Brain Tumor Segmentation and Detection Based on Deep Convolutional Wild Horse Deformable Neural Network Model

Brain tumor is one of the leading causes of mortality, and it requires early diagnosis. The diverse characteristics of brain tumors make the segmentation and detection process complicated. This paper presents a novel approach, a deep convolutional wild horse deformable neural network model (DCWDNN), for accurate disease diagnosis. The deep learning approach, UNet transformer architecture with self-attention encoder and decoder (UNet-TSA),is involved in segmenting tumor regions by capturing long-range dependencies. Moreover, the self-attention module is added to extract the fine-grained feature information with minimal computational complexity. The accuracy of the proposed tumor detection approach, UNet-TSA is enhanced with a wild horse soft computing (WHSC) based hyper parameter optimization approach. It ensures robustness to resolve problems in complex and high dimensional space. The proposed approach is evaluated with the BRaTS 2021 dataset in which the accuracy, precision, and recall performance are improved to 0.998, 0.991, and 0.992 with the Mean square error (MSE) of 0.0020. The performance of the proposed approach is compared with state-of-the-art DL approaches to show the effectiveness of the proposed approach.