Mary Yip

Conversion of mammographic images to appear with the noise and sharpness characteristics of a different detector and x-ray system

PURPOSE Undertaking observer studies to compare imaging technology using clinical radiological images is challenging due to patient variability. To achieve a significant result, a large number of patients would be required to compare cancer detection rates for different image detectors and systems. The aim of this work was to create a methodology where only one set of images is collected on one particular imaging system. These images are then converted to appear as if they had been acquired on a different detector and x-ray system. Therefore, the effect of a wide range of digital detectors on cancer detection or diagnosis can be examined without the need for multiple patient exposures. METHODS Three detectors and x-ray systems [Hologic Selenia (ASE), GE Essential (CSI), Carestream CR (CR)] were characterized in terms of signal transfer properties, noise power spectra (NPS), modulation transfer function, and grid properties. The contributions of the three noise sources (electronic, quantum, and structure noise) to the NPS were calculated by fitting a quadratic polynomial at each spatial frequency of the NPS against air kerma. A methodology was developed to degrade the images to have the characteristics of a different (target) imaging system. The simulated images were created by first linearizing the original images such that the pixel values were equivalent to the air kerma incident at the detector. The linearized image was then blurred to match the sharpness characteristics of the target detector. Noise was then added to the blurred image to correct for differences between the detectors and any required change in dose. The electronic, quantum, and structure noise were added appropriate to the air kerma selected for the simulated image and thus ensuring that the noise in the simulated image had the same magnitude and correlation as the target image. A correction was also made for differences in primary grid transmission, scatter, and veiling glare. The method was validated by acquiring images of a CDMAM contrast detail test object (Artinis, The Netherlands) at five different doses for the three systems. The ASE CDMAM images were then converted to appear with the imaging characteristics of target CR and CSI detectors. RESULTS The measured threshold gold thicknesses of the simulated and target CDMAM images were closely matched at normal dose level and the average differences across the range of detail diameters were -4% and 0% for the CR and CSI systems, respectively. The conversion was successful for images acquired over a wide dose range. The average difference between simulated and target images for a given dose was a maximum of 11%. CONCLUSIONS The validation shows that the image quality of a digital mammography image obtained with a particular system can be degraded, in terms of noise magnitude and color, sharpness, and contrast to account for differences in the detector and antiscatter grid. Potentially, this is a powerful tool for observer studies, as a range of image qualities can be examined by modifying an image set obtained at a single (better) image quality thus removing the patient variability when comparing systems.

Simulation and assessment of realistic breast lesions using fractal growth models

A new method of generating realistic three dimensional simulated breast lesions known as diffusion limited aggregation (DLA) is presented, and compared with the random walk (RW) method. Both methods of lesion simulation utilize a physics-based method for inserting these simulated lesions into 2D clinical mammogram images that takes into account the polychromatic x-ray spectrum, local glandularity and scatter. DLA and RW masses were assessed for realism via a receiver operating characteristic (ROC) study with nine observers. The study comprised 150 images of which 50 were real pathology proven mammograms, 50 were normal mammograms with RW inserted masses and 50 were normal mammograms with DLA inserted masses. The average area under the ROC curve for the DLA method was 0.55 (95% confidence interval 0.51-0.59) compared to 0.60 (95% confidence interval 0.56-0.63) for the RW method. The observer study results suggest that the DLA method produced more realistic masses with more variability in shape compared to the RW method. DLA generated lesions can overcome the lack of complexity in structure and shape in many current methods of mass simulation.

Validation of a Digital Mammography Image Simulation Chain with Automated Scoring of CDMAM Images

A wide variety of digital mammography systems are available for breast cancer imaging, each varying in physical performance. However, the relationship between physical performance assessment and clinical outcome is not clear. Thus, a means of simulating technically and clinically realistic images from different systems would represent a first step towards elucidating the impact of physical performance on clinical outcome. To this end, a framework for simulating technically realistic images has been developed. A range of simulated test objects, including CDMAM have been used to determine whether the simulation chain correctly reproduces these objects thus validating the simulation framework. Results evaluated for two digital mammography systems have been promising, with simulated images proving similar to experimental images for Modulation Transfer Function and Normalised Noise Power Spectrum measurements differing by approximately 3%.

Image resampling effects in mammographic image simulation.

This work describes the theory of resampling effects within the context of image simulation for mammographic images. The process of digitization associated with using digital imaging technology needs to be correctly addressed in any image simulation process. Failure to do so can lead to overblurring in the final synthetic image. A method for weighted neighbourhood averaging is described for non-integer scaling factors in resampling images. The use of the method is demonstrated by comparing simulated and real images of an edge test object acquired on two clinical mammography systems. Images were simulated using two setups: from idealized images and from images obtained with clinical systems. A Gaussian interpolation method is proposed as a single-step solution to modelling blurring filters for the simulation process.

Monte carlo simulation of scatter field for calculation of contrast of discs in synthetic CDMAM images

This paper reports on a further development of an image simulation chain, and in particular, the inclusion of contrast degradation across an image using scatter to primary ratios calculated using Monte Carlo simulation The Monte Carlo technique, using the Geant4 toolkit, has been implemented to model the scatter conditions when imaging the CDMAM phantom with commercial digital mammography Observed differences between linear and cellular anti scatter grid are presented and discussed These results support previous assumptions taken by Yip et al.[1].

Realistic simulation of breast mass appearance using random walk

The aim of the present work was to develop a method for simulating breast lesions in digital mammographic images. Based on the visual appearance of real masses, three dimensional masses were created using a 3D random walk method where the choice of parameters (number of walks and number of steps) enables one to control the appearance of the simulated structure. This work is the first occasion that the random walk results have been combined with a model of digital mammographic imaging systems. This model takes into account appropriate physical image acquisition processes representing a particular digital X-ray mammography system. The X-ray spectrum, local glandularity above the insertion site and scatter were all taken account during the insertion procedure. A preliminary observer study was used to validate the realism of the masses. Seven expert readers each viewed 60 full field mammograms and rated the realism of the masses they contained. Half of the images contained real, histologically-confirmed masses, and half contained simulated lesions. The ROC analysis of the study (average AUC of 0.58±0.06) suggests that, on the average, there is evidence that the radiologists could distinguish, somewhat, between real and simulated masses.

Validation of a method to convert an image to appear as if acquired using a different digital detector

A method to convert digital mammograms acquired on one system to appear as if acquired using another system is presented. This method could be used to compare the clinical efficacy of different systems. The signal transfer properties modulation transfer function (MTF) and noise power spectra (NPS) were measured for two detectors - a computed radiography (CR) system and a digital radiography (DR) system. The contributions to the NPS from electronic, quantum and structure sources were calculated by fitting a polynomial at each spatial frequency across the NPS at each dose. The conversion process blurs the original image with the ratio of the MTFs in frequency space. Noise with the correct magnitude and spatial frequency was added to account for differences in the detector response and dose. The method was tested on images of a CDMAM test object acquired on the two systems at two dose levels. The highest dose images were converted to lower dose images for the same detector, then images from the DR system were converted to appear as if acquired at a similar dose using CR. Contrast detail curves using simulated CDMAM images closely matched those of real images.

Validation of a simulated dose reduction methodology using digital mammography CDMAM images and mastectomy images

The purpose of this study is to evaluate the effect of simulated dose reduction using CDMAM and mastectomy images acquired on two digital mammography systems High dose images have been artificially degraded to reduced dose levels by systematically adding filtered noise Automated scoring has been carried out on the degraded CDMAM images and on experimental CDMAM images, taken at the same corresponding reduced doses Contrast-detail curves were derived for both, at all doses, and compared Relative difference in the contrast-detail curves was approximately 5% overall for all four doses. For the mastectomy images noise power spectra were obtained and the ratio of experimental to synthetic low dose NPS profiles averaged for all doses at 1.04 The largest differences in the NPS profiles were found at the high spatial frequencies, corresponding with the differences in the small discs in the contrast-detail curves.

Automated scoring method for the CDMAM phantom

CDMAM phantoms are widely used in the Europe to assess the performance of mammography systems utilising small size and low contrast disc details. However, the assessment of CDMAM images by human observers is slow and tedious. An automated method for scoring CDMAM images (CDCOM) is widely available to address this issue. We have developed an alternative automated scoring tool to score CDMAM images, Quantitative Assessment System (QAS), for similarly removing inter- and intra- observer variability. This provides additional valuable information about the contrast and SNR of each gold disc within the image. The QAS scores CDMAM phantom images using a scanning algorithm. QAS scoring results were compared with human observers and with CDCOM. It was found that QAS was comparable with human observers in scoring, whereas CDCOM consistently scored a higher number of discs correctly in CDMAM images compared with QAS and human observers. QAS results have been used to analyse the effects of different digital mammography system modulation transfer functions (MTFs) on fine details for a number of systems in the form of contrast degradation factor (CDF) measurements. CDF curves for experimentally acquired CDMAM images were compared with those for simulated CDMAM images to assess the accuracy of contrast measurements.

A simulation framework for the comparison of digital mammography imaging technology

With the recent developments in digital mammography, it is becoming increasingly important to be able to compare different technologies used for detecting breast cancer. By using simulation tools, it may be made possible to not only compare the images generated by different technologies but also the effect of dose levels and other imaging parameters within the same system. Images of a test phantom (CDMAM) have been simulated in this study as a proof of principle. An image simulation chain has been devised from which various doses and other generic system parameters can be simulated. Preliminary results show that the method provides a good quality initial simulation of real CDMAM images.