Table 1 Comparison of AD classification methods
Author (year) | Image modality | Pre-processing | Method | Advantage | Disadvantage |
|---|---|---|---|---|---|
Liu et al. [6] | MRI, PET | Feature extraction & selection | Autoencoder | Extracted high-level features | Difficulties in gathering various imaging modality and numerical data |
Suk et al. [7] | MRI, PET, CSF | Gray matter/white matter segmentation, feature extraction | Stacked autoencoder | Extracted and fused high-level features | |
Basheera et al. [9] | MRI | Gray matter segmentation | CNN | Focused on gray matter features | Require a precise professional knowledge |
Choi et al. [10] | MRI | Hippocampus segmentation | CNN | Improved performance using small patches as input | |
Wang et al. [8] | MRI | Spatial and intensity normalization | CNN + RELU + max pooling | Improved performance of CNN | Requires evaluation with different image acquisition environment dataset |
Feng et al. [11] | MRI, PET | Gray matter segmentation | 3D CNN + LSTM | Obtained spatial information | 3D model requires a number of image datasets for training |
Huang et al. [12] | T1-MR, FDG-PET | Hippocampus segmentation | 3D CNN | Integrated T1 weighted MR and FDG-PET as input | |
Liu et al. [13] | MRI PET | Spatial and intensity normalization | 3D CNN + cascaded 2D CNN | Extracted multi-level and multi-modal features |