A swarm intelligence-driven hybrid framework for brain tumor classification with enhanced deep features.

Area of Science:

• Medical Imaging and Artificial Intelligence
• Computational Neuroscience
• Machine Learning in Healthcare

Background:

• Accurate brain tumor classification from MRI is crucial for timely diagnosis and treatment.
• Existing methods face challenges in achieving high accuracy and interpretability.
• Automated classification systems are needed to assist radiologists and improve patient outcomes.

Purpose of the Study:

• To develop and validate a hybrid AI framework for robust four-class brain tumor classification using MRI.
• To combine deep learning features with metric learning and swarm intelligence for enhanced classification performance.
• To ensure model interpretability through explainable AI techniques.

Main Methods:

• A hybrid framework integrating Convolutional Neural Network (CNN) deep features (DenseNet201, MobileNetV2, ResNet50, ResNet101, InceptionV3), Large Margin Nearest Neighbor (LMNN) metric learning, and swarm intelligence optimization (PSO, GWO).
• Feature selection and optimization using Particle Swarm Optimization (PSO) and Grey Wolf Optimizer (GWO).
• Classification using k-Nearest Neighbor (KNN), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Random Forest (RF), with the DenseNet201-LMNN-GWO-KNN (DenseWolf-K) configuration showing optimal results.

Main Results:

• DenseNet201 achieved the highest baseline accuracy (92.66%).
• The DenseWolf-K framework (DenseNet201-LMNN-GWO-KNN) reached a superior accuracy of 99.64% on a dataset of 7,023 MRI images (glioma, meningioma, pituitary, healthy).
• The model demonstrated robustness and generalizability on an independent external dataset, supported by feature-level ranking and occlusion sensitivity maps for explainability.

Conclusions:

• The proposed DenseWolf-K framework provides a highly accurate, interpretable, and efficient solution for MRI-based brain tumor classification.
• The hybrid approach effectively combines deep learning, metric learning, and optimization techniques for improved diagnostic performance.
• This framework holds significant potential for aiding in early brain tumor detection and personalized treatment strategies.