Detection Of Brain Tumors in Magnetic Resonance Image Segmentation Using Res-UNet Convolutional Neural Network

Brain tumor disease is a serious disorder of brain function and has the potential to cause death if diagnosed too late. Brain tumors can be diagnosed using Machine Learning segmentation techniques. Machine Learning segmentation with its various architectures is widely used in the disease diagnosis process because it is considered faster and more accurate in diagnosing disease. The aim of this research is to produce brain tumor segmentation in MRI images using the Res-UNet method and determine the accuracy of brain tumor segmentation results in MRI images produced using the Res-UNet method. Res-UNet is a Convolutional Neural Network architecture which contains elements of ResNet, namely Residual Connection in the UNet architectural structure. This research uses a brain MRI image dataset entitled “The Cancer Genome Atlas Low Grade Glioma Collection (TCGA-LGG)” which consists of 3929 images and labels. Brain tumor segmentation research using Res-UNet obtained quite good results using the Tversky Index as an evaluation metric, the resulting accuracy was 92.58% Tversky score. This research concludes that the Res-UNet architecture is able to diagnose brain tumors on MRI images quite well but is not accurate enough in determining the size and shape of the brain tumor.