Brain tumor detection with bi-directional cascade Gaussian kernel feature-generative adversarial networks

Abstract

The detection of brain tumors is critical in neurology and oncology. Advanced medical imaging does not mitigate challenges such as tumor variability, the diversity of imaging data, and the demand for high computational efficiency. Methods available so far face issues in accuracy and processing speed. The solution presented in this research will address all the problems mentioned above using the Bi-directional Cascade Gaussian Kernel Feature-Generative Adversarial Networks approach. Pre-training convolutional neural networks are then used for padding, resizing, normalization, and augmentation in advance preprocessing. Afterward, segmentation of the image by the Asymmetric Compound Boundary Guidance Branch Transformer (ABGBT) promotes boundary refinement, thus reducing the uncertainty. Integration of bi-directional cascade Gaussian kernels and generative adversarial networks in BCK-GAN helps in effectively extracting features as well as the detection process. In addition, the ATTAO further optimizes the network by applying an adaptive sigmoid attenuation function to optimize hyperparameters, thereby improving overall performance. Extensive experiments were performed using Python, which yielded impressive Dice scores of 96.04% on BraTS2018, 95.53% on BraTS2019, 96.13% on BraTS2020, and 95.79% on BraTS2021 Task 1. The detection speeds are 2.8 secs, 4.63 secs, 3.62 secs, and 3.2 secs, respectively, which significantly enhances brain tumor detection accuracy and efficiency.