Open Collections

Detection of malignancy associated changes in cervical cell nuclei using feed-forward neural networks

University of British Columbia

It has been recognized that normal cells in the presence of a precancerous lesion undergo subtle changes that affect the DNA distribution in their nuclei. These changes have been termed Malignancy Associated Changes (MACs). This thesis examines the design of a classifier that separates normal slides from slides containing MACs in the presence of a severely dysplastic lesion. Classifiers were designed using discrimiriant functions and feed-forward neural net works with various structures. The discriminant function correctly separated MACs from normal cells with a classification rate of 61.6% for a 16904 cell test set. Neural network classifiers were able to achieve up to 72.5% separation for this cell-by-cell classification task when four hidden units were used. Using more than four hidden units led to a decline ill the test set performaice. The slide-by-slide classification rates were calculated for each classifier based on the distribution of classifier values for the cells on each slide. The discriminant function scored 695% on the test set containing 197 slides. The neural network classifiers all scored between 74% and 77% when used for slide-by-slide classification.

Thesis/Dissertation

application/pdf

2009-02-26

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/5141

Physics

University of British Columbia

UBCV

DSpace