Convolutional Neural Networks-Based MRI Image Analysis for the Alzheimer's Disease Prediction From Mild Cognitive Impairment

Abstract

Mild cognitive impairment (MCI) is the prodromal stage of Alzheimer's disease (AD). Identifying MCI subjects who are at high risk of converting to AD is crucial for effective treatments. In this study, a deep learning approach based on convolutional neural networks (CNN), is designed to accurately predict MCI-to-AD conversion with magnetic resonance imaging (MRI) data. First, MRI images are prepared with age-correction and other processing. Second, local patches, which are assembled into 2.5 dimensions, are extracted from these images. Then, the patches from AD and normal controls (NC) are used to train a CNN to identify deep learning features of MCI subjects. After that, structural brain image features are mined with FreeSurfer to assist CNN. Finally, both types of features are fed into an extreme learning machine classifier to predict the AD conversion. The proposed approach is validated on the standardized MRI datasets from the Alzheimer's Disease Neuroimaging Initiative (ADNI) project. This approach achieves an accuracy of 79.9% and an area under the receiver operating characteristic curve (AUC) of 86.1% in leave-one-out cross validations. Compared with other state-of-the-art methods, the proposed one outperforms others with higher accuracy and AUC, while keeping a good balance between the sensitivity and specificity. Results demonstrate great potentials of the proposed CNN-based approach for the prediction of MCI-to-AD conversion with solely MRI data. Age correction and assisted structural brain image features can boost the prediction performance of CNN.

Keywords: Alzheimer’s disease; convolutional neural networks; deep learning; magnetic resonance imaging; mild cognitive impairment.

FIGURE 1

Framework of proposed approach. The dashed arrow indicates the CNN was trained with 2.5D patches of NC and AD subjects. The dashed box indicates Leave-one-out cross validation was performed by repeat LASSO and extreme learning machine 308 times, in each time one different MCI subject was leaved for test, and the other subjects with their labels were used to train LASSO and extreme learning machine.

FIGURE 2

The demonstration of 2.5D patch extraction from hippocampus region. (A–C) 2D patches extracted from transverse (red box), coronal (green box), and sagittal (blue box) plane; (D) The 2.5D patch with three patches at their spatial locations, red dot is the center of 2.5D patch; (E) Three patches are combined into RGB patch as red (red box patch), green (green box patch), and blue (blue box patch) channels.

FIGURE 3

(A) Four random chosen 2.5D patches of one subject (who is normal control, female and 76.3 years old), indicating that these patches contain different information of hippocampus; (B) The comparison of correspond 2.5D patches of four subjects from four groups, the different level of hippocampus atrophy can be found.

FIGURE 4

The overall architecture of the CNN used in this work.

FIGURE 5

The workflow of extracting CNN-based features. The CNN was trained with all AD/NC patches, and used to extract deep features from all 151 patches of MCI subject. The feature number of each patch is reduced to P_C (P_C = 29) from 1024 by PCA. Finally, Lasso selects L_C (L_C = 35) features from P_C × 151 features for each MCI subject.

FIGURE 6

The ROC curves of classifying converters/non-converters when different features used or without age correction.