BRAIN TUMOR USING ARTIFICIAL INTELLIGENCE, Summaries of Artificial Intelligence

Download BRAIN TUMOR USING ARTIFICIAL INTELLIGENCE and more Summaries Artificial Intelligence in PDF only on Docsity!

ain Tumor Detection using

ep Learning (Custom CNN vs ResNet50)

SETTI APPALANAIDU

Marwadi University

Problem Statement

Brain tumors cause significant morbidity and mortality. Manual MRI analysis is time-consuming. Deep learning automates tumor detection.

Dataset Overview

Data Preprocessing

• Resize 224×
• Normalize intensities
• CLAHE contrast enhancement
• Augmentation: rotation, flip, elastic deformation.

ResNet50 Fine-Tuning

• Pretrained on ImageNet
• Unfreeze top 30 layers
• Dual-branch head (GAP + GMP)
• L2 Regularization & Dropout(0.5)
• Adam optimizer, LR=1e−4.

Training Progress

Confusion Matrix

ROC Curve

Misclassified Examples

Results Comparison

Custom CNN: Accuracy 81%, AUC 0. ResNet50: Accuracy 89%, AUC 0. ResNet50 demonstrates higher robustness.

Limitations & Future Work

• Small dataset; limited MRI modalities.
• Expand to T1/T2/FLAIR sequences.
• Use federated learning for privacy.
• Deploy lightweight ResNet models.

Conclusion

ResNet50 outperformed Custom CNN with higher AUC and interpretability. Grad-CAM improved transparency. AI-assisted MRI diagnosis is clinically feasible.