Search

Purpose: Large area x-ray imagers based on complementary metal-oxide-semiconductor (CMOS) active pixel sensor (APS) technology have been proposed for various medical imaging applications including digital breast tomosynthesis (DBT). The low electronic noise (50–300 e{sup −}) of CMOS APS x-ray imagers provides a possible route to shrink the pixel pitch to smaller than 75 μm for microcalcification detection and possible reduction of the DBT mean glandular dose (MGD). Methods: In this study, imaging performance of a large area (29 × 23 cm{sup 2}) CMOS APS x-ray imager [Dexela 2923 MAM (PerkinElmer, London)] with a pixel pitch of 75 μm was characterized and modeled. The authors developed a cascaded system model for CMOS APS x-ray imagers using both a broadband x-ray radiation and monochromatic synchrotron radiation. The experimental data including modulation transfer function, noise power spectrum, and detective quantum efficiency (DQE) were theoretically described using the proposed cascaded system model with satisfactory consistency to experimental results. Both high full well and low full well (LFW) modes of the Dexela 2923 MAM CMOS APS x-ray imager were characterized and modeled. The cascaded system analysis results were further used to extract the contrast-to-noise ratio (CNR) for microcalcifications with sizes of 165–400 μm at variousmore » MGDs. The impact of electronic noise on CNR was also evaluated. Results: The LFW mode shows better DQE at low air kerma (K{sub a} < 10 μGy) and should be used for DBT. At current DBT applications, air kerma (K{sub a} ∼ 10 μGy, broadband radiation of 28 kVp), DQE of more than 0.7 and ∼0.3 was achieved using the LFW mode at spatial frequency of 0.5 line pairs per millimeter (lp/mm) and Nyquist frequency ∼6.7 lp/mm, respectively. It is shown that microcalcifications of 165–400 μm in size can be resolved using a MGD range of 0.3–1 mGy, respectively. In comparison to a General Electric GEN2 prototype DBT system (at MGD of 2.5 mGy), an increased CNR (by ∼10) for microcalcifications was observed using the Dexela 2923 MAM CMOS APS x-ray imager at a lower MGD (2.0 mGy). Conclusions: The Dexela 2923 MAM CMOS APS x-ray imager is capable to achieve a high imaging performance at spatial frequencies up to 6.7 lp/mm. Microcalcifications of 165 μm are distinguishable based on reported data and their modeling results due to the small pixel pitch of 75 μm. At the same time, potential dose reduction is expected using the studied CMOS APS x-ray imager.« less

PURPOSE: The x-ray performance evaluation of digital x-ray detectors is based on the calculation of the modulation transfer function (MTF), the noise power spectrum (NPS), and the resultant detective quantum efficiency (DQE). The flat images used for the extraction of the NPS should not contain any fixed pattern noise (FPN) to avoid contamination from nonstochastic processes. The "gold standard" method used for the reduction of the FPN (i.e., the different gain between pixels) in linear x-ray detectors is based on normalization with an average reference flat-field. However, the noise in the corrected image depends on the number of flat frames used for the average flat image. The aim of this study is to modify the standard gain correction algorithm to make it independent on the used reference flat frames. METHODS: Many publications suggest the use of 10-16 reference flat frames, while other studies use higher numbers (e.g., 48 frames) to reduce the propagated noise from the average flat image. This study quantifies experimentally the effect of the number of used reference flat frames on the NPS and DQE values and appropriately modifies the gain correction algorithm to compensate for this effect. RESULTS: It is shown that using the suggested gain correction algorithm a minimum number of reference flat frames (i.e., down to one frame) can be used to eliminate the FPN from the raw flat image. This saves computer memory and time during the x-ray performance evaluation. CONCLUSIONS: The authors show that the method presented in the study (a) leads to the maximum DQE value that one would have by using the conventional method and very large number of frames and (b) has been compared to an independent gain correction method based on the subtraction of flat-field images, leading to identical DQE values. They believe this provides robust validation of the proposed method.

PURPOSE: This paper proposes the further development of a resolution modification routine which is used to simulate the presampling modulation transfer function (pMTF) of digital x-ray detectors. METHODS: It suggests a method to reconstruct anisotropic two dimensional (2D) pMTF matrices from the experimentally measured horizontal and vertical 1D pMTFs. In this study, the horizontal dimension of the detector is 17.3 cm, while the vertical one is 24 cm. This matrix is multiplied with the 2D Fourier transform of the super-sampled ideal input image to simulate blurring. Then, the restored image is sampled to form the pixels of the digital image. The authors suggest convolution with the comb function instead of the rectangular function to avoid the correction with the sinc function required by the latter. It is demonstrated that this correction is avoided when the comb function is used. Moreover, this study suggests a way to effectively sample the images in the case when the ratio between the "analog" pitch of the super-sampled input image and the pixel pitch of the digital x-ray detector is a semi-integer. RESULTS: The validation of the simulation algorithm demonstrated that when the comb function was used the average absolute difference between the pMTF measured from the output images and the input ones was less than 1%, while this was of 13% when the rectangular function was used. When a sinc correction was applied in the latter case the difference decreased again to less than 1%. CONCLUSIONS: The developed modification routine provides the means to simulate the spatial resolution of digital x-ray detectors under a wider range of conditions.