Deep Belief Networks for Multi-Class Brain Tumor Classification with Improved Diagnostic Accuracy

The proposed research work investigates the use of Deep Belief Networks (DBNs) for the multi-class classification of brain tumors to improve diagnostic accuracy in medical imaging. Brain tumors present significant difficulties in identification and classification due to their varied morphologies and overlapping characteristics. DBNs, characterized by their multi-layered structure of restricted Boltzmann machines, are used to automatically extract hierarchical characteristics from magnetic resonance images of brain. The proposed technique consists of a two-phase training process: first, unsupervised network pre-training to extract pertinent features, followed by supervised fine-tuning to enhance classification performance. The DBN model's efficacy is compared to traditional machine learning techniques using an extensive dataset of brain tumor images. The results demonstrate that the DBN technique improves current approaches for accuracy, sensitivity, and specificity across several tumor types, including gliomas, meningiomas, and pituitary tumors. The proposed DBN achieves 97.9% accuracy, outperforming existing machine learning algorithms with a 7–18% enhancement in brain tumour classification, demonstrating greater diagnostic accuracy. The results highlight the efficacy of DBNs as a powerful instrument for automated brain tumor classification, offering significant assistance to radiologists and enhancing diagnostic processes. It supports the increasing evidence for using deep learning methods in clinical practices to improve patient care in oncology.