Fig 1.
Scheme of the proposed model.
Table 1.
Recent meta-heuristic optimization in deep learning approaches applied to brain tumor analysis.
Fig 2.
Visualization of the BraTS2023 dataset collection, the range of MRI modalities and ground truth segmentation.
Fig 3.
Augmented MRI modalities and corresponding segmentation labels.
Fig 4.
Final preprocessed MRI modalities and corresponding labels.
Fig 5.
Proposed architecture.
Fig 6.
Sample image of proposed method results.
Table 2.
Corresponding sample image metric values.
Table 3.
Quantitative metrics comparison for Sample 100 (validation case).
Fig 7.
Sample image of original and corresponding image for ground truth.
Fig 8.
Comparison of different algorithms (Sample 100).
Fig 9.
Comparison of HD95 for different algorithms on Sample 100.
Table 4.
Quantitative metrics comparison for Sample 77 (validation case).
Fig 10.
Comparison of different algorithms (Sample 77).
Fig 11.
Comparison of HD95 for different algorithms on Sample 77.
Fig 12.
Example of no tumor prediction segmentation.
Fig 13.
Training dynamics of the proposed GA-ResUNetGAN model.
(a) Training loss curve with convergence; (b) Training accuracy showing rapid improvement; (c)-(e) Validation DSC curves for TC, WT, and ET indicating high segmentation performance.
Table 5.
Comparison of key metrics with and without augmentation (identical training budget in both conditions).
Fig 14.
Error of segmentation because of over fitting results of sample image.
Table 6.
Accurate hyperparameter settings for GA-ResUNetGAN model.
Table 7.
5-fold cross-validation: Best validation Dice score per fold (mean ± std across folds).
Table 8.
10-Table 9Table 10fold cross-validation: Best validation Dice score per fold (mean ± std across folds).
Table 9.
Ablation of architectural components and their effect on segmentation performance.
Table 10.
Wilcoxon signed-rank test comparing GA-ResUNetGAN with baseline models on BraTS2023 validation (n = 22).
Table 11.
Model complexity: trainable parameters, FLOPs per 96 × 96 × 96 patch, and inference time per full MRI volume.
Fig 15.
Comparison of CPU time, system time, and total execution time (in seconds) for different segmentation models over two training epochs.
Average runtime per epoch is annotated above each bar.
Table 12.
Comparison of Brain Tumor Segmentation Models on BraTS Datasets.