Search

We examined the effects of visual deprivation at birth on the development of the corpus callosum in a large group of congenitally blind individuals. We acquired high-resolution T1-weighted MRI scans in 28 congenitally blind and 28 normal sighted subjects matched for age and gender. There was no overall group effect of visual deprivation on the total surface area of the corpus callosum. However, subdividing the corpus callosum into five subdivisions revealed significant regional changes in its three most posterior parts. Compared to the sighted controls, congenitally blind individuals showed a 12% reduction in the splenium, and a 20% increase in the isthmus and the posterior part of the body. A shape analysis further revealed that the bending angle of the corpus callosum was more convex in congenitally blind compared to the sighted control subjects. The observed morphometric changes in the corpus callosum are in line with the well-described cross-modal functional and structural neuroplastic changes in congenital blindness.

The in vivo detection of subpial cortical gray matter lesions in multiple sclerosis is challenging. We quantified the spatial extent of subpial decreases in the magnetization transfer ratio (MTR) of cortical gray matter in subjects with multiple sclerosis, as such reductions may indicate regions of cortical demyelination. We exploited the unique geometry of cortical lesions by using two‐dimensional parametric surface models of the cortex instead of traditional three‐dimensional voxel‐wise analyses. MTR images were mapped onto intermediate surfaces between the pial and white matter surfaces and were used to compute differences between secondary‐progressive MS (n = 12), relapsing‐remitting MS (n = 12), and normal control (n = 12) groups as well as between each individual patient and the normal controls. We identified large regions of significantly reduced cortical MTR in secondary‐progressive patients when compared with normal controls. We also identified large regions of reduced cortical MTR in 11 individual patients (8 secondary‐progressive, 3 relapsing‐remitting). The secondary‐progressive patients showed larger areas of abnormally low MTR compared with relapsing‐remitting patients both at the group level and on an individual basis. The spatial distributions of abnormal MTR preferentially involved cingulate cortex, insula, and the depths of sulci, in agreement with pathological descriptions of subpial gray matter lesion distribution. These findings suggest that our method is a plausible in vivo imaging technique for quantifying subpial cortical demyelinating lesions in patients with multiple sclerosis and, furthermore, can be applied at the typical clinical field strength of 1.5 T. Hum Brain Mapp 35:3402–3413, 2014. © 2013 Wiley Periodicals, Inc.

While the automatic detection of AD has been widely studied, the problem of the prediction of AD (or its early detection) has been less investigated. This might be explained by the fact that the prediction problem is clearly more challenging since the anatomical changes are more subtle. However, from a clinical point of view the prediction of AD is the key question since it is in that moment when treatment is possible. The potential use of structural MRI as imaging biomarker for Alzheimer’s disease (AD) for early detection has become generally accepted, especially the use of atrophy of entorhinal cortex (EC) and hippocampus (HC). Therefore, in this study, we analyze the capabilities of the recently proposed method, SNIPE (Scoring by Nonlocal Image Patch Estimator), for the early detection of AD to analyze EC and HC atrophy over the entire ADNI database (834 subjects). During validation, the detection (AD vs. CN) and the prediction (pMCI vs. sMCI) efficiency of SNIPE were studied. The obtained results showed that SNIPE obtained competitive or better results than HC volume, cortical thickness and TBM. Moreover, results indicated that MRI grading-based biomarkers are more relevant than volume-based biomarkers. Finally, the success rate obtained by SNIPE was 90% for detection (AD vs. CN) and 74% for prediction (pMCI vs. sMCI).

International audience; Predicting Alzheimer's disease (AD) in individuals with some symp-toms of cognitive decline may have great influence on treatment choice and guide subject selection in trials on disease modifying drugs. Structural MRI has the potential of revealing early signs of neurodegeneration in the human brain and may thus aid in predicting and diagnosing AD. Surface-based cortical thickness measurements from T1-weighted MRI have demonstrated high sensi-tivity to cortical gray matter changes. In this study, we investigated the possibil-ity of using patterns of cortical thickness measurements for predicting AD in subjects with mild cognitive impairment (MCI). Specific patterns of atrophy were identified at four time periods before diagnosis of probable AD and fea-tures were selected as regions of interest within these patterns. The selected re-gions were used for cortical thickness measurements and applied in a classifier for testing the ability to predict AD at the four stages. The accuracy of the pre-diction improved as the time to conversion from MCI to AD decreased, from 70% at 3 years before the clinical criteria for AD was met, to 76% at 6 months before AD. These results show that prediction accuracies of conversion from MCI to AD can be improved by learning the atrophy patterns that are specific to the different stages of disease progression. This has the potential to guide the further development of imaging biomarkers in AD.

International audience; We propose a new method combining a population-specific nonlinear template atlas approach with non-local patch-based structure segmentation for whole brain segmentation into individual structures. This way, we benefit from the efficient intensity-driven segmentation of the non-local means framework and from the global shape constraints imposed by the nonlinear template matching.

Subthalamic nucleus (STN) deep brain stimulation (DBS) has been demonstrated as an efficient surgical treatment in patients with Parkinson's disease suffering from severe disabilities in their motor symptoms (tremor, akinesia, rigidity) [1]. Sometimes, patients with STN DBS have secondary neuropsychological and/or psychiatric effects because the STN has a small size and a functional subdivision into motor, associative and limbic parts [8]. Therefore, targeting is an important step of the neurosurgical procedure as the DBS could induce undesirable side-effects. To improve targeting, many authors created either anatomical atlases [2,4] or multimodal databases [5-7]. They demonstrated that multimodal databases improved targeting accuracy over anatomical atlases. We presented a validation automatic segmentation of the basal ganglia based on a manual segmentation of these structures completed on a 3T MRI-template. This is the first step to increase accuracy of the targeting of DBS in order to improve the clinical outcome of DBS procedures.

Cortical thickness measurements based on MRI have the potential to accurately detect changes in the cortical gray matter. During the last decade, several methods have been developed to estimate cortical thickness using surface based reconstruction. Due to the high sensitivity of reconstruction methods to acquisition artifacts, noise in MRI signal may directly affect the obtained measurements. Usually, this noise is reduced using various smoothing filters. Recently, the non-local means (NLM) filter has demonstrated very high denoising performance compared to these traditional filters. In this study, we investigated the sensitivity to noise of cortical surface reconstruction and the ability of the NLM filter to reduce the effect of noise.

International audience; Longitudinal measures of brain volume are powerful tools to assess the anatomical changes underlying on-going neurodegenerative processes. In different neurological disorders, such as in multiple sclerosis, Alzheimer's disease and Parkinson's disease, the neurodegenerative aspect may result in subtle anatomical brain changes before the appearance of clinical symptoms. Large longitudinal brain imaging datasets are now accessible to investigate such structural changes over time and to evaluate their use as biomarkers of prodromal disease. However, manual segmentation is long and tedious and although automatic methods exist, they are often performed in a cross-sectional manner where each visit is analysed independently. With such analysis methods, bias, error and longitudinal noise may be introduced. MR scanner noise and physiological effects can also introduce additional variability. Therefore, we developed a specific pipeline for longitudinal brain image analysis. To avoid any bias, an individual subject template is created and used as a reference within the pipeline. Then, the pair-wise deformation fields of each visit to the individual template are used to estimate the variation between individual time-points.

Background: While widely used to detect morphological differences between groups, Voxel-Based Morphometry (VBM) is based on the assumption of one-to-one anatomical mapping between subjects and Gaussian distributions of focal tissue densities during statistical testing. To make data fit this model, tissue densities are blurred with large kernels at the expense of focal accuracy. To these issues, we propose a new Patch-Based Morphometry (PBM) method derived from our recently proposed innovative method to detect fine anatomical changes in MRI called Scoring by Nonlocal Image Patch Estimator . SNIPE takes advantage of non-local analysis to handle the one-to-many mapping between brain anatomies. In this study, we extend SNIPE to the whole brain before comparing populations with PBM scores.Methods: We randomly selected 50 MRI from cognitive normal (CN) subjects and 50 MRI from AD patients from the ADNI database. Step 1: the 100 images were processed as described in (inhomogeneity correction, intensity normalization and rigid registration to MNI-ICBM152-nonlinear). Through a leave-one-out procedure, SNIPE was applied on each of the MRI scans using 30 images from each population as training templates. Step 2: all grading maps were nonlinearly registered to the MNI-ICBM152-nonlinear template with ANIMAL . Step 3: a non-parametric Kruskall-Wallis test was performed at each voxel to estimate statistical differences between populations.Results: Examples of grading maps are presented in Figure 1. Figure 2 shows the p-values overlaid on the MNI-template. Maximum differences between AD and CN were found in hippocampus and para-hippocampal areas, entorhinal cortex and in the temporal lobe around the lateral sulci and insula. Moreover, diffuse differences appear within the gray matter. These results are consistent with previous VBM results . We also noted an important difference around the superior mammillary notches as previously reported in volumetric studies . Conclusion: In this proof of concept study, we showed that PBM produces results consistent with previously published VBM studies. However, contrary to VBM, these results were obtained without a blurring step since PBM can work at the voxel resolution. Further work will investigate optimal parameters for SNIPE and the possibility of using multivariate tests.

International audience; The early detection of Alzheimer's disease (AD) is a key step to accelerate the development of new therapies and to diminish the associated socio-economic burden. To address this challenging problem, several biomarkers based on MRI have been proposed. Although numer- ous efforts have been devoted to improve MRI-based feature quality or to increase machine learning methods accuracy, the current AD prog- nosis accuracy remains limited. In this paper, we propose to combine both high quality biomarkers and advanced learning method. Our ap- proach is based on a robust ensemble learning strategy using gray matter grading. The estimated weak classifiers are then fused into high infor- mative anatomical sub-ensembles. Through a sparse logistic regression, the most relevant anatomical sub-ensembles are selected, weighted and used as input to a global classifier. Validation on the full ADNI1 dataset demonstrates that the proposed method obtains competitive results of prediction of conversion to AD in the Mild Cognitive Impairment group with an accuracy of 75.6%.