A team of researchers has announced a major advancement in the automated detection of brain tumours, unveiling a hybrid artificial intelligence system that dramatically enhances the accuracy of MRI-based diagnosis. The innovative framework integrates deep learning, metric learning, and swarm-intelligence techniques, offering a powerful new tool for medical imaging.
Integrating Deep Learning and Optimisation
The study evaluated five leading pre-trained convolutional neural networks (CNNs), DenseNet201, MobileNetV2, ResNet50, ResNet101, and InceptionV3, on a diverse dataset of 7,023 MRI images. These images were categorised into four classes: glioma, meningioma, pituitary tumour, and healthy brain tissue.
DenseNet201 emerged as the most accurate baseline model, achieving 92.66% accuracy. The research team then applied a Large Margin Nearest Neighbour (LMNN) algorithm to increase feature separability, followed by two swarm-intelligence optimisers, Particle Swarm Optimisation (PSO) and Grey Wolf Optimiser (GWO), to refine feature selection. The optimised features were then input into four classifiers: k-Nearest Neighbour (KNN), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Random Forest (RF).
DenseWolf-K Sets a New Benchmark
Among all tested configurations, the combination of DenseNet201, LMNN, GWO, and KNN, dubbed “DenseWolf-K”, achieved an unprecedented 99.64% accuracy rate. This outstanding performance marks DenseWolf-K as the optimal configuration for precise and reliable brain tumour classification.
The model’s robustness and generalisability were confirmed using an independent external dataset, highlighting strong performance across varied imaging conditions. For transparency and clinical trust, explainability tools were incorporated. Feature-level ranking and occlusion sensitivity mapping, two explainable artificial intelligence (XAI) techniques, were used to reveal the most influential features for each prediction.
With its high accuracy, low false negative rates, and enhanced interpretability, the DenseWolf-K framework represents a major step forward in medical diagnostics. Experts suggest this hybrid approach could accelerate early tumour detection, support radiologists’ clinical decisions, and set a new standard for AI-assisted medical imaging.
Reference
Yonar A. A swarm intelligence-driven hybrid framework for brain tumor classification with enhanced deep features. Scientific Reports. 2025;15:37543.
