Please use this identifier to cite or link to this item:
Title: Segmentation of Ultrasound Breast Images
Researcher: Shrinath Pravin Mahadeorao
Guide(s): Kekre H. B.
Keywords: Local adaptive
texture feature images
University: Narsee Monjee Institute of Management Studies
Completed Date: 2014
Abstract: Breast cancer is the leading cause of death amongst women worldwide. However early detection can reduce the mortality rate substantially. B-mode ultrasound (US) imaging modality is one of the reliable tools used to detect and diagnose the breast cancer. In fact it can detect the cancer at an early stage where effective treatment is possible. Computer aided diagnosis system helps radiologists in the newlinedetection and diagnosis of the region of interest accurately. Image segmentation is one of the vital steps used in automated computer aided diagnosis system. Indeed accuracy of the overall diagnosis is newlinedepends on the detection and demarcation of region of interest. However US image characteristics such as varying echogenicity,heterogeneous texture patterns, irregular shape, fuzzy tumor newlineboundary makes segmentation challenging. Moreover poor quality images due to inherent artifact such as speckle, attenuation makes segmentation more challenging. Many methods are suggested in the literature for speckle removal before implementation of region extraction algorithms. However in order to reduce the complexity of the segmentation, we suggested to omit speckle removal step newlinedeliberately from the proposed segmentation algorithm. Therefore we newlineuse original ultrasound images directly as input to the segmentation process.During the literature survey and problem formulation we observed that US breast images have extreme random gray level distribution. This phenomenon is the major hurdle in achievement of accurate newlinesegmentation. We found that unsupervised learning (clustering) has a great potential in solving such problems. Here we proposed total six algorithms based on clustering for segmentation. Initially we proposed thresholding based clustering on texture feature images. In this method texture has been analyzed by using selected texture parameters proposed by Haralick. Primarily texture feature images are generated using Correlation, Variance, Sum variance and Sum average parameters.
Appears in Departments:Department of Technology Management

Files in This Item:
File Description SizeFormat 
00. title page.pdfAttached File157.32 kBAdobe PDFView/Open
01. declaration.pdf1.91 MBAdobe PDFView/Open
02. certificate.pdf2.61 MBAdobe PDFView/Open
03. certificate of examining committee.pdf2.17 MBAdobe PDFView/Open
04. table of contents.pdf113.96 kBAdobe PDFView/Open
05. abstract.pdf114.07 kBAdobe PDFView/Open
06. acknowledgments.pdf76.8 kBAdobe PDFView/Open
07. list of figures.pdf99.17 kBAdobe PDFView/Open
08. list of tables.pdf89.84 kBAdobe PDFView/Open
09. abbreviations.pdf81.24 kBAdobe PDFView/Open
10. chapter 1.pdf269.31 kBAdobe PDFView/Open
11. chapter 2.pdf240.06 kBAdobe PDFView/Open
12. chapter 3.pdf154.92 kBAdobe PDFView/Open
13. chapter 4.pdf448.54 kBAdobe PDFView/Open
14. chapter 5.pdf505.72 kBAdobe PDFView/Open
15. chapter 6.pdf1.09 MBAdobe PDFView/Open
16. chapter 7.pdf647.15 kBAdobe PDFView/Open
17. chapter 8.pdf118.79 kBAdobe PDFView/Open
18. reference.pdf129.53 kBAdobe PDFView/Open

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