Table 1 Summary of the techniques used for tumor classification and detection.
From: Detection and classification of brain tumor using a hybrid learning model in CT scan images
References | Methodology | Dataset | Strengthen/Result | Limitation |
|---|---|---|---|---|
Alamin et al.32 | ResNet50V2 and Fine-tuned by attaching layers | Figshare MRI brain tumor dataset | The model is highly effective in accurately Diagnosing brain tumors | The main shortcoming of this work is the lack of sharper images and a better DL architecture, which prevents even better performance outcomes |
Takowa et al.33 | Parallel deep convolutional neural network | Figshare dataset | Extracts both low-level and high-level features; the structure is accurate and efficient | The accuracy of each category of tumor is not mentioned, resizing to 32 × 32 is so small image |
Kumar et al.34 | Modified ResNet50 and HOG | Rembrandt Dataset | Optimal computational efficiency, low complexity, ResNet50, the proposed model will converge more quickly if identity is nearer the optimal value | Low accuracy and performance compared to existing methods. They only performed on MRI images |
Fatima et al.35 | Fine-tuned EfficientNetB2 | CE-MRI brain tumor dataset | Generalizability using cross-validation, Improving the data appeared to have potential for improving the usefulness of the improved EffectiveNetB2. | They only performed on MRI images, high network complexity, Imbalance data, They only performed on MRI images. |
Muhammad et al.36 | Combination of GoogLeNet, ResNet, and Desnet121 models | Figshare dataset | Used statistical and image processing methods to improve an image’s visual quality. Methods of data augmentation reduce overfitting of the network | It is not generalizable; a more representative dataset is required, and They only performed on MRI images |
Priya et al.2 | Hybrid Alexnet-Gated Recurrent Unit (Alexnet-GRU) | Kaggle datasets of Brain MRI Images | While dropout layers prevent overfitting and maintain the integrity of the feature map, GRU overcomes the gradient vanishing problem and improves model performance | They only performed on MRI images, AlexNet-GRU challenges in obtaining high-quality, diverse, and annotated datasets, especially for rare cancer, to accurately label brain tumor images |
Rastogi et al.7 | Multi-Branch Networks with Inception Block and five-fold cross validation | Br35H dataset | Its supremacy over the existing state-of-the-art models in the field | limited data, Complex architecture, No mentioning of training and testing time, The accuracy of each category of tumor is not mentioned |
Kar et al.31 | Deep neural network with multi-head attention mechanism | Brain Tumor MRI 44c, Brain Tumor MRI 17c, and “Brain Tumor Multimodal CT and MRI | Integrates CT and MRI with multi-head attention to achieve high accuracy and robust multimodal brain tumor classification | Lacks validation on real hospital-acquired clinical datasets, limiting its generalizability to real-world settings |
Loganayagi et al.37 | Hybrid Deep Maxout-VGG-16 model | Figshare dataset | Combines advanced segmentation and feature extraction with DM-VGG-16 to achieve strong tumor detection accuracy on MRI | Model performance (≈ 90%) is lower than state-of-the-art and lacks multimodal validation or clinical deployment evidence |
Amreen38 | Proposes a multi-path CNN with varied kernel sizes and feature selection (Chi-square, mutual info, correlation) to classify brain tumors from MRI scans | Kaggle datasets of Brain MRI Images | Achieves high multiclass classification accuracy (96.03%) with fewer trainable parameters using a lightweight multi-path CNN | Needs further validation on larger clinical datasets and doesn’t incorporate multimodal imaging |
