Indian Journal of Science and Technology

Abstract

Objective: Medical Image Processing needs to have images to be processed in a more meaningful way for the diagnostic purposes. In the MRI diagnosis, the exact status of Brain tumor is not obtained. The objective of this research work is to find an MRI scanned image of brain is cancerous or not, because the benign stage is not visible so easily. Methods: In the research work, a novel approach is applied using Nearest Neighbour classification with Fuzzy Logic, directly accessing on the intensity of the pixels without omitting any region of the image. The Bayesian Classifier is further applied to classify the segmented regions such as normal and abnormal. The sequential matching is performed with every region using its feature space. It is also novel because it’s using similarity measures of the trained set images and not on the geometric shape modeling. Findings: The MRI scanned image is applied with NN classifier using along with Fuzzy logic gives 12% to 18% is better than Multi region Graph cut. Contrast to existing NN classifier, the execution time in BCEV (Bayesian Classifier using Eigen Vector)is low and the variance is 10-20% low in the proposed BCEV. Compared with an existing and other works, the proposed efficient sequential pattern matching algorithm for classified brain image system provides an efficient estimation and the variance is 5-10% high in the proposed ESPM than when compared to BCEV, 8-10% higher than NN classifier and fuzzy logic method and 10-15% higher than existing graph cut method. Application/Improvement: The results show that our proposed approach gives high level of accuracy, execution faster and higher flexibility in segmentation, classification and in matching.

Keywords: Classification, Image Segmentation, Pattern Matching, Similarity Measure