Séminaire du 23/04/2015, 14h00

jeudi 23 avril 2015, 14h00

New patch‐based estimators for early detection of Alzheimer's disease

Conférencier : Pierrick Coupé (LABRI)

Résumé : The diagnosis of Alzheimer's disease (AD) at pre-clinical stages or the prediction of conversion of patients with mild cognitive impairment (MCI) to AD is a very challenging problem receiving attention because of the immense associated social and economic costs. Several biomarker candidates have already been studied in depth with the goal of achieving this task. Nowadays, measures of neuronal injury and neurodegeneration are among the most important biomarkers of AD. Cerebral atrophy caused by the progressive neurodegeneration can be measured in detail by magnetic resonance imaging (MRI). Optimizing such MRI-based biomarkers for detection and prediction of AD may have a significant impact on early diagnosis of patients as well as being valuable tools when designing therapeutic studies of individuals at risk of AD to prevent or alter the progression of the disease. In this talk, new patch-based methods will be presented to accurately detect and predict Alzheimer's disease (AD).

First, a new patch-based label fusion (PBL) method for structure segmentation will be detailed. Inspired by recent work in image denoising, our patch-based label fusion involves patch comparison where the weight assigns to each label depends on the similarity between the current patch and the training patch. The search of similar training patches is based on nonlocal strategy to better handle the inter-subject variability and to capture registration inaccuracies. In a limited computational time this method achieves state-of-the-art segmentation accuracy. Consequently, since its introduction, our PBL has been intensively studied and many improvements have been proposed. Some of them will be presented during this talk with of focus of our recent near real PBL method: OPAL. Then, an innovative extension of this method to structure scoring will be discussed. This new method simultaneously performs segmentation and scoring of structures to efficiently capture the anatomical alterations caused by AD. Known as SNIPE (Scoring by Non-local Image Patch Estimator), this scoring measure is based on a nonlocal means framework to estimates the similarity of a new MRI compared to several populations of training MRI. With the nonlocal framework, SNIPE is able to handle inter-subject variability by enabling a one-to-many mapping between the subject's anatomy and those of the training templates. Moreover, by employing the patch-based comparison principle, SNIPE can detect subtle anatomical changes caused by the disease. Finally, a validation of the ADNI database (>800 subjects) will be presented. Detection and prediction aspects will be investigated using several methods. Finally, a comparison with recent studies on MRI-based biomarkers will be discussed.

Références :

• P. Coupé, J. V Manjon, V. Fonov, J. Pruessner, M. Robles, D. L. Collins. Patch-based Segmentation using Expert Priors: Application to Hippocampus and Ventricle Segmentation. NeuroImage, 54(2): 940–954, 2011.
• P. Coupé, S. F. Eskildsen, J. V. Manjon, V. Fonov, D. L. Collins and ADNI. Simultaneous Segmentation and Grading of Anatomical Structures for Patient's Classification: Application to Alzheimer's Disease. NeuroImage, 59(4):3736–3747, 2012
• P. Coupé, S. F. Eskildsen, J. V. Manjon, V. Fonov, J. C. Pruessner, M. Allard, D. L. Collins and ADNI. Scoring by Nonlocal Image Patch Estimator for Early Detection of Alzheimer's Disease. NeuroImage: Clinical, 1(1):141–152, 2012.
• V.-T. Ta, R. Giraud, D. L. Collins, P. Coupé. Optimized PatchMatch for Near Real Time and Accurate Label Fusion. MICCAI'14, pages 105-112, 2014.