In this study, we adapt and apply model observers within the framework of realistic detection tasks in breast
tomosynthesis (BT). We use images consisting of realistic masses digitally embedded in real patient anatomical
backgrounds, and we adapt specific model observers that have been previously applied to digital mammography (DM).
We design alternative forced-choice experiments (AFC) studies for DM and BT tasks in the signal known exactly but
variable (SKEV) framework. We compare performance of various linear model observers (non-prewhitening matched
filter with an eye filter, and several channelized Hotelling observers (CHO) against human.
A good agreement in performance between human and model observers can be obtained when an appropriate internal
noise level is adopted. Models achieve the same detection performance across BT and DM with about three times less
projected signal intensity in BT than in DM (humans: 3.8), due to the anatomical noise reduction in BT. We suggest that,
in the future, model observers can potentially be used as an objective tool for automating the optimization of BT
acquisition parameters or reconstruction algorithms, or narrowing a wide span of possible parameter combinations,
without requiring human observers studies.
We studied the influence of signal variability on human and model observer performances for a detection task with
mammographic backgrounds and computer generated clustered lumpy backgrounds (CLB). We used synthetic yet
realistic masses and backgrounds that have been validated by radiologists during previous studies, ensuring conditions
close to the clinical situation. Four trained non-physician observers participated in two-alternative forced-choice (2-AFC)
experiments. They were asked to detect synthetic masses superimposed on real mammographic backgrounds or CLB.
Separate experiments were conducted with sets of benign and malignant masses. Results under the signal-known-exactly
(SKE) paradigm were compared with signal-known-statistically (SKS) experiments. In the latter case, the signal was
chosen randomly for each of the 1,400 2-AFC trials (image pairs) among a set of 50 masses with similar dimensions, and
the observers did not know which signal was present. Human observers' results were then compared with model
observers (channelized Hotelling with Difference-of-Gaussian and Gabor channels) in the same experimental conditions.
Results show that the performance of the human observers does not differ significantly when benign masses are
superimposed on real images or on CLB with locally matched gray level mean and standard deviation. For both benign
and malignant masses, the performance does not differ significantly between SKE and SKS experiments, when the
signals' dimensions do not vary throughout the experiment. However, there is a performance drop when the SKS signals'
dimensions vary from 5.5 to 9.5 mm in the same experiment. Noise level in the model observers can be adjusted to
reproduce human observers' proportion of correct answers in the 2-AFC task within 5% accuracy for most conditions.
In this study we estimated human observer templates associated with the detection of a realistic mass signal
superimposed on real and simulated but realistic synthetic mammographic backgrounds. Five trained naïve observers
participated in two-alternative forced-choice (2-AFC) experiments in which they were asked to detect a spherical mass
signal extracted from a mammographic phantom. This signal was superimposed on statistically stationary clustered
lumpy backgrounds (CLB) in one instance, and on nonstationary real mammographic backgrounds in another. Human
observer linear templates were estimated using a genetic algorithm. An additional 2-AFC experiment was conducted
with twin noise in order to determine which local statistical properties of the real backgrounds influenced the ability of
the human observers to detect the signal.
Results show that the estimated linear templates are not significantly different for stationary and nonstationary
backgrounds. The estimated performance of the linear template compared with the human observer is within 5% in
terms of percent correct (Pc) for the 2-AFC task. Detection efficiency is significantly higher on nonstationary real
backgrounds than on globally stationary synthetic CLB.
Using the twin-noise experiment and a new method to relate image features to observers trial to trial decisions, we found
that the local statistical properties preventing or making the detection task easier were the standard deviation and three
features derived from the neighborhood gray-tone difference matrix: coarseness, contrast and strength. These statistical
features showed a dependency with the human performance only when they are estimated within an area sufficiently
small around the searched location. These findings emphasize that nonstationary backgrounds need to be described by
their local statistics and not by global ones like the noise Wiener spectrum.
In this work we investigated the digital synthesis of images which mimic real textures observed in mammograms. Such images could be produced in an unlimited number with tunable statistical properties in order to study human performance and model observer performance in perception experiments.
We used the previously developed clustered lumpy background (CLB) technique and optimized its parameters with a genetic algorithm (GA). In order to maximize the realism of the textures, we combined the GA objective approach with psychophysical experiments involving the judgments of radiologists. Thirty-six statistical features were computed and averaged, over 1000 real mammograms regions of interest. The same features were measured for the synthetic textures, and the Mahalanobis distance was used to quantify the similarity of the features between the real and synthetic textures. The similarity, as measured by the Mahalanobis distance, was used as GA fitness function for evolving the free CLB parameters. In the psychophysical approach, experienced radiologists were asked to qualify the realism of synthetic images by considering typical structures that are expected to be found on real mammograms: glandular and fatty areas, and fiber crossings.
Results show that CLB images found via optimization with GA are significantly closer to real mammograms than previously published images. Moreover, the psychophysical experiments confirm that all the above mentioned structures are reproduced well on the generated images. This means that we can generate an arbitrary large database of textures mimicking mammograms with traceable statistical properties.
This work compares the detector performances of the recent Kodak Min-R EV 190/Min-R EV and current Kodak Min-R 2190/Min-R 2000 mammography screen-film combinations with the Kodak CR 850M system using the new EHR-M and standard HR plates. Basic image quality parameters (MTF, NNPS and DQE) were evaluated according to ISO 9236-3 conditions (i.e. 28 kV; Mo/Mo; HVL = 0.64 mm eq. Al) at an entrance air kerma level of 60 μGy. Compared with the Min-R 2000, the Kodak Min-R EV screen-film system has a higher contrast and an intrinsically lower noise level, leading to a better DQE. Due to a lower noise level, the new EHR-M plate improves the DQE of the CR system, in comparison with the use of the standard HR plate (30 % improvement) in a mammography cassette. Compared with the CR plates, screen-film systems still permit to resolve finer details and have a significantly higher DQE for all spatial frequencies.