Modelling FTIR spectral sata with Type-I and Type-II fuzzy sets for breast cancer grading

Naqvi, Shabbar (2014) Modelling FTIR spectral sata with Type-I and Type-II fuzzy sets for breast cancer grading. PhD thesis, University of Nottingham.

[img]
Preview
PDF - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Download (5MB) | Preview

Abstract

Breast cancer is one of the most frequently occurring cancers amongst women throughout the world. After the diagnosis of the disease, monitoring its progression is important in predicting the chances of long term survival of patients. The Nottingham Prognostic Index (NPI) is one of the most common indices used to categorise the patients into different groups depending upon the severity of the disease. One of the key factors of this index is cancer grade which is determined by pathologists who examine cell samples under a microscope. This manual method has a higher chance of false classification and may lead to incorrect treatment of patients. There is a need to develop automated methods that employ advanced computational methods to help pathologists in making a decision regarding the classification of breast cancer grade. Fourier transform infra-red spectroscopy (FTIR) is one of the relatively new techniques that has been used for diagnosis of various cancer types with advanced computational methods in the literature. In this thesis we examine the use of advanced fuzzy methods with the FTIR spectral data sets to develop a model prototype that can help clinicians with breast cancer grading.

Initial work is focussed on using the commonly used clustering algorithms k-means and fuzzy c-means with principal component analysis on different cancer spectral data sets to explore the complexities within them. After that, a novel model based on Type-II fuzzy logic is developed for use on a complex breast cancer FTIR spectral data set that can help clinicians classify breast cancer grades. The data set used for the purpose consists of multiple cases of each grade. We consider two types of uncertainty, one within the spectra of a single case of a grade (intra -case) and other when comparing it with other cases of same grade (inter-case). Features have been extracted in terms of interval data from various peaks and troughs. The interval data from the features has been used to create Type-I fuzzy sets for each case. After that the Type-I fuzzy sets are combined to create zSlices based General Type-II fuzzy sets for each feature for each grade. The created benchmark fuzzy sets are then used as prototypes for classification of unseen spectral data. Type-I fuzzy sets are created for unseen spectral data and then compared against the benchmark prototype Type-II fuzzy sets for each grade using a similarity measure. The best match based on the calculated similarity scores is

assigned as the resultant grade.

The novel model is tested on an independent spectral data set of oral cancer patients.

Results indicate that the model was able to successfully construct prototype fuzzy sets for the data set, and provide in-depth information regarding the complexities of the data set as well as helping in classification of the data.

Item Type: Thesis (University of Nottingham only) (PhD)
Supervisors: Garibaldi, J.M.
Miller, S.M.
Subjects: Q Science > QA Mathematics > QA150 Algebra
Faculties/Schools: UK Campuses > Faculty of Science > School of Computer Science
Item ID: 14321
Depositing User: EP, Services
Date Deposited: 22 Dec 2014 11:48
Last Modified: 14 Sep 2016 08:41
URI: http://eprints.nottingham.ac.uk/id/eprint/14321

Actions (Archive Staff Only)

Edit View Edit View