Brain Tumor Classification using Deep learning Algorithm

A deep learning-based hybrid system for the classification of brain tumors using magnetic resonance imaging (MRI). The proposed method integrates three prominent convolutional neural network (CNN) architectures—VGG16, InceptionV3, and ResNet50—with a Deep Belief Network (DBN) to form a comprehensive ensemble approach. The proposed system begins with pre-processing of MRI images to enhance their quality and ensure uniform input formatting. This step includes normalization, noise reduction, and contrast enhancement, which are crucial for improving the performance of subsequent feature extraction stages. The features extracted by these three CNN models are concatenated to form a rich, high-dimensional feature vector that encapsulates diverse spatial and textural information. This fused feature vector is then fed into a Deep Belief Network (DBN), which serves as the final classifier. The DBN is composed of multiple layers of Restricted Boltzmann Machines (RBMs) that are pre-trained in a greedy layer-wise fashion and then fine-tuned using supervised learning. It further refines the high-level feature representations and outputs the probability distribution over four predefined tumor classes: Glioma, Meningioma, Pituitary, and No Tumor. By combining the strengths of multiple CNN architectures with the powerful classification capabilities of DBNs, the proposed hybrid approach enhances diagnostic accuracy. This ensemble strategy exploits the complementary nature of the models, improving robustness and reliability in tumor detection. Overall, the system lays the groundwork for developing automated, real-time brain tumor diagnostic tools that can aid radiologists and clinicians in making faster and more accurate decisions.