Brain Tumor Segmentation and Classification Using Deep Learning with Hybrid KFCM Clustering and LuNet Classifier

Abstract

Brain tumor identification and classification play a critical role in early diagnosis and treatment planning. This research presents a hybrid deep learning-based architecture for accurately classifying and segmenting brain tumors from MRI images. The T1-W CEMRI dataset is used as the input data set for the suggested approach. The data set is first preprocessed by turning it into greyscale images with a fixed pixel resolution. Techniques for noise reduction are used to improve image quality and maintain structural details that are essential for diagnosis. Using a Hybrid Kernelized Fuzzy C-Means (KFCM) clustering technique, which efficiently defines tumor borders by utilizing both spatial and intensity information in a kernel-induced space, tumor regions are segmented. This makes segmentation more robust, particularly in areas with poor contrast or noise. A deep convolutional neural network called VGG-16 is used for feature extraction in order to extract high-level spatial characteristics from the segmented brain tumor images. After extraction, a lightweight convolutional neural network architecture called LuNet which has been trained to differentiate between meningioma, glioma, and pituitary tumors is used to classify the retrieved tumor sections. According to experimental data, the suggested model achieves an astonishing 98% classification accuracy, showing its potential for accurate tumor diagnosis. By fusing the advantages of deep learning and conventional clustering, this hybrid framework provides a very effective and precise method for analyzing brain tumors in clinical settings.