Ensemble sparse classification of Alzheimer's disease

Abstract

The high-dimensional pattern classification methods, e.g., support vector machines (SVM), have been widely investigated for analysis of structural and functional brain images (such as magnetic resonance imaging (MRI)) to assist the diagnosis of Alzheimer's disease (AD) including its prodromal stage, i.e., mild cognitive impairment (MCI). Most existing classification methods extract features from neuroimaging data and then construct a single classifier to perform classification. However, due to noise and small sample size of neuroimaging data, it is challenging to train only a global classifier that can be robust enough to achieve good classification performance. In this paper, instead of building a single global classifier, we propose a local patch-based subspace ensemble method which builds multiple individual classifiers based on different subsets of local patches and then combines them for more accurate and robust classification. Specifically, to capture the local spatial consistency, each brain image is partitioned into a number of local patches and a subset of patches is randomly selected from the patch pool to build a weak classifier. Here, the sparse representation-based classifier (SRC) method, which has shown to be effective for classification of image data (e.g., face), is used to construct each weak classifier. Then, multiple weak classifiers are combined to make the final decision. We evaluate our method on 652 subjects (including 198 AD patients, 225 MCI and 229 normal controls) from Alzheimer's Disease Neuroimaging Initiative (ADNI) database using MR images. The experimental results show that our method achieves an accuracy of 90.8% and an area under the ROC curve (AUC) of 94.86% for AD classification and an accuracy of 87.85% and an AUC of 92.90% for MCI classification, respectively, demonstrating a very promising performance of our method compared with the state-of-the-art methods for AD/MCI classification using MR images.

Figure 1

The framework of the random patch-based subspace ensemble classification method.

Figure 2

Classification accuracies of SVM and SRC with respect to different numbers of top ranked features selected for AD classification.

Figure 3

Classification results using different patch sizes at four sampling rates: (a) 20%, (b) 40%, (c) 60%, and (d) 80%.

Figure 3

Classification results using different patch sizes at four sampling rates: (a) 20%, (b) 40%, (c) 60%, and (d) 80%.

Figure 4

SRCs ensemble classification results using five different sampling rates with the patch size set to 7×7×7 voxels.

Figure 5

Comparison of ensemble classification results using SVM with 9×9×9 patch size and SRC with 7×7×7 patch size at four sampling rates.

Figure 6

The diversity-accuracy diagrams of SVM and SRC classifiers. The patch size, sampling rate and ensemble size are set to 7×7×7, 60% and 20, respectively. The x-axis represents average accuracy of a pair of classifiers, and y-axis represents diversity of a pair of classifiers evaluated by the kappa measure. The blue and red dashed vertical lines show the ensemble accuracies of SRC and SVM, respectively. The blue and red hexagrams denote the centroids of SRC and SVM classifier clouds, respectively.

Figure 7

ROC curves of five different methods for AD classification.

Figure 8

Classification accuracy of SVM and SRC with respect to different numbers of top ranked features selected for MCI classification.