Modified U-Net with Attention gates and FTL for Lesion Segmentation

Brain injury is described as damage, destruction anywhere on the brain. It can be because of injury. Or it can be because of other disease which causes inflammations, brain not functioning, or may be because of destruction in the brain, which can be tissues. The wound can be placed on one side of the brain or spread. Lesion Segmentation is function of separating lesions form normal part of the brain, it is to detect the lesions which are present on the body. Lesions can be described as injuries, like brain injuries, etc. A new architecture is proposed, which is Modified U-Net with Attention Gates and Focal Loss (FTL), it is a combination of two Models which are combined to form on model. These models are based on UNet architecture, which is U shaped, hence the name UNet. The first model is UNet with encoder as VGG-19. To get more data information, Attention UNet is added at the end and for dealing with the class imbalance data FLT is used, to improve the performance and accuracy of previous models like U-Net, Attention U-Net for better performance in dealing with Lesion Segmentation.