Deep Learning Framework for Multi-Grade Brain Tumor Detection Using Latent Graph Encoder Coupled Triple Generative Adversarial Networks with Walk-Spread Algorithm

A brain tumor is an aberrant cell development that may be benign or malignant. Tor form, size, location, and overlapping elements of visual characteristics associated with multi-grade BT identification therefore vary. In order to tackle these problems, a model for Multi-Grade Brain Tumor Detection (MGBTD) called Latent Graph Encoder Coupled Triple Generative Adversarial Networks with Walk-Spread Algorithm (LGECTGA2Nets+WSA) is proposed. To improve image quality, MRI images from the BRATS 2018 & Figshare datasets are first pre-processed using the Hybrid Recursive Reversible Box Filter-Based Fast Adaptive Bilateral Filtering (HRRBF-FABF) approach. Then, robust spatial-frequency feature extraction is done using the Discrete Quaternion Quadratic Phase Fourier Transform (DQQPFT). Tumor segmentation is performed with a Convolution-Transformer (CT), followed by classification using LGECTGA2Nets. The Walk-Spread Algorithm has been used to obtain model optimization. Python-based system has an accuracy and sensitivity value of 99.9% and 99.8% respectively, which is better when compared to other systems used in detecting and classifying multiple grades of tumors.