Use this URL to cite or link to this record in EThOS:
Title: Image texture analysis of transvaginal ultrasound in monitoring ovarian cancer
Author: Ab. Hamid, Bidi
ISNI:       0000 0004 2732 1402
Awarding Body: Cardiff University
Current Institution: Cardiff University
Date of Award: 2011
Availability of Full Text:
Access from EThOS:
Access from Institution:
Ovarian cancer has the highest mortality rate of all gynaecologic cancers and is the fifth most common cancer in UK women. It has been dubbed “the silent killer” because of its non-specific symptoms. Amongst various imaging modalities, ultrasound is considered the main modality for ovarian cancer triage. Like other imaging modalities, the main issue is that the interpretation of the images is subjective and observer dependent. In order to overcome this problem, texture analysis was considered for this study. Advances in medical imaging, computer technology and image processing have collectively ramped up the interest of many researchers in texture analysis. While there have been a number of successful uses of texture analysis technique reported, to my knowledge, until recently it has yet to be applied to characterise an ovarian lesion from a B-mode image. The concept of applying texture analysis in the medical field would not replace the conventional method of interpreting images but is simply intended to aid clinicians in making their diagnoses. Five categories of textural features were considered in this study: grey-level co-occurrence matrix (GLCM), Run Length Matrix (RLM), gradient, auto-regressive (AR) and wavelet. Prior to the image classification, the robustness or how well a specific textural feature can tolerate variation arises from the image acquisition and texture extraction process was first evaluated. This includes random variation caused by the ultrasound system and the operator during image acquisition. Other factors include the influence of region of interest (ROI) size, ROI depth, scanner gain setting, and „calliper line‟. Evaluation of scanning reliability was carried out using a tissue-equivalent phantom as well as evaluations of a clinical environment. iii Additionally, the reliability of the ROI delineation procedure for clinical images was also evaluated. An image enhancement technique and semi-automatic segmentation tool were employed in order to improve the ROI delineation procedure. The results of the study indicated that two out of five textural features, GLCM and wavelet, were robust. Hence, these two features were then used for image classification purposes. To extract textural features from the clinical images, two ROI delineation approaches were introduced: (i) the textural features were extracted from the whole area of the tissue of interest, and (ii) the anechoic area within the normal and malignant tissues was excluded from features extraction. The results revealed that the second approach outperformed the first approach: there is a significant difference in the GLCM and wavelet features between the three groups: normal tissue, cysts, and malignant. Receiver operating characteristic (ROC) curve analysis was carried out to determine the discriminatory ability of textural features, which was found to be satisfactory. The principal conclusion was that GLCM and wavelet features can potentially be used as computer aided diagnosis (CAD) tools to help clinicians in the diagnosis of ovarian cancer.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available