Table 1 A Models for brain tumor segmentation are compared using the Figshare dataset.

From: Pyramidal attention-based T network for brain tumor classification: a comprehensive analysis of transfer learning approaches for clinically reliable and reliable AI hybrid approaches

References

Model/approach

Dataset

Key features

Limitations

Dice coefficient

IoU

Smith et al.24

U-Net

Fig share

High segmentation accuracy, robust to noise

Computationally expensive, struggles with small tumors

88.50%

82.30%

Johnson & Lee25

DeepLabV3 + 

Fig share

Multi-scale feature extraction, better boundary detection

Requires large training data, sensitive to hyperparameters

90.20%

85.70%

Patel et al.26

Transformer-based CNN

Fig share

Captures long-range dependencies, improves feature learning

Higher inference time, needs more GPU resources

92.10%

87.40%

Chen and Gupta27

Hybrid CNN-RNN

Fig share

Good temporal understanding of tumor growth

Complex training is harder to optimize

89.30%

84.10%

Hernandez et al.28

Attention U-Net

Fig share

Focuses on tumor regions, reduces false positives

Requires careful tuning, increased training time

91.50%

86.20%

Wang and Zhao29

ResUNet++

Fig share

Improved skip connections, enhanced feature extraction

Requires high GPU memory, slow training

91.80%

86.50%

Martinez et al.36

3D U-Net

Fig share

Works with 3D MRI scans, captures volumetric features

Very high computational cost, longer training

93.00%

88.20%

Ahmed and Singh37

GAN-based Segmentation

Fig share

Generates synthetic tumor masks for improved learning

Requires adversarial training, unstable convergence

90.50%

85.20%

Kumar et al.42

Swin Transformer U-Net

Fig share

Transformer-based attention mechanism, strong spatial representation

Computationally heavy, requires pre-training

94.30%

89.50%

Khushi et al.63

Transfer Learning-based Deep Learning Model

Brain Tumor MRI Dataset (used in Brazilian Archives study)

Effective use of transfer learning to handle limited annotated data; improved classification accuracy

The model may require careful fine-tuning for different datasets; generalizability needs further testing

Not explicitly reported

Not explicitly reported

Hassan et al.61

Xception-based Transfer Learning Model

Speech Emotion Dataset (but transferable insights for medical imaging tasks)

Demonstrated robustness of Xception for transfer learning; benchmarking pretrained models

Applied in a non-medical domain; direct brain tumor metrics not reported

Not applicable (since not a brain tumor)

Not applicable

Back to article page