Table 1 Summary of brain stroke diagnosis studies using machine learning and deep learning (2024–2025).
From: An effective brain stroke diagnosis strategy based on feature extraction and hybrid classifier
Author | Year | Method | Dataset | Accuracy/performance | Dataset source/notes |
|---|---|---|---|---|---|
Abdi et al. | 2025 | CNN with interpretability (LIME, saliency maps) | 2501 CT images (internal), 9900 (external) | 97.2% (internal), 89.73% (external) | Internal + external datasets |
Yu et al. | 2025 | Multimodal DL (facial motion + speech) | Video data (CPSS/FAST emulation) | Sensitivity = 93.12%, Accuracy = 79.27% | Simulated/emulated dataset |
Dhakan et al. | 2025 | Ensemble ML model on structured data | 5110 patient records | Outperformed individual classifiers | Structured tabular data |
DeepRETStroke | 2025 | DL on retinal fundus images | 895,000 + retinal images | AUC = 0.901 | Public dataset (retinal images) |
Abulfaraj et al. | 2024 | SqueezeNet v1.1, MobileNet V3-Small, CatBoost | Brain Stroke CT Image Dataset | 99.1% | Public (Kaggle) |
Sabir and Ashraf | 2024 | DCNN model with feature fusion | Brain CT images | 96.5% | Not specified |
Dubey et al. | 2024 | Boosting algorithms (GB, ADB, XGB), explainable via LIME and SHAP | Brain CT images | 96.97% (train), 92.13% (test) | Not specified |
Polamuri et al. | 2024 | Enhanced CNNs (DenseNet121, ResNet50, VGG16) | MRI images | Outperformed baseline CNNs | Not specified |
Khan et al. | 2024 | Survey of CNN, transformer, and hybrid models for segmentation | CT and MRI images | Qualitative review | Review paper |
