Brain Tumor Segmentation, Detection and Grading in MRI Images

Abstract

The most common malignant brain tumours are gliomas, and they come in a variety of grades, each of which has a significant impact on the patient's chance of survival. Magnetic resonance imaging (MRI) tumour grading and segmentation are normal and crucial for treatment preparation and diagnosis. A deep learning approach was developed to meet this clinical need, that associates tumour segmentation using U-net which is a convolutional neural network (CNN) and tumour grading using transfer learning using a Vgg19 and a completely associated classifier. T1-postcontrast, FLAIR and T1-precontrast MRI images of 110 patients with LGG were used to train and evaluate. DSC for segmentation model's and tumour detection accuracy are 0.875 and 0.937, correspondingly. At the MRI image level, the grading model classifies LGG with specificity, accuracy, sensitivity, and of 0.922, 0.907, and 0.893, correspondingly. In MRI images this study shows conventional tool for automated and simultaneous LGG tumour segmentation, detection, and grading in clinical settings.