We propose an automatic segmentation method based KNN,exploring small 3*3 kernels. The use of small kernels allows designing a deeper architecture, besides having a positive effect against over fitting, given the fewer number of weights in the network. We also investigated the use of intensity normalization as a pre-processing step, which though not common in KNN-based segmentation methods, proved together with data augmentation to be very effective for brain tumor segmentation in MRI images.
The use of symmetric distributions may not be satisfying: the amount of outlying data has to be the same in each dimension, and this has no reason to occur in multi-contrast MRI data. The proposed use of the MST distribution addresses specifically this problem. This is illustrated with statistical tests that reject Gaussianilty in both the healthy and pathological data cases. At last, one inconvenient of clustering approaches is that they usually require to use an atlas and to fix some sort of hyper parameters indicating for instance the expected number of outliers. In our novelty detection approach, we also have to set thresholds to distinguish intralesion classes but we propose a data driven way based on model selection tools and extreme value theory.In the paper differs from existing work in various aspects.
• The most of existing methods has ignored the poor quality images like images with noise or poor brightness.
• Neural network based brain tumor detection may provide better results; but due to training and testing phase it will comes up with some potential overheads i.e. poor in case of time complexity.
Our proposed method is first mixture model is fitted to the healthy subjects voxels. This reference model is used to detect voxels which exhibit abnormal MR features with respect to the reference model, in the healthy and pathological subjects. A second mixture model is fitted to the detected abnormal voxels and yields a clustering of these voxels into several classes. This improvement was obtained by minimizing the data heterogeneity from multi-site multi-scanner MRI acquisitions. Using morphological operation we have segment the image to detect the tumor. In feature extraction method we have to use textural features of affected region. Finally we classify whether tumor is present or not by using convolutional neural network.
• High accuracy is obtained and time consumption for detecting the tumor is very less.
• More secure
• High efficiency.
1. Watershed algorithm
Processor Type : Pentium -IV
Speed : 2.4 GHZ
Ram : 128 MB RAM
Hard disk : 20 GB HD
Operating System : Windows 7
Software Programming Package : Matlab R2014a
Natarajan P, Krishnan.N, Natasha Sandeep Kenkre, Shraiya Nancy, Bhuvanesh Pratap Singh, "Tumor Detection using threshold operation in MRI Brain Images" , IEEE International Conference on Computational Intelligence and Computing Research, 2012.