A hybrid optimized attention-based deep learning approach for enhancing brain tumor classification using magnetic resonance imaging

Brain tumor is a life-threatening disease triggered by the uncontrolled development and spreading of tumorous cells in the brain tissues. Timely diagnosis and prompt interventions are crucial to improve patient outcomes. Recently, Deep Learning (DL) methods have gained popularity in diagnosing brain tumors using Magnetic Resonance Imaging (MRI) samples. Nevertheless, conventional approaches faced issues, like poor tumor segmentation, feature representation, learning flexibility, and parameter tuning, which impact classification accuracy. To overcome this, the research proposes an Optimized Dual Learning-enabled Hybrid Attention-based Convolutional Memory Network (ODL-HA-CMNet) for the classification of brain tumors using MRI. The inclusion of the Optimized Hybrid Attention (OHA) based VNet segmentation facilitates deliberative analysis on tumorous regions, while attending to tumor-specific patterns. Here, the usage of the hybrid attention mechanism stimulates the focus on discriminative tumor characteristics while avoiding irrelevant patterns. Notably, the training strategy based on dual learning enhances system scalability and improves the model’s ability to adapt to newer data through knowledge retention. Further, adaptive tuning of hyperparameters using the Leader Guidance Foraging Optimization (LGFO) guarantees steady convergence, thereby enhancing the segmentation and classification accuracy. ODL-HA-CMNet is trained and tested using two benchmarks, where the highest level of performance, with 98.81% of accuracy, 99.11% of specificity, 98.93% of F1-Score, 99.30% of precision, and 98.56% of sensitivity, is attained on the BraTS2020 dataset, under the training percentage of 90%. Similarly, with the BraTS2017 dataset, ODL-HA-CMNet attained a remarkable 98.72% of accuracy, 99.03% of specificity, 98.97% of F1-Score, 99.42% of precision, and 98.53% of sensitivity with the training percentage of 90%.