Advance of AI-Based Predictive Models for Diagnosis of Alzheimer's Disease (AD) in Healthcare
Table 1.
Summary on Machine Learning-Driven for EarlyDetection of Alzheimer's Disease in Healthcare
|
| References |
Methodology |
Dataset |
Performance |
Limitations & Future Work |
|
|
| Afzal et al., 2019 |
Transfer learning + data augmentation on 3D MRI |
OASIS |
98.41% (single view), 95.11% (3D view) |
Class imbalance in dataset; need for balanced multiclass classification of AD stages |
|
| Ahmed et al., 2019 |
CNN ensemble on TVPs of left/right hippocampus |
GARD |
90.05% accuracy |
Focused only on hippocampus; small dataset; overfitting still a concern |
|
| Silva et al., 2019 |
Deep feature extraction + classical ML classifiers (RF, SVM, K-NN) |
MIRIAD |
RF: 88.32%, SVM: 96.07%, K-NN: 87.45% |
Limited region of brain (30 slices); classification only between AD vs. HC |
|
| Altaf et al., 2018 |
Hybrid of clinical + texture features; BoVW model |
ADNI |
Binary: 98.4%, Multi-class: 79.8% |
Moderate performance in multi-class classification; relies on handcrafted features |
|
| Mahyoub et al., 2018 |
ML classifiers on lifestyle, demography, and medical history |
Custom tabular dataset |
Sensitivity: 0.741, Specificity: 0.515 (test) |
Poor generalization; low precision; limited data modalities |
|
| Padole et al., 2018 |
Graph CNN using resting-state fMRI |
ADNI |
92.44% accuracy |
Focused only on fMRI; computationally intensive graph construction |
|
|
|
