BibTeX entry

@PHDTHESIS{201006Mary_Yip,
  AUTHOR={Mary Yip},
  TITLE={Image Simulation Framework for Digital Mammography Systems},
  SCHOOL={University of Surrey},
  MONTH=Jun,
  YEAR=2010,
  URL={http://www.bmva.org/theses/2010/2010-yip.pdf},
}

Abstract

A wide range of technology is available in commercial digital mammography; these vary greatly in physical performance measures such as Modulation Transfer Function (MTF) and Noise Power Spectrum (NPS). However, the relationship between a digital mamography system’s physical performance and its impact on the clinician’s ability to detect and characterise cancer from its images is unclear. Use of clinical trials to compare digital mammography technology would be expensive and time-consuming. Simulation of images acquired on such digital mammography systems provides a powerful way to compare and optimise technology. The aim of this work has been to develop and validate an image simulation framework to generate images as if acquired from specific digital mammography systems. Signal and noise transfer properties of various systems were characterised using detector response, MTF and NPS measurements. Straight-edge and flat-field images were synthesized to validate the image simulation framework with experimentally acquired images. Excellent agreement was seen between synthetically generated and corresponding experimentally acquired images. Theory was developed to investigate the resampling effects upon measured MTF using the image simulation framework. The image simulation framework was further extended to generate CDMAM phantom images. The CDMAM phantom provides a method to quantify the detectability of details imaged on a mammography system based on an observer’s ability to detect discs of varying contrast and diameter. Simulation of the CDMAM phantom image based on MTF and NPS alone was seen to be insufficient to create physically accurate CDMAM phantom images. Non-uniformity effects such as heel-effect, local pixel variance, geometric magnification and scatter effects were modelled to improve the image simulation framework. Reducing the radiation dose to the patient whilst preserving sufficient information in the resultant mammogram to diagnose cancer is a very important issue. Lower exposure levels can be used in digital mammography systems compared with screen- film. However, there is ambiguity as to the optimal exposure level as well as the effects upon the characterisation and detection of cancer lesions from such images. A multiple noise source model was developed taking into account the major sources of noise: electronic, quantum and structural, thus increasing the exposure range at which the system noise can be accurately simulated. Comparison of ‘dose reduced’ experimentally acquired flat-field, CDMAM phantom and mastectomy images with corresponding lower dose experimentally acquired images showed excellent results. This marks the first study to have simulated such images for direct comparison in dose-reduction.