Advancements in Brain Tumor Detection Using Hybrid Deep Learning Models

Tumors are abnormal masses of tissue that form when cells grow and divide, or do not die when they should. These tumors can grow in any part of the body, and when such growth occurs within the brain, it is referred to as a brain tumor. Brain tumors consist of atypical cell growths within the brain or the spinal canal in the middle. These growths may be malignant (cancerous) or benign (noncancerous) and differ significantly in their growth patterns and potential to inflict damage. These tumors present significant health challenges, impairing essential brain functions, and requiring early detection for effective treatment. This research proposes a robust deep-learning framework designed for the accurate MRI identification and categorization of brain cancers. By using the BRATS 2020 dataset, this study develops a comprehensive pipeline that begins with the meticulous acquisition of MRI data. This is followed by advanced stages of data augmentation and preparing the input data for further machine learning by improving its quality and correctness. The core methodology employs a U-Net model for to accurately segment brain tumors, ensuring detailed delineation of tumor regions, the extracted features are analyzed using the architectures of DenseNet and EfficientNet, adept at handling complex image data. The proposed framework achieves a Dice Coefficient of 0.97, the following performance metrics: for Lower Grade Glioma, an accuracy of 98.96%, and for High-Grade Glioma, accuracy stands at 98.94%. These metrics affirm the robustness in distinguishing between tumor grades with high reliability. This notable degree of precision highlights the possibility of this advanced deep-learning approach to enhance diagnostic processes and improve patient outcomes by facilitating early and accurate tumor detection.