Use this URL to cite or link to this record in EThOS:
Title: Morphological quantitation software in breast MRI : application to neoadjuvant chemotherapy patients
Author: Ahmed, Arfan
ISNI:       0000 0004 2751 5303
Awarding Body: University of Hull
Current Institution: University of Hull
Date of Award: 2012
Availability of Full Text:
Access from EThOS:
Access from Institution:
The work in this thesis examines the use of texture analysis techniques and shape descriptors to analyse MR images of the breast and their application as a potential quantitative tool for prognostic indication. Textural information is undoubtedly very heavily used in a radiologist’s decision making process. However, subtle variations in texture are often missed, thus by quantitatively analysing MR images the textural properties that would otherwise be impossible to discern by simply visually inspecting the image can be obtained. Texture analysis is commonly used in image classification of aerial and satellite photography, studies have also focussed on utilising texture in MRI especially in the brain. Recent research has focussed on other organs such as the breast wherein lesion morphology is known to be an important diagnostic and prognostic indicator. Recent work suggests benefits in assessing lesion texture in dynamic contrast-enhanced (DCE) images, especially with regards to changes during the initial enhancement and subsequent washout phases. The commonest form of analysis is the spatial grey-level dependence matrix method, but there is no direct evidence concerning the most appropriate pixel separation and number of grey levels to utilise in the required co-occurrence matrix calculations. The aim of this work is to systematically assess the efficacy of DCE-MRI based textural analysis in predicting response to chemotherapy in a cohort of breast cancer patients. In addition an attempt was made to use shape parameters in order to assess tumour surface irregularity, and as a predictor of response to chemotherapy. In further work this study aimed to texture map DCE MR images of breast patients utilising the co-occurrence method but on a pixel by pixel basis in order to determine threshold values for normal, benign and malignant tissue and ultimately creating functionality within the in house developed software to highlight hotspots outlining areas of interest (possible lesions). Benign and normal data was taken from MRI screening data and malignant data from patients referred with known malignancies. This work has highlighted that textural differences between groups (based on response, nodal status, triple negative and biopsy grade groupings) are apparent and appear to be most evident 1-3 minutes post-contrast administration. Whilst the large number of statistical tests undertaken necessitates a degree of caution in interpreting the results, the fact that significant differences for certain texture parameters and groupings are consistently observed is encouraging. With regards to shape analysis this thesis has highlighted that some differences between groups were seen in shape descriptors but that shape may be limited as a prognostic indicator. Using textural analysis gave a higher proportion of significant differences whilst shape analysis results showed inconsistency across time points. With regards to the mapping this work successfully analysed the texture maps for each case and established lesion detection is possible. The study successfully highlighted hotspots in the breast patients data post texture mapping, and has demonstrated the relationship between sensitivity and false positive rate via hotspot thresholding.
Supervisor: Gibbs, Peter; Pickles, Martin Darren Sponsor: Yorkshire Cancer Research Campaign
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: Medicine