The proposed algorithm can also solve the single label classification problem. For this purpose, like a multi-label state, first the nonparallel hyper planes of each label isare created on kernel space. Then with the, using of the margin data on kernel space and, the prior probability of labels and the statistical information ofin the neighborhoods of the test sample, we can decide on the label of a test sample.
To evaluate the proposed algorithm on the single label classification problem, it is tested to detect abnormal brain structures in MRI brain data than containingthat contain scans of high-high and low- grade glioma cases.
For this purpose, we use the MICCAI Brain Tumor Segmentation Challenge (BRATs 2015) training data, The. This data set contains about 300 high-high and low- grade glioma cases. Each data set has T1 MRI, T1 contrast enhanced MRI, T2 MRI, and T2 FLAIR MRI volumes with 1mm^3 voxel resolution.
Figure 9 shows Anan example of these images.