Accurate Brain Tumor Detection Using Laplacian Based GAN Model

The very complexity of the brain and the processes it operates with makes its tumors one of the most complicated malignancies that are difficult to detect and heal. Both benign and malignant kinds of tumors could seriously impair ordinary brain function. Improved results entail early, accurate diagnoses but traditional methods of biopsies and X-rays have disadvantages and risks. Improvements in deep learning (DL) and machine learning (ML) have helped diagnose brain tumors using non-invasive methods like magnetic resonance imaging (MRI). This paper applies Graph Neural Networks (GNN) and Laplacian-based image processing to propose an automated brain tumor recognition system depend on the MRI data from the BraTS 2020 dataset to classify brain lesions as benign or malignant. Before segmenting the images with a U-Net-based model to delineate brain lesions, the technique pre-processed MRI images using Laplacian filtering for better edge identification and noise reduction. Data imbalance was handled with a hybrid Laplacian-based U-Net GAN model, and a Fast Graph Convolutional Network (GCN) was applied to identify malignant cells. To improve generalization and avoid overfitting, dropout layers and data augmentation were applied. The model succeeded in identifying the tumor with a 90% accuracy on 1000 datasets, but when it was increased to 1500 datasets, the overfitting caused the result to decrease to 75% and stabilize at 70%. Although the issue is still being faced with bigger datasets, the findings show promise for early timely detection of tumors.