Leveraging Deep Learning for Efficient Brain Tumor Classification

The buildup of abnormal cells in certain brain tissues is known as a brain tumor. It is among the most dangerous and lethal cancers that can occur at any age. Identification of brain tumors involves determining the part of the brain that is impacted by the tumor by looking at its size, shape, border, and location. This project aims to develop an innovative network for the classification of brain tumors. The primary focus is on enhancing classification through precise Preprocessing, segmentation, Extraction of features and lastly classification approach. The workflow begins with an optimized image enhancement process to improve the input image quality. For segmentation, a customized deep learning architecture is proposed, steering away from the conventional U-Net model. This new segmentation technique is designed to capture intricate tumor boundaries and shapes effectively. Distinctive properties such as Modified LGDiP, loop characteristics, Size of tumor and the MEdian Binary Pattern (MBP) are retrieved after segmentation. These features collectively contribute to a rich representation of the tumor’s visual and spatial attributes. To categorize tumors, a hybrid classification model combining CNN and LSTM is used.