Brain Tumors Classification on Magnetic Resonance Images by Hybrid Model

Objective: To provide fast and reliable tumor diagnosis system for radiologists. Methods: An ensemble model–Hybrid Ant Bee Colony (HABC) is proposed to classify adult brain tumor classes Meningioma (MEN, Class-1), Glioma (GLI, Clss-2) and Pituitary tumor (PIT, Class-3) obtained from internet repository with 3064 T1 post-contrast MR brain tumor images. Strategy for searching features is analogous to foraging habits of ants and bees. The algorithm generates neighboring food sources (feature subsets) and evaluates them, while utilizing the ant pheromone trail to collate the best food sources from all feature subsets. The brain tumors on magnetic resonance images are identified by their relative intensity (iso, hypo, and hyper) and textural features-367 in number. Results: HABC utilizes 216 features and achieves 96% classification accuracy. Conclusion: The statistical parameters used to measure performance in the present research, clearly indicate an advantage of implementing HABC for feature selection and classification of 3064 tumor slices into 3 classes. Significance: The proposed technique performs better than both Ant Colony Optimization (ACO) and Artificial Bee colony (ABC) as it modifies all feature subsets by sorting individual features in order of least error in classification, hence it reduces misclassification rate thereby increasing overall classification accuracy, sensitivity and specificity.