Brain Tumor Image processing:A glance to recent studies

Image processing has been focused in several studies in brain tumor detection from the brain Magnetic Resonance Imaging (MRI) in order to improve accuracy of experts manual inspection. This work is an uptodate concise review of learning machine techniques in order to analysis their strengths and weaknesses for detection of malignant tissues and improvement of experts diagnostic capability. Many methods were discussed as follows: Threshold-based (Global/Local) , Region-based (Region-growing , Watershed) , Pixel-based (Fuzzy C Means, Markov Random Fields) , Model-based (Parametric Deformable Models, Level Sets ) , The atlas-based segmentation , K-NN techniques , Neural network, K-mean algorithms. Also , hybrid techniques were proposed including Combination of K-means and fuzzy c-means , FKSRG , Multi-region + multi-reference framework ,Generative probabilistic model + spatial regularization , probabilistic model plus localization , Non-rigid registration / atlas/ MRF , SVM / CRF , Decision Forests / tissue-specific Gaussian mixture models , SVM / Kernel feature selection , etc. We found that the machine learning approaches integrated with other approaches can offer a higher detection success rate , accuracy and sensitivity rates. But they are time consuming and it is better to improve this matter in the future works.