Convolutional Neural Networks for Multimodal Brain MRI Images Segmentation - A Comparative Study.

Manual segmentation of brain tumors from MRI images is very frustrating and time consuming for medical doctors, and relies on accurate segmentation of regions of interests. Convolutional Neural Networks (CNN) based segmentation has gained a huge amount of attention over the last few years due to its speed and automated aspect. As the CNN models are becoming more efficient for image analysis and processing, they increasingly defeat previous state-of-the-art classical machine learning algorithms. Through this study, we provide an overview of CNN-based segmentation models for quantitative brain MRI images segmentation. As this has become a fast expanding field, we will not survey the entire existing landscape of methods but we will focus on the three best outperforming algorithms according to evaluation parameters. Firstly, we review the current conventional methods and deep learning architectures used for segmentation of brain lesions. Next, we perform deep performance comparison based on accuracy and loss function of some relevant selected CNN methods. Finally, a critical analysis of the current study is made to identify all pertinent issues and limitations to work on.