Design of FCSE-GAN for Dissection of Brain Tumour in MRI

Brain tumour is the collection or growth of abnormal cells in brain. The tumours can develop from nerve cells, brain cells, membranes, glands, glial cells, etc. For treating patients with brain tumours, it is important to segment the brain tumour area accurately. The Generative adversarial network (GAN) is a promising deep neural network technique which has two neural networks namely Generator and Discriminator which acts opposite to each other. This paper proposes FCSE-GAN (Feature Concatenation based Squeeze and Excitation-GAN) for segmenting the brain tumour area in MRI. The proposed FCSE-GAN uses ResNet as basic neural network architecture. It includes feature concatenation technique with generator for generating sharp MRI images and Squeeze and excitation block with discriminator for segmenting the brain tumour area. The experiments were conducted using Brain MRI image dataset from Kaggle on WGAN-GP, Info-GAN and FCSE-GAN architectures. The experimental results shows that FCSE-GAN yields better accuracy, precision, recall and F1-score when compared to WGAN-GP and Info-GAN.