An automatic scheme for brain tumor region detection from 3D MRI data based on enhanced intensity variation

At present, brain tumor segmentation from 3D MRI data has become much popular in biomedical instrumentation. Various automatic and semi-automatic methods have been developed for this purpose but with huge computational burden due to the enormous volume of 3D data. Therefore, an effective automatic approach for detecting a tentative Region of Interest (ROI), in which the presence of tumor is guaranteed, is highly demanding as it can help to investigate brain tumor with reduced computation time. In this paper, an automatic brain tumor region detection scheme is developed based on the variation of intensity distribution in 3D volumetric tumor and non-tumor region. First, an efficient scheme is developed utilizing the CDF of the intensity distribution which drastically reduces a large volume of non-tumor data. Next, in order to enhance separability between tumor and non-tumor region of the brain, a 3D mean filtering operation is carried out utilizing a spherical window. After that, the roughness on the surface of the mean-filtered volume is reduced by implementation of a smoothing operation on its surface. Next, a voxel-wise analysis is performed. For each voxel, an unsupervised classification is performed whether it is a tumor voxel or non-tumor voxel based on intensity distribution inside the voxel. The non-tumor voxels are discarded and a precise ROI is extracted by taking the surviving voxels which ensures the presence of the whole tumor within that region.