Docteur de l'université Automatic Segmentation and Shape Analysis ...

More documents

Recommendations

Info

110 Chapter 5 Quantitative shape analysis of hippocampus in AD separation between AD and NC increases the discrimination power of SSM. The predictors extracted on the subregions also showed stronger correlation with the memory-related measurements such as Logical Memory, Auditory Verbal Learning Test (AVLT) and the memory subscores of Alzheimer Disease Assessment Scale (ADAS). The results of the localization of the SSM improving the discrimination of shape analysis have been previously partially published in “Increasing the discrimination power of shape analysis in Alzheimer’s disease detection by localized statistical shape model,” in Alzheimer’s Association International Conference on Alzheimer’s Disease (ICAD), 2011, and “Detecting hippocampal shape changes in Alzheimer’s disease using statistical shape models,” in Medical Imaging 2011: Image Process- ing. Full results have been submitted to the journal NeuroImage for publication. 5.1 Overview of the method We aim to extract hippocampal shape descriptors capturing both global and local shape changes linked to the AD pathology by SSM, so as to improve the performance of disease classification using additional shape information. We try to model the localized disease-related shape changes by performing the shape analysis upon the regions that are found to be affected by AD. We first identify the hippocampal surface subregions that are significantly different between the AD group and the controls via Hotelling’s T 2 test. Then we model the variations on these subregions using SSMs. Compared to the shape analysis of hippocampi via LoCA (Xie et al., 2009) which localizes the shape components while not specifically targeting the area related to the disease, we focus our shape analysis only on the regions that we marked as different, extracting the principal shape components on these regions. With the use of machine learning techniques, classifiers are able to learn the difference between AD and NC from the shape information as quantified by shape models. Extra information concerning AD pathology can be obtained by adding hippocampal shape descriptors in addition to the volumetry. We use the
Chapter 5 Quantitative shape analysis of hippocampus in AD 111 morphological variation on these regions as variables to describe the AD pathology as they improve the discrimination between the classes and are correlated with the measures of memory decline associated with the disease. We model the morphology of hippocampus by SSMs, and use the shape descriptors derived from the SSM to detect the effect of AD on the hippocampal size and its shape. Descriptors produced by different models serve as features for machine learning algorithms and are evaluated in terms of their prediction performance in distinguishing AD from NC. The method used to extract relevant shape information from SSMs can be divided into two steps: 1) the localization step, and 2) the shape modeling step. The processing pipeline is shown in Figure 5.1. In the localization step, we build up the correspondence on hippocampal surface over a training set of both NC and AD subjects. Once the correspondence problem is solved, all the landmarks on the hippocampal surfaces are aligned by Procrustes analysis (Gower, 1975). A statistical test can be performed on each landmark to evaluate the significance of the difference between the distributions of aligned points of the NC group and the AD group. The resulting significance map produced by the statistical tests in the localization step can be thresholded to obtain a surface mask of landmarks. In the shape modeling step, we apply these masks to the hippocampal surface to select a subset of hippocampal landmarks separating the subpopulations at a given statistical significance. The selected subregional landmarks are again aligned by Procrustes analysis and a PCA is performed on the aligned subregional landmarks. The coefficient of principal components describing local shape of hippocampus can thus be calculated. In both the localization step and the shape modeling step, the shapes or selected shape patches are aligned by Procrustes analysis, which can be performed either through rigid-body transformations or through similarity transformations. In our experiments, both alignments are performed at each step and their performances are evaluated and compared. We evaluate the subregional shape models based
Page 1:
UNIVERSITE DE BOURGOGNE U.F.R. SCIE
Page 5 and 6:
Résumé L’objectif de cette thè
Page 7 and 8:
Abstract The aim of this thesis is
Page 9 and 10:
TABLE DES MATIÈRES Acknowledgement
Page 11:
TABLE DES MATIÈRES xi 4.1.4.3 Spec
Page 15 and 16:
LISTE DES FIGURES 1.1 Cortical atro
Page 17:
LISTE DES FIGURES xvii 5.6 The accu
Page 20 and 21:
2 Chapter 1 Introduction The resear
Page 22 and 23:
4 Chapter 1 Introduction .β.α .γ
Page 24 and 25:
6 Chapter 1 Introduction 1.1.2 Hipp
Page 26 and 27:
8 Chapter 1 Introduction Figure 1.4
Page 28 and 29:
10 Chapter 1 Introduction Functiona
Page 30 and 31:
12 Chapter 1 Introduction backgroun
Page 32 and 33:
14 Chapter 1 Introduction Figure 1.
Page 35 and 36:
Chapter 2 Literature Review In this
Page 37 and 38:
Chapter 2 Literature Review 19 Figu
Page 39 and 40:
Chapter 2 Literature Review 21 Figu
Page 41 and 42:
Chapter 2 Literature Review 23 does
Page 43 and 44:
Chapter 2 Literature Review 25 inte
Page 45 and 46:
Chapter 2 Literature Review 27 In t
Page 47 and 48:
Chapter 2 Literature Review 29 Wolz
Page 49 and 50:
Chapter 2 Literature Review 31 . Pr
Page 51 and 52:
Chapter 2 Literature Review 33 The
Page 53 and 54:
Chapter 2 Literature Review 35 2.2.
Page 55 and 56:
Chapter 2 Literature Review 37 2.2.
Page 57 and 58:
Chapter 2 Literature Review 39 Apar
Page 59:
Chapter 2 Literature Review 41 spec
Page 62 and 63:
44 Chapter 3 Hippocampal segmentati
Page 64 and 65:
Page 66 and 67:
Page 68 and 69:
Page 70 and 71:
Page 72 and 73:
Page 74 and 75:
Page 76 and 77:
Page 78 and 79: 60 Chapter 3 Hippocampal segmentati
Page 90 and 91: 72 Chapter 4 Statistical shape mode
Page 118 and 119: 100 Chapter 4 Statistical shape mod
Page 127: Chapter 5 Quantitative shape analys
Page 131: Chapter 5 Quantitative shape analys
Page 134 and 135: 116 Chapter 5 Quantitative shape an
Page 155 and 156: Chapter 6 Conclusions Topics in med
Page 157 and 158: Chapter 6 Conclusions 139 Given the
Page 159: Chapter 6 Conclusions 141 6.2.3 Sha
Page 162 and 163: 144 LIST OF PUBLICATIONS using dive
Page 164 and 165: 146 BIBLIOGRAPHY Alzheimer, A. (191
Page 166 and 167: 148 BIBLIOGRAPHY Baillard, C., Hell
Page 168 and 169: 150 BIBLIOGRAPHY of post mortem mag
Page 170 and 171: 152 BIBLIOGRAPHY B. (2008). Three-d
Page 172 and 173: 154 BIBLIOGRAPHY Cootes, T. F., Tay
Page 174 and 175: 156 BIBLIOGRAPHY head in MR images
Page 176 and 177: 158 BIBLIOGRAPHY Evans, A. C., Coll
Page 179 and 180:
BIBLIOGRAPHY 161 Ghanei, A., Soltan
Page 181 and 182:
BIBLIOGRAPHY 163 Heckemann, R. A.,
Page 183 and 184:
BIBLIOGRAPHY 165 R. C. (2003). MRI
Page 185 and 186:
BIBLIOGRAPHY 167 Konrad, C., Ukas,
Page 187 and 188:
BIBLIOGRAPHY 169 Mahony, R. and Man
Page 189 and 190:
BIBLIOGRAPHY 171 disease, mild cogn
Page 191 and 192:
BIBLIOGRAPHY 173 Petersen, R. C., S
Page 193 and 194:
BIBLIOGRAPHY 175 Rorden, C. and Bre
Page 195 and 196:
BIBLIOGRAPHY 177 Shen, K., Bourgeat
Page 197 and 198:
BIBLIOGRAPHY 179 Barillot, C., Hayn
Page 199 and 200:
BIBLIOGRAPHY 181 Tzourio-Mazoyer, N
Page 201 and 202:
BIBLIOGRAPHY 183 Vos, F. M., de Bru
Page 203 and 204:
BIBLIOGRAPHY 185 Woolrich, M. W., J
show all

Docteur de l'université Automatic Segmentation and Shape Analysis ...

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?