The effect of background structure on the detection of low contrast objects in mammography

The visual task of mammographic interpretation is considered in terms of the detection of a signal in the presence of noise, where the noise is taken to include a structure noise contribution from those areas of the imaged breast which do not contain the signal. The structure noise in a variety of typical mammographic parenchymal patterns was quantified and related to the area under observation using a statistical analysis of digitized image samples. Images of an anthropomorphic phantom were also analysed to establish it as a suitable test background for a series of contrast detail detection experiments. These experiments were performed with and without a structured background over a wide range of film dose, but at a fixed average film optical density. The presence of structure noise was found to reduce the detectability of low contrast objects, the effect becoming progressively smaller as the object size is reduced. Where the structured background was used, even large changes in dose to the film were found to produce little change in the overall contrast detail result except at the smallest detail diameters. These results are discussed in relation to existing theories of visual perception. It is suggested that the correlation between patient dose and cancer detection rate may be poorer than previously thought, as the detection task for objects larger than approximately 1 mm in diameter is dominated by the structure noise of the background parenchymal pattern rather than quantum noise.