Alzheimer's Disease Neuroimaging Initiative (ADNI) (2011): Discrimination of AD and normal subjects from MRI: Anatomical versus statistical regions

GTC, Aragon Institute of Engineering Research, Universidad de Zaragoza, Zaragoza, Spain.
Neuroscience Letters (Impact Factor: 2.03). 10/2010; 487(1):113-7. DOI: 10.1016/j.neulet.2010.10.007
Source: PubMed


This work is a feature-extraction and classification study between Alzheimer's disease (AD) patients and normal subjects. Voxel-wise morphological features of brain MRI are defined as the Jacobian determinants that measure the local volume change between each subject and a given atlas. The goal of this work is to determine the region of interest (ROI) which is best suited for classification. Two types of ROIs are considered: anatomical regions, that were automatically segmented in the atlas (amygdalae, hippocampi and lateral ventricles); and statistical regions, defined from group comparison statistical maps. Classification performance was assessed with five classifiers on 20 pairs of matched training and test groups of subjects from the ADNI database. In this study the statistical masks provided the best classification performance.

3 Reads
  • Source
    • "Our general objective is to assess the accuracy of our automated high-dimensional morphometry technique to the hypothetical prediction of future clinical status from MRI when examining previously acquired data in a cohort of MCI subjects from the large, multicentric ADNI dataset, compared to the currently known clinical status for these subjects, under various conditions. Specifically, we will want to test the following hypotheses, which would need to hold true for any methodology: (a) that intensity standardization and tissue classification improve the system's robustness and hence performance , in a multicentric setting; (b) that a medial temporal lobe VOI is the best for the differentiation of CTRL from either probable AD or MCI progressing to probable AD, as opposed to whole-brain VOIs [43]; (c) that the methodology remains highly accurate even with large, ostensibly heterogeneous datasets. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Introduction. Medial temporal lobe atrophy assessment via magnetic resonance imaging (MRI) has been proposed in recent criteria as an in vivo diagnostic biomarker of Alzheimer's disease (AD). However, practical application of these criteria in a clinical setting will require automated MRI analysis techniques. To this end, we wished to validate our automated, high-dimensional morphometry technique to the hypothetical prediction of future clinical status from baseline data in a cohort of subjects in a large, multicentric setting, compared to currently known clinical status for these subjects. Materials and Methods. The study group consisted of 214 controls, 371 mild cognitive impairment (147 having progressed to probable AD and 224 stable), and 181 probable AD from the Alzheimer's Disease Neuroimaging Initiative, with data acquired on 58 different 1.5 T scanners. We measured the sensitivity and specificity of our technique in a hierarchical fashion, first testing the effect of intensity standardization, then between different volumes of interest, and finally its generalizability for a large, multicentric cohort. Results. We obtained 73.2% prediction accuracy with 79.5% sensitivity for the prediction of MCI progression to clinically probable AD. The positive predictive value was 81.6% for MCI progressing on average within 1.5 (0.3 s.d.) year. Conclusion. With high accuracy, the technique's ability to identify discriminant medial temporal lobe atrophy has been demonstrated in a large, multicentric environment. It is suitable as an aid for clinical diagnostic of AD.
    International Journal of Alzheimer's Disease 08/2014; 2014:278096. DOI:10.1155/2014/278096
  • Source
    • "Some approaches classically used for FS in neuroimaging include univariate methods (e.g., -test, Anova, Wilcoxon) as filters to select features for classification ([9], [10]), as well as multivariate approaches, e.g., recursive feature elimination ([7], [11]), hybrid FS and nonlinear SVM classification [12], reverse feature elimination methods [13], sparse logistic regression [14], and perturbation method [15]. Additionally, alternative feature extraction approaches based on neuroanatomical landmarks have also been applied to neuroimaging (e.g., using summarizations from regions of interest, as in [16]). The later approach can produce interesting results when there is prior knowledge about anatomical regions or brain tissues (i.e., gray or white matter) involved in the specific disorder studied. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Feature selection (FS) methods play two important roles in the context of neuroimaging based classification: potentially increase classification accuracy by eliminating irrelevant features from the model and facilitate interpretation by identifying sets of meaningful features that best discriminate the classes. Although the development of FS techniques specifically tuned for neuroimaging data is an active area of research, up to date most of the studies have focused on finding a subset of features that maximizes accuracy. However, maximizing accuracy does not guarantee reliable interpretation as similar accuracies can be obtained from distinct sets of features. In the current paper we propose a new approach for selecting features: SCoRS (survival count on random subsamples) based on a recently proposed Stability Selection theory. SCoRS relies on the idea of choosing relevant features that are stable under data perturbation. Data are perturbed by iteratively sub-sampling both features (subspaces) and examples. We demonstrate the potential of the proposed method in a clinical application to classify depressed patients versus healthy individuals based on functional magnetic resonance imaging data acquired during visualization of happy faces.
    IEEE Transactions on Medical Imaging 01/2014; 33(1):85-98. DOI:10.1109/TMI.2013.2281398 · 3.39 Impact Factor
    • "In a typical example of machine learning applied to brain MRI, an algorithm is trained on a set of MRI features, such as regional volumes and cortical thickness, to create a classifier which predicts the correct diagnostic outcome for new observations. To reduce the high dimensionality of the MRI features and to stabilize predictions, shrinkage methods, such as principal components analysis or partial least squares have sometimes been added prior to the training step (Franke et al., 2010; Pelaez-Coca et al., 2011; Phan et al., 2010; Teipel et al., 2007). MRI data pose an especially challenging problem for predictions because of the difficulty in finding a good representation of brain features that makes them easily extractable and reveals their essential structure. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Modern machine learning algorithms are increasingly being used in neuroimaging studies, such as the prediction of Alzheimer's disease (AD) from structural MRI. However, finding a good representation for multivariate brain MRI features in which their essential structure is revealed and easily extractable has been difficult. We report a successful application of a machine learning framework that significantly improved the use of brain MRI for predictions. Specifically, we used the unsupervised learning algorithm of locally linear embedding (LLE) to transform multivariate MRI data of regional brain volume and cortical thickness to a locally linear space with fewer dimensions, while also utilizing the global nonlinear data structure. The embedded brain features were then used to train a classifier for predicting future conversion to AD based on a baseline MRI. We tested the approach on 413 individuals from the Alzheimer's Disease Neuroimaging Initiative (ADNI) who had baseline MRI scans and completeclinical follow-ups over 3 years with following diagnoses: Cognitive normal (CN; n= 137), stable mild cognitive impairment (s-MCI; n=93), MCI converters to AD (c-MCI, n=97), and AD (n=86). We found classifications using embedded MRI features generally outperformed (p<0.05) classifications using the original features directly. Moreover, the improvement from LLE was not limited to a particular classifier but worked equally well for regularized logistic regressions, support vector machines, and lineardiscriminant analysis. Most strikingly, using LLE significantly improved (p = 0.007) predictions of MCI subjects who converted to AD and those who remained stable (accuracy/sensitivity/specificity: = 0.68/0.80/0.56). In contrast, predictions using the original features performed not better than by chance (accuracy/sensitivity/specificity: = 0.56/0.65/0.46). In conclusion, LLE is a very effective tool for classification studies of AD using multivariate MRI data. The improvement in predicting conversion to AD in MCI could have important implications for health management and for powering therapeutic trials by targeting non-demented subjects who later convert to AD.
    NeuroImage 06/2013; 83. DOI:10.1016/j.neuroimage.2013.06.033 · 6.36 Impact Factor
Show more