Efficient dendronic creation, visualisation and analysis for the detection of stellates in digitised mammograms

Publication Type:
Thesis
Issue Date:
2006
Full metadata record
Files in This Item:
Filename Description Size
Thumbnail01Front.pdf6.49 MB
Adobe PDF
Thumbnail02Whole.pdf107.51 MB
Adobe PDF
Breast cancer can be controlled and treated successfully when detected at early stages. Since survival rates are highest and recurrence as well as treatment costs are lowest if the cancer is detected and treated at an early stage, it is critical to diagnose breast cancer in its earliest stage. However, the mammographic appearance of normal breast tissue varies widely and the signs of breast cancer are subtle. The evaluation of screening mammograms is a very difficult task where the radiologist must balance the requirement of high sensitivity for abnormalities (leading to high cancer detection) and high specificity for normality (keeping unnecessary biopsies to a minimum). Computer Aided Detection of abnormalities in breast screening can potentially improve individual radiologists’ performance, however, despite a great deal of research literature on image processing in mammography, the detection of cancerous mass lesions is still very difficult for many reasons. Masses are often of varying size, shape and density, at the same time exhibiting poor image contrast. In addition, many mass lesions and normal parenchymal tissues surrounding them look similar on mammograms. A novel method utilising dendronic analysis of mammograms has been carried out in this research. A dendrone is a hierarchical thresholding structure that can be automatically generated from a complex image. The dendrone structure captures the connectedness of objects and sub-objects during successive brightness thresholding. Based upon connectedness and changes in intensity contours, dendronic representations of objects in images capture the course to fine unfolding of finer and finer detail, which is invariant to lighting, scale and placement of the object within the image. Complex images can be autonomously analysed using the output of the dendrone to determine if they contain the signatures of particular target objects of interest, in this case, the signatures of stealth-like stellate mass lesions within mammograms. Dendronic analysis of mammograms has been overlooked due to the computational expense in the creation of the image dendrogram. Thus, in this research, a modified creation algorithm has been developed to dramatically decrease creation times. Further, a threedimensional data visualisation implementation has been developed to better visualise and understand the dendronic representation of the image. In the analysis of 50 mammogram images containing 15 cases of cancerous stellate mass lesions, the method proposed in this research detected 86.7% of the stellates. However, it is also evident that the number of false positives produced required further classification of the output. These results indicate that dendronic analysis of digitised mammogram images has the potential to provide a robust tool to aid in the screening of breast cancer. The analysis is completely automated with very little a-priori information required and shows that dendronic analysis is an excellent tool in the detection of stellates. Moreover, the technique can now be performed within a clinically acceptable timeframe.
Please use this identifier to cite or link to this item: