A Modified U-Net Based Architecture for Brain Tumour Segmentation on BRATS 2020

Abstract

The segmentation of brain tumours plays a significant role in the analysis of medical imaging. For a precise diagnosis of the condition, radiologists employ medical imaging. In order to recognise brain tumours from medical imaging, the radiologist's work must be challenging and complex. There are various distinct steps that may be used to identify brain tumours using magnetic resonance imaging (MRI). In the field of medical imaging, segmentation is the key stage. Segmentation is carried out after classification and image analysis. The appropriate segmentation is crucial since a brain tumour's incorrect detection might have a number of negative effects Method: In this work, the multimodal Brain tumour segmentation challenge was employed (MICCAI BRATS). We removed the brain tumour from the MRI images using the BRATS 2020 dataset, which is openly accessible. In this collection, there are 371 NiFTI-format folders. Convolutional neural networks (CNNs), a kind of deep learning based on an encoder-decoder model, are used in the proposed method to separate the tumours. Results: Accuracy = 0.9759, loss = 0.8240, and IOU = 0.6413 indicate that the proposed model is successful. The proposed model performs better when compared to the state-of-art segmentation models used in this study.

CrossTransUnet: A New Computationally Inexpensive Tumor Segmentation Model for Brain MRI

“…In the specific case of brain tumors, most of the challenges and developments have been based on the BraTs database that includes axial resonance images with their respective segmentation masks, performed by expert radiologists. For example, Kajal and Mittal [37] use this database for the development of a 3D network inspired by the UNet network. The authors compare the performance of their model with different state-of-the-art networks, concluding that their model outperforms other models in terms of accuracy and IoU (accuracy of 98.19% and IoU of 65.88%).…”