Date of Award

2017

Degree Type

Thesis

Degree Name

Doctor of Philosophy

School

School of Science

First Advisor

Leisa Armstrong

Second Advisor

Amiya Tripathy

Field of Research Code

080106

Abstract

Many developing countries are still facing challenges with limited access to fertility health services. Women face problems in conceiving due to many factors such as increasing age. In vitro fertilization (IVF) treatments can assist these women but are considered too expensive. Medical pathology laboratories are searching for novel technologies that can improve microscopic slide testing of female ovarian reproductive tissues. Current electronic methods for the assessment of human ovaries are not suitable for analysis of ovarian reproductive tissues. Ultrasound method cannot be used to identify small ovarian Non-Growing Follicles (NGFs) that are responsible for reproduction. A computer assisted approach to overcome the problems associated with manual microscopic analysis of ovarian reproductive tissues could be beneficial in increasing the accuracy and speed of the analysis. Few studies have reported on the use of images and other artificial intelligence techniques for ovarian tissue samples and have mostly concentrated on the analysis of cancer cells or ovarian animal tissues which are different from human ovarian reproductive tissues. Other studies using human ovarian reproductive tissues have been limited in terms of accuracy. This research examines the possibility of developing an automated computer approach which will improve the practices of these pathology laboratories to analyse female ovarian reproductive tissues and assist medical practitioners to provide necessary fertility treatment. The objective of the research was to study existing computerized methods used in various tissue assessments; identify the gaps and limitations; and to propose a novel method on digitized colour images acquired from ovarian reproductive tissue slides. The following major research question has been addressed by the research study: “How to develop an automated approach to assist pathology experts to identify ovarian reproductive NGFs (Non-growing follicles) or simply ovarian reproductive tissues using digital images acquired from type P63 (counter and non-counter stained) histopathology ovarian biopsy slides?” In order to answer this question, the research was carried out in a number of phases to examine existing computerized techniques for impact assessment of the ovarian reproductive tissue analysis. The research used a mixed method approach based on a case study using experimental and engineering methodologies. The study also employed quantitative and statistical data analysis methods. The research was carried out as a series of research activities including data collection, image processing, development of proposed approach, assessment factors (different magnification and different stains), validation of results with manual microscopic analysis results and development of the framework. A series of 7 different approaches were examined which started with basic image analysis technique. Modification and further medications were carried out to find the best possible approach which maintains the “gold standard” criteria in comparison to manual microscopic analysis results. A novel proposed approach was developed which used two phases: (1) phase one include pre-processing (intensity correction, filter operation, colour image segmentation, intensity clustering, feature extraction approach to find out the most suitable features); and (2) phase two for identifications of potential ovarian NGFs using shape, size and colour features that were extracted in phase one. It was found that the accuracy rate was above 90% for all magnifications and stains used in this research study which maintains the “gold standard” criteria in comparison to manual microscopic analysis results. To increase the accuracy rate and to diminish the false error rate classification approach was incorporated. The proposed approach established the most effect techniques in comparison to existing available approaches. A novel intensity correction was proposed and incorporated at the beginning of pre-processing, fast reliable novel filter operation was developed and incorporated for filter operation, colour image segmentation was considered to use the colour features for identification of region of interest (ROIs) from other tissues, extraction of features to capture NGFs’ characteristics, and incorporation of domain knowledge to identify NGFs. Validation was carried out with experts’ manual microscopic analysis results and similar regions were analysed to minimize the experts’ observation variability issues to improve the accuracy rate. A prototype software tool was developed in MATLAB platform, which enables a non-expert to easily use and analyse the ovarian reproductive tissues without changing any processing parameter automatically by giving the image magnification and image type as input parameter. The proposed approach was found to reduce the time and effort required for the analysis without any human intervention. The novelty of the research is that the approach was fully automated; non-experts will be able to use this approach for analysis; and no change of processing parameter is essential for new image batch/batches. The approach was also accurate, reliable and provided repeatable results in comparison to manual microscopic analysis results. Further work could explore the modification of tissue parameters that could be used for other tissue analysis.

Available for download on Monday, November 28, 2022

Share

 
COinS