Please use this identifier to cite or link to this item:
Title: Mammogram Mass Classification Using Linear Differential Analysis Sequential Feature Selection Technique
Authors: Jain, Shipra
Supervisor: Singla, Sunil Kumar
Keywords: Mamogram;Fuzzy
Issue Date: 11-Aug-2015
Abstract: Cancer is one of the most deadly diseases affecting the world at large today. Exact cause of this disease is not known, only timely detection can help in curbing the repercussion of this disease. Breast Carcinoma is one such cancer affecting women population largely. Its early diagnosis is the only solution to decrease its wrath. Mammography is the most reliable method used for its early diagnoses. Radiologists are working diligently in curing this disease, but because of the low ratio of number of radiologist to the mammograms diagnosed, probability of misdiagnose increases which can be reduced by repeated x-ray image analysis. Computer Aided Detection techniques are being developed these days as an aid for radiologist as second opinion on crucial matters. Computer detection is used to reduce number of misdiagnosed cases which eventually help in decreasing the effect of trauma a patient undergoes because of incorrect malignancy diagnoses. In the present work done computer aided analysis of mammogram images has been done. Mammogram image database is taken online from mini-MIAS society. Images are first segmented by using fuzzy thresholding from which region of interest has been obtained. Two types of features i.e. texture and shape features of lesion image have been extracted. Best combination of texture feature has been selected using 〖LDA〗_sfs. Masses are classified into malignant and benign class using artificial neural network. Classification accuracy of 97.5% has been achieved by using the combination of proposed feature set.
Description: M.E. (Electronic Instrumentation and Control Engineering)
Appears in Collections:Masters Theses@EIED

Files in This Item:
File Description SizeFormat 
3556.pdf2.11 MBAdobe PDFView/Open    Request a copy

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.