Performance analysis of hybrid bio-inspired algorithms for classifying brain tumors in imbalanced magnetic resonance imaging datasets

Abstract

Magnetic resonance imaging (MRI) is a substantial imaging procedure for diagnosing brain tumors. However, brain tumor classification continues challenging due to the unequal distribution of classes within datasets, complicating precise diagnosis and classification. This research focuses on the class imbalance in medical image datasets by proposing a hybrid bio-inspired algorithm for brain tumor classification. A rider optimization and particle rider mutual information-based dendritic-squirrel search algorithm combined with an artificial immune classifier is developed and tested on imbalanced datasets generated from BRATS and SimBRATS. Experimental outcomes are encouraging, For the imbalanced BRATS dataset, the rider optimization- based classifier achieved an accuracy of 94.84%, sensitivity of 92.96%, and specificity of 94.95%. The particle rider mutual information-based classifier outperformed others with 96.25% accuracy, 94.33% sensitivity, and 94.85% specificity. For the imbalanced SimBRATS dataset, the rider optimization-based classifier achieved 94.95% accuracy, 92.05% sensitivity, and 94.04% specificity. The particle rider mutual information-based classifier excelled with 96.35% accuracy, 94.42% sensitivity, and 95.44% specificity. These findings suggest that the proposed algorithm effectively addresses class imbalance in medical image datasets, offering a robust solution for brain tumor classification. The particle rider mutual information-based classifier shows promise for enhancing diagnostic accuracy in clinical settings, demonstrating the efficacy of hybridized bio-inspired algorithms in managing imbalanced datasets.