Table 1 Summary of reviewed studies on brain tumor classification.

Nassar et al.¹⁵

Combination of 5 TL models

3064 T1W-CE MRI (Figshare) images from 233 patients

99.31

Developed an efficient automated method for classifying brain tumors, combining strengths of five different models for improved performance.

Agarwal et al.¹⁶

ODTWCHE and Inception V3

3064 T1W-CE MRI (Figshare) images

98.89

Proposed an Auto Contrast Enhancer, Tumor Detector, and Classifier for improved contrast and early diagnosis of brain tumors, outperforming models like AlexNet, VGG-16, and ResNet-50.

Talukder et al.¹⁷

Xception, ResNet50V2, InceptionResNetV2, DenseNet201

Figshare (3064 MRI)

99.68

Proposed TL models with ResNet50V2 achieving the highest accuracy for brain tumor classification.

Dahan et al.¹⁸

ResNet50 CNN and Marine Predator Algorithm (MPA)

3064 T1W-CE MRI (Figshare) images

98.72

Developed a three-step model using ResNet50 for feature extraction and MPA for feature fusion, achieving high accuracy in detecting brain tumors in challenging images.

Islam et al.¹⁹

EfficientNetB0, EfficientNetB3

3064 T1W-CE MRI (Figshare) images

99.69

Introduced a deep learning approach using EfficientNet for enhanced classification, with EfficientNetB3 achieving the highest accuracy, outperforming many state-of-the-art methods.

Tummala et al.²⁰

ViT models (L/16, B/16, L/32, B/32)

Figshare (3-class MRI)

98.70

Showcased the effectiveness of ensemble ViT models in three-class brain tumor classification.

Abd-Ellah et al.²¹

BTC-fCNN (custom CNN architecture)

Figshare (3064 MRI)

98.86

Developed a custom 13-layer CNN architecture, BTC-fCNN, outperforming existing CNN strategies for brain tumor detection.

Maruf et al.²²

EfficientNetB3, DenseNet121, EfficientNetB2/B4/B5

Figshare (3064 MRI)

98.98

EfficientNetB3 outperformed 26 other CNN models, showing excellent performance in brain tumor classification.

Asif et al.²³

DenseNet121, ResNet152V2, Xception, DenseNet201, InceptionResNetV2

Figshare MRI Dataset

99.67

Designed a DL architecture for both three-class and four-class classification, outperforming state-of-the-art methods.

Nassar et al.²⁴

GoogleNet, ShuffleNet, SqueezeNet, AlexNet, NASNet-Mobile

Figshare (3064 MRI)

99.31

Introduced a majority voting technique with multiple models, achieving favorable results in brain tumor classification.

Saeedi et al.²⁵

2D CNN, Convolutional Autoencoder

Figshare MRI Dataset

96.47

Proposed 2D CNN and autoencoder networks with multiple convolution and pooling layers for brain tumor classification.

Ait Younes et al.²⁶

CNN with Bayesian Optimization

Figshare MRI Dataset

98.70

Applied Bayesian optimization for hyperparameter tuning, achieving superior performance in brain tumor classification.

Ayadi et al.²⁷

CNN (custom architecture)

Figshare, Radiopaedia, Rembrandt

97.22

Evaluated on multiple datasets, the CNN model achieved strong classification accuracy with minimal preprocessing.