Robust Segmentation of 3D Brain MRI Images in Cross Datasets by Integrating Supervised and Unsupervised Learning

With the rapid development of machine learning technology in recent years, image segmentation technology based on supervised learning or unsupervised learning has also made important progress and achieved many successful applications, such as the applications to medical imaging in the same time. However, both supervised and unsupervised segmentation methods have their own strong and weak points. In order to address this dilemma, in this paper, we proposed a robust method for 3D image segmentation that can not only maintain the advantages of the two kinds of learning methods, but also overcome their disadvantages, by integrating supervised and unsupervised learning technologies into one method effectively. The proposed method has been applied to brain MRI image segmentation with a variety of experiments on several open 3D brain MR1 datasets. Experimental results obtained in the work show that the proposed method, with strong adaptability and robustness, outperforms other state of the art segmentation approaches including both the supervised and unsupervised ones when applied to a new MRI dataset or a cross dataset without needing to be retrained by using the annotation information of the dataset.