Digital Archives Initiative
Memorial University - Electronic Theses and Dissertations 5
menu off  add document to favorites : add page to favorites : reference url back to results : previous : next
 
 Search this object:
  
 0 hit(s) :: previous hit : next hit
  View:    
  previous page : next page
Document Description
TitleDiscrete cosine transform based feature extraction for computer aided detection of suspicious x-ray mammogram images
AuthorFlynn, Matthew T. (Matthew Thomas), 1985-
DescriptionThesis (M.Sc.)--Memorial University of Newfoundland, 2011. Medicine
Date2011
Paginationx, 122 leaves : ill.
SubjectBreast--Diseases--Diagnosis; Breast--Radiography; Cancer--Diagnosis--Computer simulation; Diagnosis, Radioscopic--Computer simulation
DegreeM.Sc.
Degree GrantorMemorial University of Newfoundland. Faculty of Medicine
DisciplineMedicine
LanguageEng
NotesBibliography: leaves 114-122.
AbstractOne of the best ways to decrease breast cancer mortality is through early detection. X-ray mammography is widely used to screen women with an increased risk of breast cancer. Computer aided detection (CAD) programs have been developed in an effort to boost efficiency and accuracy, but studies have shown that the CAD programs currently in use are not particularly effective. -- In this project, a new CAD algorithm was developed. The two main components of the method were the use of whole image classification and a novel feature extraction step using the discrete cosine transform. The features were generated from moments of the mean of square sections centered on the origin of the transform. Feature vectors were then run through k-nearest neighbour and naive Bayesian classifiers. -- It was found that the discrete cosine transform could be used to manually filter suspicious characteristics from images. Features extracted from the images were found to change dramatically when a mass was introduced into the image. Using a k-nearest neighbour classifier, sensitivities as high as 98% with a specificity of 66% was achieved. With a naive Bayesian classifier, sensitivities as high as 100% were achieved with a specificity of 64%.
TypeText
Resource TypeElectronic thesis or dissertation
FormatImage/jpeg; Application/pdf
SourcePaper copy kept in the Centre for Newfoundland Studies, Memorial University Libraries
RightsThe author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.
CollectionElectronic Theses and Dissertations
Scanning StatusCompleted
PDF File(6.41 MB) -- http://collections.mun.ca/PDFs/theses/Flynn_MathewT.pdf
CONTENTdm file name12930.cpd