Enhanced brain tumor classification framework using deep learning

The increasing prevalence of brain tumors calls for the development of accurate and reliable diagnostic tools. Whereas traditional techniques offer some benefits, they can hardly detect or accurately classify the type of a tumor at an early stage, crucial for proper treatment planning. For this purpose, deep learning methods have been taken into consideration but are often prohibitive in nature due to architectures not well-suited for handling complex, heterogeneous datasets and requirements for large numbers of labeled data samples. This paper will introduce an advanced deep learning framework for increase the classification accuracy of different classes of brain tumors, such as glioma, meningioma, no tumor, and pituitary. To achieve this, it uses an enriched comprehensive dataset, with data augmentation done through Generative Adversarial Networks, to boost model performance and address the problem of limited labeled data. In this paper we used ResNet18 model to extract features, because it has been proven to be very effective for medical image and sample feature extraction of a complex nature, hence improving computational efficiency and performance. DMFN model was then formulated using a number of the ResNet 18 models used for the purpose where 3 models of Res Net 18 had to act on the pair - wise .The model shows excellent improvement on the BRATS2021 Dataset over existing techniques to reduce the training loss to 0.1963, validate loss to 0.1382, and validation accuracy to 98.36%. These results underscore the potential of our approach to move forward the diagnostics of the brain tumors. The combination of GAN for data augmentation, combined with the innovative use of PCA-PSO for FS and DMFN for classification, provides an overall robust framework that allows further applications with other medical imaging tasks in order to bring about improved clinical outcomes and further the medical image analysis field.