A novel approach to brain tumor classification using a ResNet50v2-GRU hybrid model

Brain tumors present a significant medical challenge, necessitating early and accurate detection for effective treatment. While current diagnostic methods are effective, they often require considerable manual effort and are prone to human error, underscoring the need for innovative technology-driven solutions. Deep learning has emerged as a promising solution for automated tumor classification from Magnetic Resonance Imaging (MRI) scans. This study introduces a novel hybrid ResNet50v2-GRU model, combining a Convolutional Neural Network (CNN) for feature extraction with a Gated Recurrent Unit (GRU) for pattern discernment. The proposed model achieves a classification accuracy of 98.25% on the ‘Brain Tumor MRI Images’ dataset [Aditya Komaravolu, 2021] comprising 7023 images of Glioma, Meningioma, Pituitary and ‘No Tumor’, outperforming the replicated state-of-the-art benchmark models—ResNet50v2 (97.86%), Xception (95.65% ), and InceptionResNetV2 (97.10%) [Talukder et al., 2023]. These models were retrained on the same dataset to create a comparison against the proposed model. The proposed CNN-RNN hybrid model demonstrates competitive performance when compared to the other models, emphasizing its potential to elevate diagnostic accuracy and propel advancements in medical image analysis, particularly in the context of neuroimaging.