<![CDATA[Automated brain tumor classification from magnetic resonance imaging (MRI) has become an essential component in advancing computer-aided diagnosis. However, many deep learning approaches prioritize accuracy alone while overlooking two key requirements for real-world medical deployment: the reliability of predicted confidence scores and the computational efficiency required for clinical integration. This study proposes a multi-objective bio-inspired hyperparameter optimization framework to produce convolutional neural network (CNN) models that are accurate, well-calibrated, and computationally efficient. The model is optimized using a Multi-Objective Particle Swarm Optimization (MOPSO) algorithm that jointly minimizes validation error, Expected Calibration Error (ECE), and inference latency. Experiments were conducted on a four-class Brain Tumor MRI dataset, and the optimized configuration achieved a test accuracy of 95 percent, an ECE of 1.48 percent, and a sub-millisecond inference latency of 0.88 milliseconds per sample. Grad-CAM visualizations further confirm that the model’s decisions are guided by clinically relevant tumor regions. The results demonstrate that multi-objective hyperparameter optimization offers a robust pathway for developing trustworthy, efficient, and interpretable artificial intelligence systems for medical imaging applications.]]>
Copyrights © 2025
