Feature-enhanced deep learning technique with soft attention for MRI-based brain tumor classification

Brain tumor classification using Magnetic Resonance Imaging (MRI) is a pivotal area in medical diagnostics, with the potential to influence early detection and subsequent treatment strategies. Over the years, various machine learning and deep learning models have been proposed to enhance the accuracy of MRI-based tumor detection. In this evolving landscape, this paper introduces a distinctive deep-learning model that harnesses the power of a soft attention mechanism. We employ a meticulously designed Convolutional Neural Network (CNN) comprising four convolution layers. One of the significant innovations in our approach is the method of feature extraction. Instead of extracting features solely from the final layer, as is common in many models, our approach aggregates and combines features from all layers. This ensures that the vital characteristics intrinsic to each layer are not lost but rather amalgamated into a robust and comprehensive feature vector. The incorporation of a soft attention mechanism at the terminal stages ensures that the most salient and clinically relevant features are emphasized, enhancing classification accuracy. To validate the efficacy of our proposed model, we employed standard datasets for training and testing. A comparative analysis with existing state-of-the-art models affirms the superiority and potential of our approach in the realm of MRI-based brain tumor classification.