Brain Image segmentation is the predominate approach used to conquer an isolated anomaly portion of MRI image through proper thresholding strategy adapted in the deep learning model. Hence, it can converge to desire fitness value and thereby, finds optimistic solution to best threshold effectively. However, it may face high computational complexity because of meta-heuristic search of multi-level thresholding in order to console accuracy as well as efficiency. In this paper, an effective segmentation of abnormalities on magnetic resonance imaging (MRI) images using Double Optimized Convolution Neural Network (DOCNN) is proposed. In this method, first, Ant Bee Colony (ABC) algorithm along with Sugeno Fuzzy (SF) logic set is used to extract the salient features by knowing depth of penetration into tissues textures. Thereby, a cell infectious radius and spreading ratio is easily calculated. Second, a best optimistic threshold value is identified without creating any additional complexity by using back widow spider algorithm. When compared to other existing approaches such as Equilibrium optimization (EO), Particle swarm optimization (PSO), Slap swarm optimization (SSO), Genetic algorithm (GA), Firefly algorithm (FA), and Gray wolf optimization (GWO), With Other Deep Learning Models (CNN, RNN, LSTM) the proposed algorithm takes less computational complexity and high accuracy in terms of infectious portion detection. It has proved by estimated parameters such as specificity, precision, and sensitivity, Jaccard index and Dice coefficient index.