Showing papers by "Pierrick Coupé published in 2014"

Nonlocal intracranial cavity extraction

Abstract: Automatic and accurate methods to estimate normalized regional brain volumes from MRI data are valuable tools which may help to obtain an objective diagnosis and followup of many neurological diseases. To estimate such regional brain volumes, the intracranial cavity volume (ICV) is often used for normalization. However, the high variability of brain shape and size due to normal intersubject variability, normal changes occurring over the lifespan, and abnormal changes due to disease makes the ICV estimation problem challenging. In this paper, we present a new approach to perform ICV extraction based on the use of a library of prelabeled brain images to capture the large variability of brain shapes. To this end, an improved nonlocal label fusion scheme based on BEaST technique is proposed to increase the accuracy of the ICV estimation. The proposed method is compared with recent state-of-the-art methods and the results demonstrate an improved performance both in terms of accuracy and reproducibility while maintaining a reduced computational burden.

Optimized patchMatch for near real time and accurate label fusion.

Abstract: Automatic segmentation methods are important tools for quantitative analysis of magnetic resonance images. Recently, patch- based label fusion approaches demonstrated state-of-the-art segmentation accuracy. In this paper, we introduce a new patch-based method using the PatchMatch algorithm to perform segmentation of anatomical structures. Based on an Optimized PAtchMatch Label fusion (OPAL) strategy, the proposed method provides competitive segmentation accuracy in near real time. During our validation on hippocampus segmentation of 80 healthy subjects, OPAL was compared to several state-of-the-art methods. Results show that OPAL obtained the highest median Dice coefficient (89.3%) in less than 1 sec per subject. These results highlight the excellent performance of OPAL in terms of computation time and segmentation accuracy compared to recently published methods.

Nonlocal regularization for active appearance model: Application to medial temporal lobe segmentation.

Abstract: The human medial temporal lobe is an important part of the limbic system, and its substructures play k ey roles in learning, memory, and neurodegeneration. T he medial temporal lobe includes the hippocampus, amygdala, parahippocampal cortex, entorhinal cortex, and perirhinal cortex ‐ structures that are compl ex in shape and have low between-structure intensity c ontrast, making them difficult to segment manually in magnetic resonance images. This paper presents a new segmentation method that combines active appearance modeling and patch-based local refinement to automatically segme nt specific substructures of the medial temporal lo be including hippocampus, amygdala, parahippocampal cortex, and entorhinal/perirhinal cortex from MRI data. Appearance modeling, relying on eigen-decomposition to analyze statistical variations in image intensity and shape information in study population , is used to capture global shape characteristics o f each structure of interest with a generative model. Patc h-based local refinement, using nonlocal means to compare the image local intensity properties, is ap plied to locally refine the segmentation results al ong the structure borders to improve structure delimitation . In this manner, nonlocal regularization and globa l shape constraints could allow more accurate segment ations of structures.

NICE: Non-local Intracranial Cavity Extraction

Abstract: Automatic and accurate methods to estimate normalized regional brain volumes from MRI data are valuable tools which may help to obtain an objective diagnosis and follow-up of many neurological diseases. To estimate such regional brain volumes, the Intracranial Cavity Volume (ICV) is often used for normalization. However, the high variability of brain shape and size due to normal inter-subject variability, normal changes occurring over the lifespan, and abnormal changes due to disease makes the ICV estimation problem challenging. In this paper, we present a new approach to perform ICV extraction based on the use of a library of pre- labeled brain images to capture the large variability of brain shapes. To this end, an improved non-local label fusion scheme based on BEaST technique is proposed to increase the accuracy of the ICV estimation. The proposed method is compared with recent state-of-the-art methods and the results demonstrate an improved performance both in terms of accuracy and reproducibility while maintaining a reduced computational burden.

Anatomically Constrained Weak Classifier Fusion for Early Detection of Alzheimer’s Disease

Abstract: 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 numerous efforts have been devoted to improve MRI-based feature quality or to increase machine learning methods accuracy, the current AD prognosis accuracy remains limited. In this paper, we propose to combine both high quality biomarkers and advanced learning method. Our approach is based on a robust ensemble learning strategy using gray matter grading. The estimated weak classifiers are then fused into high informative 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%.

A unified assessment of fully automated hippocampus segmentation methods

Abstract: changes. Methods: Participants were 211 cognitively normal (clinical dementia rating [CDR]1⁄40, Montreal Cognitive Assessment [MOCA]>24 ) volunteers aged at least 60 years with a family history of AD in a first-degree relative. Persons aged 55-59 were also admitted if their relative had dementia onset within 15 years of their own age. Participants were evaluated cognitively using the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) without age adjusted norms. We investigated participants’ sense of smell using the University of Pennsylvania Smell Identification Test (UPSIT).26 participants were enrolled in a randomized trial and presently have 3-month cognitive follow-up data available. Results: Participants were 72.5% women; 32% were APOE ε4 carriers. Mean age was 64.5 6 s.d. 7.5 yrs, mean education was 15.3 6 s.d. 3.5 years; mean MOCA score at eligibility was 28.1 6 s.d. 1.6. In a subset of 89 participants, we found that the UPSIT was significantly associated at baseline with the RBANS immediate memory, delayed memory, and total index scores (Table 1). In an overlapping subset of 103 participants with MRI scans, we found that the UPSIT was significantly associated with hippocampal volumes bilaterally (Table 1). In the subset of participants with 3-month RBANS follow-up scores, we found that low UPSIT at baseline predicted a decline in attention sub-score (r1⁄4-0.429 sig1⁄40.037 df1⁄422) after controlling for age, and gender. Conclusions: Observations to date suggests that people with good higher-order processing of odors have better cognitive scores and increased hippocampal volumes. Our data substantiate other reports of olfactory identification abnormalities in pre-clinical Alzheimer’s disease and suggest the potential utility of olfactory testing as a ‘‘biomarker‘‘ to track progress of pre-symptomatic AD.

Non Local Spatial and Angular Matching : a new denoising technique for diffusion MRI

Abstract: Diffusion Weighted Images (DWIs) datasets suffer from low Signal-to-Noise Ratio (SNR), especially at high b-values. Acquiring data at high b-values contains relevant information and is now of great interest for connectomics studies. High noise levels bias the measurements because of the non-Gaussian nature of the noise, which in turn can lead to a false and biased estimation of the diffusion parameters. Therefore, high SNR DWIs is important in order to draw meaningful conclusions in subsequent data or group analyses. The acquired DWIs differ between themselves, but still share the same underlying structure. It is also known that natural images are redundant and can be sparsified. We thus propose to use the redundancy of DWIs as a sparse representation to reduce the noise level and achieve a higher SNR using dictionary learning and sparse coding, without the need for additional acquisition time. We show quantitative results on the ISBI 2013 HARDI challenge phantom.

Impact of DWI denoising on Track-Density Imaging

Abstract: Track-Density Imaging (TDI) has been proposed as super-resolution method for diffusion-weighted images (DWI). Based on fiber-tracking (FT), TDI enables to produce high-resolution white matter images. However, the sensitivity to noise of FT negatively impacts TDI accuracy and reproducibility. In this study, we analyzed whether the use of DWI denoising is able to improve TDI robustness and reproducibility.