Multi-class brain tumor classification using deep learning ‎and tumor segmentation using image processing techniques

Abstract

With the rising occurrence of brain cancers worldwide, accurate and timely tumor diagnosis is crucial. In this study, we provide a robust and ‎efficient brain tumor classification and segmentation system that uses deep learning and digital image processing for tumor segmentation. ‎We use convolutional neural networks to extract specific information from images. This will eventually eliminate the requirement for manual ‎tumor extraction and detection. We employ a diverse collection of JPEG pictures of three types of tumors: glioma, pituitary, and meningioma-‎mas. Our CNN architecture is meant to record diverse patterns of the image, which assists our model in distinguishing between types of ‎tumors and allows it to classify distinct types of tumors. In addition, we applied digital image processing for tumor segmentation, employing techniques such as contrast, erosion, dilatation, and many more. To improve the quality of our model, we applied several data augmentation approaches and dropouts in our CNN architecture to identify tumors more precisely and robustly. According to our findings, using ‎deep learning technology for cancer categorization can improve the accuracy and speed with which brain tumors are detected. This can help ‎to reduce the manual labor necessary to locate tumors.