A Thesis Report On BREAST TUMOR DETECTION BY SOFT COMPUTING TOOLS

Abstract

The research work is based on the title Breast Tumor Detection by newlineSoft Computing Tools and the same was presented in this thesis. Breast newlinecancer is one of the major threats to females; if spotted out earlier can be newlinecured whereas it can only be treated when identified in metastatic state. newlineOncologist depends on the reports given by the radiologist, who analyse newlineX-ray image called Mammogram. Detection of tumor in early stage is not newlineeasy to read from mammogram. Dense tissue and noise may disturb the newlinereadability of mammogram. newlineThere is a need of collaboration of engineering in medical field to newlinefulfil this problem. Proposed method is to detect tumor by engrossing newlinecomputer vision in medical field. Computer aided diagnosis involves newlinepipeline of work such as pre-processing, image enhancement, newlinesegmentation of image, extracting various features/character and newlineclassifier to categorize the result. newlineIn pre-processing various filter characters were analysed for their newlineperformance in dense mammogram and Duo-Median filtered is tailored newlinefor high density impulsive noise. newlineMammogram images were enhanced by cropping the unwanted newlineregion and also by removing pectoral muscle which is bright portion in newlinemammogram. This portion often segmented as ROI that diminish the newlineclassification efficiency. newlineEnhanced mammogram was classified with the combination of newlineFuzzy C Means clustering and Fuzzy Min Max Neural Network classifier. newlineThe same was optimized for minimum error by modified Grey Wolf newlineAlgorithm that gives is accuracy of detection upto 95.65 %. In another newlinecombination of detection pipeline modified Watershed that is Marker newlineControlled Watershed Algorithm is implemented for segmentation newlinefollowed by Deep Convolution Neural Network with multiple convolution, newlinepooling, rectification and fully connected layers for classification process. newlineThis gives an accuracy upto 98.4 % and other evaluation parameters newlinewere satisfactory newline