Simulation of calcification clusters in observer performancestudies for optimisation of digital mammography

Breast cancer is the most common cancer in women in the UK. Breast screening using mammography imaging is performed to detect cancers early and reduce the death rate from breast cancer. It is important that the effect of new imaging technologies on cancer detection is known prior to their use in breast screening. Clinical trials can measure this, however these are expensive and time consuming. In comparison, observer studies with simulated cancers can be performed in a fraction of the time. This thesis focuses on the simulation of calcification clusters and their use in observer studies. Once inserted into breast images the simulated calcification clusters have been shown to look realistic and have the correct contrast and sharpness. The calcification clusters have been used in observer studies to compare two different types of digital detector-computed radiography (CR) and direct digital (DR) systems, different dose levels and different image processing algorithms. Calcification detection is significantly poorer when using a CR system compared with a DR system, and also sensitive to dose used. It has also been shown that image processing has a significant impact on calcification detection. However, this difference in calcification detection is smaller than the difference in calcification detection due to differences in detector type or dose. Image processing was not found to significantly impact detection of non-calcification cancers. The calcification detection measured using observer studies was compared to the threshold gold thicknesses measured with the CDMAM phantom. Threshold gold thickness was found to be relevant to calcification detection, however the acceptable and achievable threshold limits set in the European quality control protocol using this phantom need revising. The results of this work are important, providing evidence that can be used when selecting the optimal digital detector, dose and image processing.