To develop a parameter-efficient CNN architecture for MRI-based brain tumor classification that effectively balances accuracy, computational cost, and deployment efficiency.
Key Findings:
PruDensNet achieved a validation accuracy of 97.27% and test accuracy of 96.05% on a four-class brain tumor MRI benchmark, with detailed per-class metrics demonstrating superior performance.
Outperformed matched-capacity CNN and Transformer baselines in terms of accuracy and per-class metrics.
Demonstrated a favorable accuracy-footprint trade-off suitable for cost- and latency-sensitive clinical workflows.
Interpretation:
The results indicate that PruDensNet is a viable option for clinical settings where computational efficiency is crucial, without compromising diagnostic accuracy.
Limitations:
Results require external validation to confirm generalizability across diverse clinical environments, emphasizing the need for varied clinical settings.
Performance may vary based on hardware-specific implementations and configurations.
Conclusion:
PruDensNet represents a significant advancement in efficient brain tumor classification using MRI, balancing accuracy and resource constraints for practical clinical applications, particularly in resource-limited settings.
