Table 1 Some recent state-of-the-art works in the field of Alzheimer’s disease detection.
From: A hybrid filtering and deep learning approach for early Alzheimer’s disease identification
Author (s) & Year | Work done | Dataset | Performance | Visualization | Limitations |
|---|---|---|---|---|---|
Metrics | with XAI | ||||
Hybrid model merging | ADNI | Hybrid model achieved | x | Most current models performed | |
LeNet and AlexNet | 93.58% accuracy | less than 90% in classification task | |||
CNN and LSTM | Kaggle | Attained an accuracy | x | Weight decay issues may arise, | |
of 98.5% | optimal solution does not cover | ||||
VGG-16-based CNN | ADNI | Achieved an accuracy | x | Classifying pMCI and sMCI is | |
with Transformer | of 77.2% | difficult due to subtle differences | |||
Customized AlexNet & | ADNI | Gained accuracy of 96.61% | x | Model’s interpretability & | |
InceptionV2 architecture | and AUC of 0.9663 | explainability limited | |||
CNN architecture | OASIS | Achieved an accuracy of | x | Lack of generalizability & | |
99.68% | interpretability | ||||
Lightweight DL Model | Kaggle | Achieved an accuracy of | x | Potential overfitting due to | |
95.93% | the small dataset size | ||||
Enhanced EfficientNetB7 | ADNI | Achieved an accuracy of | x | Lack of interpretability | |
98.2% | |||||
Densenet201, EfficientNet | Kaggle | Achieved an accuracy of | Yes | Combination of different TL | |
and AlexNet | 99.83% | prone to rise complexity | |||
CNN with RNN | ADNI | Achieved an accuracy of | Yes | Potential for overfitting due | |
98.45% | to complex models | ||||
Ensemble DL models | Kaggle | Achieved an accuracy of | Yes | Potential biases in dataset | |
96% | |||||
U-net+GAN | ADNI | Achieved an accuracy of | Yes | Scalability challenges with multiple models | |
95% | |||||
Attention DL model | Kaggle | Achieved an accuracy of | Yes | Subpar performance in specific cases | |
95.28% |
