From: XGBoost-SHAP-based interpretable diagnostic framework for alzheimer’s disease
Authors | Method | Subjects | Modalities | Performance | Limitation | |
---|---|---|---|---|---|---|
Lin et al. (2020) [20] | ELM | 110 pMCI vs. 205 sMCI | MRI, PET, CSF, and genes | ACC: 84.7% | The multiclassification task for AD failed. | |
Tufail et al. (2020) [21] | DL | 90 nonAD vs. 90 AD | sMRI | Fluctuated accuracy: 0.62–0.65 | ||
Akter et al. (2022) [22] | SMOTE-ENN + AD-CovNet (long short-term memory-multilayer perceptron) | 754 AD individuals with and without COVID-19 | Demographic and clinical from medical records | ACC: 86% AUC: 0.857 | The method of imbalanced data potentially causes data leakage (overoptimism or overfitting). | |
Lin et al. (2021) [23] | LDA + ELM | 200 NC vs. 441 MCI vs. 105 AD | MRI, PET, CSF, and genes | ACC: 66.7% F1 score: 0.649 | The 3 way classification performance is poor due to imbalanced problem. | |
Basheera et al. (2020) [26] | CNN | 28 NC vs. 32 MCI vs. 65 AD | MRI | ACC | AD-NC: 100% AD-MCI: 96.2% NC-MCI: 100% AD-MCI-NC: 86.7% | The internal mechanism is complex and poorly interpretable. |
Hu et al. (2021) [27] | DL-based networks | 823 NC vs. 552 FTD vs. 422AD | MRI | ACC | FTD vs. FTD_NC: 93.45% AD vs. AD_NC: 89.86% FTD vs. AD vs. NC: 91.83% FTD vs. AD: 93.05% |