ENHANCED BRAIN TUMOR SEGMENTATION AND CLASSIFICATION USING CAPSULE NETWORK AND EFFICIENTNETB3

Abstract

Brain tumors pose a significant threat to both young and older individuals. Despite advancements in machine learning (ML) and deep learning (DL) techniques for brain tumor segmentation and detection, challenges persist, including low performance, high computational complexity, and insufficient data. This study proposes an efficient multi-modal brain tumor segmentation and classification method using a combination of Capsule Network and EfficientNetB3. Initially, T1, T2, and Flair MRI images are pre-processed using Upgraded Mean Filtering (Up-MFil), image resizing, and HSV color channel conversion. Segmentation is then performed using Gannet-based Kapur's Thresholding (GKT), followed by feature extraction and optimal feature selection using the Gannet Optimal EfficientNetB3 model. Finally, the brain tumors are classified into High Grade Glioma (HGG) and Low Grade Glioma (LGG) using the Fine-Tuned Hybrid Deep Convolutional Capsule Network (FT-HDC²Net). The proposed technique is evaluated on the BraTS 2018 dataset, achieving 98% accuracy, 96.52% precision, and 94.03% recall. Compared to other related techniques, the proposed method demonstrates superior performance in terms of accuracy, precision, recall, F1-score, specificity, and Kappa. Additionally, the method achieves low error rates and fast computational times, highlighting its potential for efficient brain tumor segmentation and detection.