Quantum-inspired optimization of transformer-capsule networks for accurate brain tumor segmentation and...

Area of Science:

• Medical imaging analysis
• Artificial intelligence in oncology
• Computational neuroscience

Background:

• Early and accurate brain tumor diagnosis is critical for patient survival.
• Current deep learning methods for brain tumor analysis suffer from limitations in spatial dependency modeling, feature redundancy, and structural relationship representation, reducing diagnostic reliability.

Purpose of the Study:

• To propose a novel Quantum-Inspired Optimization of Transformer-Capsule Networks (QI-TCN) framework to address the limitations in current deep learning-based brain tumor diagnosis.
• To enhance the accuracy, reliability, and interpretability of automated brain tumor diagnosis.

Main Methods:

• Integration of Swin Transformer U-Net (Swin-UNet) for hierarchical segmentation.
• Application of Quantum-Inspired Harris Hawks Optimization (QHHO) for hyperparameter tuning and feature selection.
• Utilization of EfficientNetV2 for feature extraction and Graph Attention Capsule Network (GACN) for classification.

Main Results:

• Achieved high classification accuracies (99.02% on BraTS 2019, 99.15% on BraTS 2020) and F1-scores (98.97%, 99.10%).
• Demonstrated superior performance over baseline methods with statistically significant improvements (p < 0.001).
• Achieved 96.8% zero-shot and 98.4% fine-tuned accuracy on the Figshare Brain Tumor Dataset.

Conclusions:

• The proposed QI-TCN framework offers a robust, interpretable, and clinically viable solution for automated brain tumor diagnosis.
• The novel integration of quantum-inspired optimization and advanced deep learning architectures significantly enhances diagnostic performance.
• This framework represents a significant advancement in leveraging AI for neurological disease diagnostics.