Image Segmentation Approaches to Detect Abnormalities in Brain MRI Images using CNN & U-Net

Abstract

It is a challenging and crucial task in medical research to recognize and define brain cancers via Magnetic Resonance Imaging (MRI). Inspite new model, this paper comes with a solution for drawbacks in the (CNN+DWA (Distance Wise Attention)) model with the hybrid model, it has two models which are U-NET and (CNN+DWA). Even though CNN is the best model for brain tumor identification, it has one exception case, when the brain tumor is more than 1/3rd of the brain then it gives inaccurate values. In normal cases as usually, CNN models are used for analysis if an exception case has occurred then only in that condition this U-NET model comes into the picture, otherwise, this model is just beside without disturbing analysis of CNN. The CNN Model suggests using a pre-processing method that only affects a tiny portion of the MRI image as opposed to looking at the entire picture. It, therefore, resolves the fitting problems in the Cascading Deep Learning model and speeds up computation. In the second stage, a straightforward and effective convolutional neural network (C-Conv Net/CNN) is suggested to deal with a smaller portion of each slice's brain MRI images. This CNN model uses two different approaches to mine both local and global characteristics. Additionally, the DWA mechanism has been employed to enhance the accuracy of brain tumor segmentation as compared to contemporary models. The DWA approach takes into account the effects of a brain tumor being present in a critical region of the brain. U-NET Model, which is already exited but in addition to that included error value. This exception case is calculated by a model based on accuracy and computational time. It maintains accuracy and efficiency by adding error value in exceptional cases only.