A Professional Analysis and Evaluation of Computed Tomography Brain Tumor Images using SDNN for Segmentation and SOM-LS-SVM for Classification

摘要

Image processing is an area fascinating for researching, medical image applications in particular. The study of the Computed Tomography (CT) images considers image segmentation a very important and vital part in identifying the different kinds of tumor. Many researchers have presented many Neural Network (NN) algorithms for segmentation. However, those algorithms are prone to over fitting and require greater computational assets. Computed Tomography (CT) images play a vital role in the diagnosis of cerebral stroke as compared with the Magnetic Resonance (MR) images. As to the algorithms in general, the ones based on the Support Vector Machine (SVM) are preferable to the other algorithms for high dimensional spaces. This paper proposes new techniques of segmentation and classification for distinguishing brain tumors from CT brain images. Sparse Deep Neural Networks (SDNN) technique is utilized for the segmentation process based on the distributed representations. In SDNN, varying numbers of layers and their sizes can be used to provide different amounts of abstraction. The classification process follows the segmentation process by way of the performance of the Self Organizing Map-Least Squares-Support Vector Machine (SOM-LS-SVM). This technique aptly distinguishes both the uncertain and the diverse samples from the low density regions. The comparative analysis in terms of segmentation accuracy, classification accuracy, sensitivity, specificity, computation time, Mean Square Error (MSE) and Peak Signal to Noise Ratio (PSNR) is carried out between the proposed methodology and the other popular specified techniques such as K-NN with SVM, SVM, ANN. Experimental results unambiguously prove that in both the segmentation and the classification as well, the proposed technique, that is the SDNN followed by the SOM-LS-SVM produces far better results than the other techniques do.