Refs. | Methodology | Features | Database | Advantages | Limitations |
---|---|---|---|---|---|
Nazir, et al. (2019) | Convolutional neural network (CNN) | Automatic deep features | CT images | Specificity 99%, sensitivity 95%, and accuracy 98% claimed | No benchmark datasets were experimented. Only local dataset used. Difficult to generalize results |
Amin et al. (2020a) | SVM and 3D–2D hybrid CNN combination | Spectral and Spatial features | vivo HS brain cancer dataset | 80% accuracy attained | No benchmark dataset was experimented. Also accuracy is low |
Saba et al. (2018b) | 2-level histogram-based morphometry (HBM) | 3D histogram of oriented gradients (HOG) | sMRI | (AUC) of > 0.75 in each dataset, highest AUC of 0.849 in case of ETH site | Traditional machine learning used that is not suitable for large datasets. Also accuracy is low |
Ma et al. (2022) | Normal Pressure Hydrocephalus segmentation and classification (NPH-SC) model and CCN | Skull features and segmentation by marker-based watershed approach | MRI Brain | NPH-SC model: sensitivity 96%, specificity 100%, validation accuracy 97%. CNN test accuracy 98% | Experiments are performed on small dataset & validity of the approach seems not good on large dataset |
Mathur et al. (2020) | CNN | geometric data augmentation techniques | MRIs | Accuracy 98.8% and specificity 99% | No benchmark dataset used & experiments performed on augmented data |
Jabeen et al. (2018) | Feedforward multilayer neural network | PCA + ANN | MRIs | Accuracy 75% | Small dataset employed for experiments |