Brain Tumor Segmentation and Classification Using Binomial Thresholding-Based Bidirectional-Long-Short Term Memory

Abstract

A brain tumor arises when abnormal cells develop in the brain, leading to an elevated risk of illness and mortality due to the accelerated growth of these tumor cells. Magnetic resonance imaging (MRI) is utilized for detecting and diagnosing brain tumors by providing images of the brain's internal structure. However, brain tumors lead to the death of numerous lives due to inaccurate segmentation and classification of brain tumors. In this research, the binomial thresholding-based bidirectional-long-short term memory (BT-Bi-LSTM) is proposed for accurate segmentation and classification. Initially, the image is acquired from BRATS 2019 and BRATS 2020 datasets and then normalization and contrast limited adaptive histogram equalization (CLAHE) approaches are established in preprocessing. The BT technique is employed for segmenting tumor portions from the pre-processed images. The graylevel Co-occurrence matrix (GLCM) and local ternary pattern (LTP) are employed to extract the features. Finally, the Bi-LSTM is used to classify the types of brain tumors. The BT-Bi-LSTM achieves better accuracy, precision, recall, and f1-score of 99.76%, 99.52%, 99.31%, and 98.69% for BRATS 2019 dataset compared to the existing approaches like DNN-based mathematical approach, tumor localization enhancement approach and U-net architecture, and Hybrid Convolution Neural Network (HCNN). When compared to tumor localization enhancement approach and U-net architecture, CNN, hybrid Deep CNN with k-means clustering, and 2D U-net, the BT-Bi-LSTM achieves better accuracy, precision, and recall of 99.89%, 99.76%, and 99.62% for BRATS 2020 dataset respectively.