Brain tumor segmentation in multi-modal MRI: A comparative study

A brain tumor is considered as an abnormal growth or lesion inside of the brain or near it. These tumors can be cancerous; malignant, known as Glioma due to abnormal growth of glial cells, or it can be noncancerous or benign as well. As such, the early detection and treatment of brain tumors is a crucial task today. Various technologies exist and are continuously being developed to capture high quality images of brain tissue, allowing experienced medical professionals to diagnose tumors in its early stage, so that they can take action to provide the necessary treatment based on their findings. With emerging interests and technological advancements in Artificial Intelligence and Machine Learning models, the task of manually classifying and segmenting medical images can become less burdensome. Automated diagnosis and classification of brain tumors using deep learning models can provide a way to overcome the problems of manual segmentation. This work aims to compare various studies involving brain tumor segmentation using deep learning methods on various aspects like accuracy the data used and hyper parameters of those architectures for task of brain tumor segmentation. The work compares existing model proposed using U-Net ,Link-Net,PSP-Net and FPN. Where the the performance stydy is compared using Accuracy, Dice Coefficient, Sensitivity, Precision and Specificity. The result shows PSP-Net achieves the highest Dice coefficient, indicating superior segmentation accuracy at the expense of computational intensity.