body problem. The particle nodes correspond to regions of the cortex. The locations of particles are defined as the centers of the respective regions on the cortex and their masses are proportional to each region's volume. The force of attraction is modeled on the gravitational force, and explicitly made proportional to the elements of a connectivity matrix derived from diffusion imaging data. We present experimental results of the simulation on a population of 110 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI), consisting of healthy elderly controls, early mild cognitively impaired (eMCI), late MCI (LMCI), and Alzheimer's disease (AD) patients. Results show significant differences in the dynamic properties of connectivity networks in healthy controls, compared to eMCI as well as AD patients.