Table 1 Literature Review.
Ref | Model Type | Method / Model Used | Dataset | Innovation | Strength | Research Gap |
|---|---|---|---|---|---|---|
Hybrid DL + ML | CNN + SVM/RF with multimodal data | BRATS | Combined multimodal MRI features with robust selection | High precision and reduced overfitting | Limited to small dataset; lacks real-time validation | |
CNN (Explainable) | CNN-TumorNet + XAI | Custom MRI | Integrates explainability into CNN | Enhances trust and interpretability | Requires clinical deployment and scalability testing | |
CNN | Optimized CNN | BRATS 2020 | Automated hyperparameter tuning | High consistency across folds | Needs comparison with TL models | |
CNN + ML | AlexNet + SVM/KNN/RF | BRATS MRI | Combines DL features with ML classifiers | Better feature generalization | Feature fusion can be further optimized | |
Transfer Learning | ResNet, VGG16 | BRATS, Kaggle | Uses TL for MRI with fine-tuning | High accuracy with low training cost | Dataset imbalance not addressed | |
Hybrid DL + ML | CNN + XAI + ML | BRATS, TCIA | Introduces explainability with hybrid fusion | Very high accuracy and interpretability | High computational cost | |
ML (Unsupervised + SVM) | K-means + SVM | Custom MRI | Combines segmentation and ML classification | Improved edge detection | Not end-to-end DL; lacks automation | |
CNN + Attention | CNN + Soft Attention | Kaggle MRI | Attention mechanism for ROI focus | Better tumor localization | High model complexity | |
CNN | CNN | BRATS, Harvard Dataverse | Improved CNN architecture | Robust detection accuracy | Limited explainability | |
ML | SVM + Novel Feature Extraction | BRATS 2021 | Unique handcrafted features | Simple and interpretable | Lower performance than CNN | |
Hybrid | CNN + Random Forest | Kaggle MRI | Combines CNN and ML | Balanced accuracy and efficiency | Needs end-to-end optimization | |
CNN | CNN | BRATS 2020 | End-to-end automated CNN pipeline | Robust classification | No interpretability layer | |
Hybrid DL + ML | CNN + SVM + Genetic Algorithm | BRATS 2020 | Genetic feature selection | High efficiency | Computationally heavy | |
Ensemble DL + ML | CNN Ensemble + SVM/RF | BRATS 2018 | Feature-level ensemble | Improved robustness | Needs real-time evaluation | |
CNN + Segmentation | ResNet50 + U-Net | TCGA-LGG, TCIA | Combines classification & segmentation | Superior localization & accuracy | High GPU requirements | |
Hybrid DL + ML | CNN + ML | BRATS 2021 | Fuses ML with DL for optimization | Reduced complexity | Limited transferability | |
CNN (Explainable) | Explainable CNN + Grad-CAM | BRATS 2020 | Visual interpretability integration | High clinical potential | Lower accuracy vs. non-XAI models | |
Hybrid | CNN-SVM + PSO | Kaggle MRI | PSO optimization for feature weights | Improved hybrid performance | Limited generalization | |
Ensemble DL | VGG + ResNet + DenseNet | BRATS | Multi-architecture ensemble | Exceptional classification accuracy | High computational cost | |
CNN | EfficientNet | Custom MRI | EfficientNet optimization | Lightweight and accurate | Needs more diverse data | |
Ensemble DL + ML | CNN + Ensemble ML | BRATS 2019 | Combined multiple feature sources | Enhanced model stability | Lacks real-time validation | |
Transfer Learning | DenseNet121 | BRATS, TCGA | Transfer learning with DenseNet | Accurate on small data | Model explainability missing | |
CNN | CNN + Optimized Parameters | BRATS | Classifier optimization | Improved efficiency | Dataset generalization needed | |
Hybrid DL + ML | CNN + SVM/KNN/RF | BRATS 2021 | CNN feature extraction + ML classification | Easy to interpret | Lower accuracy than TL models | |
Transfer Learning | CNN + TL + SVM | BRATS | Combined TL and ML techniques | Improved adaptability | Limited dataset variation | |
CNN + GDD | Deep Learning + GDD | Kaggle MRI | Introduces GDD feature approximation | Accurate classification & survival prediction | Complex preprocessing stage |
