The effect of image quality on cancer detection in mammography

Breast screening is now undertaken using digital mammography rather than screen film mammography systems. However, there is a wide range in image qualities associated with the use of the different digital detectors available and there is evidence of reduced cancer detection rates with powder phosphor computed radiography (CR) systems. Further work is required to investigate the link between clinical implications and measured image quality. However, clinical studies can be expensive and time-consuming and suffer from various confounding factors. Studies using simulation can be faster and reduce the number of confounding factors. This thesis develops a methodology to adapt patient mammograms to appear as if acquired with a different detector. The adaption methodology accounts for differences in sharpness, glare, noise, scatter and beam quality and has been validated. The procedure can be used for adapting both patient images and mathematically created images. A set of mammograms containing inserted calcification clusters and real noncalcification lesions has been adapted to appear with the image quality associated with four types of detector and used in an observer study to compare cancer detection rates. If the results were applied to a screening service, the predicted cancer detection rate would be between 15 and 38% less for powder phosphor CR compared to digital radiography (DR) technologies. Needle image plate CR is a newer technology and was significantly better than powder phosphor CR but the predicted cancer detection rate would be between 6 and 15% lower than for the DR system at the same dose. It is shown that measurements of threshold gold thickness using the CDMAM test object correlate well with the cancer detection rates from the observer study. It is concluded that threshold gold thickness is suitable for setting limiting image quality criteria, as in the European Guidelines, but that the actual limits may need reviewing.