We have developed a method for improving the accuracy of image-based diagnosis by providing the radiologist with image-reading and decision-making aids. The image-reading aid is a checklist of visual features to guide the radiologist through a systematic assessment of the features of a detected abnormality. The decision aid is a computer-based statistical decision rule that accepts the feature assessments made by the radiologist, and returns an advisory estimate of the probability that disease is present. I discuss the steps involved in developing and testing these aids, and present illustrative examples of their successful application in mammography. I describe other possible benefits of a feature-based diagnostic system in improving report standardization, training of radiologists, quality assurance, and automated report-writing. Finally, I discuss potentially greater improvements in diagnostic accuracy that may be realized through the future development of hybrid systems in which a radiologist and computer interact, and systems combining features from multiple tests or sources.

We propose a novel technique for estimation of image noise amplitude without a priori signal information. Knowledge of the normalized noise distribution is used to construct an approximate Wiener filter parametrized by the estimated noise amplitude. For a given noise amplitude, the resulting signal estimate is subtracted from the image to produce a sample noise estimate. The estimated noise amplitude is varied in order to maximize the probability that the noise estimate is a sample of the known noise distribution with the estimated variance. Probability is measured by the (chi) ² distribution. The technique is tested for several images by adding stationary zero-mean Gaussian noise with varying amplitude. The variation of estimated versus added noise variance is very nearly linear with unit slope for all of the images tested. The estimated noise variance for images with no added noise is generally small compared to the signal power unless the signal power spectrum is nearly white.

Because of the complex response of the human visual system, typical measurements of image system quality such as the detective quantum efficiency, mean transfer function, and signal-to- noise ratio cannot always be used to determine conditions for optimal perceptual image quality. Using a model of the human vision system, the ViDEOS/Sarnoff Human Vision Discrimination Model (HVM), this work demonstrates that human vision models provide a promising quantitative measure of image perceptual quality. The model requires an image and a matching reference image in order to determine the perceptual difference between the images at each point. A simple model of a digital amorphous silicon medical x-ray system is used to create the necessary images as a function of various design parameters. The image pairs are then analyzed by the HVM. In all cases the dependence of perceived image quality closely follows measures of image quality as determined by the HVM for many image system design variations. Increasing the detector size actually increases the image quality in the presence of either readout or input noise. The model was also used to optimize the image system for a specific task optimization. As an example, the effect of system design parameters on tumor identification in mammographic images is determined.

The simplest task for evaluation of image quality is detection of a known signal on a known background. For linear, shift-invariant imaging systems with stationary, Gaussian noise, performance of the ideal observer on this task is determined by the frequency-dependent noise- equivalent quanta (NEQ), defined as the ratio of the square of the system modulation transfer function (MTF) to the noise power spectrum (NPS). It is the purpose of this paper to show that a closely analogous expression applies without the assumption of shift-invariance or noise stationarity. To get this expression, we describe an object of finite support exactly by a Fourier series. The corresponding system description is the Fourier crosstalk matrix, the diagonal elements of which constitute a generalized MTF. Since this matrix is not diagonal, calculation of the ideal-observer performance requires a double integral over the frequency domain, but if we average the task performance over all possible locations of the signal, the off-diagonal elements average to zero and a single sum results. With one approximation, this expression takes the same form as in the case of shift-invariant imaging and stationary noise.

Variance bounds are attractive for use in assessing quantitative system performance because they tell us how well a system can perform an estimation task without specifying a particular estimator. The most familiar of the variance bounds is the Cramer-Rao Bound also known as the Fisher-Information Bound. While suitable for many estimation problems, the Cramer-Rao Bound is often unachievable when applied to the very noisy data from imaging systems in nuclear medicine. This leads to overly optimistic estimates of system performance. As a result, some researchers have turned to the Barankin Bound as an alterative to the Cramer-Rao Bound. Our main result is that if no unbiased estimator exists, the Barankin Bound is infinite and computational methods for finding the Barankin Bound are unstable. This result is most conveniently seen in a simple 1D test problem we have developed here for demonstration purposes. The implications of this work are that caution must be used in applying the Barankin Bound to a given problem. In the absence of an unbiased estimator, this bound is misleading and other measures must be used.

Statistical estimation techniques for medical images require well-founded models for the data formation and acquisition process. In our work, we attempt to quantify the appropriate empirical noise model for two alternative digital chest radiography modalities. In this study, we acquired image sets from the same clinical x-ray unit using photostimulable phosphor plates (PPP) and film/screen. The images were acquired in the range of 1-9 mR. The PPP data, at 1760 by 2140 pixels, were transferred directly to our workstations for analysis, and the films were digitized at 2048 by 2480 pixels (0.2 mm resolution for both modalities). 1024 by 1024 regions of interest (ROIs) without edge effects or artifacts were then extracted. Using Fourier transform methods, the noise power spectra were formed for all the images with smoothing applied near the axes to exclude frequency components due to beam profile, detector inhomogeneity and other structural noise. The noise power spectra were then integrated to find the variance at the exposure in question. This corresponds to an estimate of the non-structural variance. The data for all images obtained with one method were finally combined to form the variance function (variance as a function of the mean) for the modality. The data show that for the range of exposure considered, the PPP and digitized film modalities exhibit significant differences in noise structure. Subtle differences in the noise power spectra were seen. Of particular interest to our application, however, the variance function in the two cases was seen to differ in its functional form. The data from the photostimulable phosphor plates correspond well to a linear model (variance proportional to the mean) throughout most of the range investigated, whereas the digitized film variance function corresponds better to a higher-degree polynomial fit over the range considered. This study indicates that separate models may be warranted for different sources of digital chest radiographs and also suggests the type of assumption that, based on empirical findings, may be useful in such a framework.

The irradiation field is the diagnostic region of radiograph which has been exposed directly to x rays and has not been shielded from the source by x-ray opaque material. Such material may have been placed to shield vulnerable regions of the patient from unnecessary exposure, and also to partition the radiograph into sub-images, thereby permitting multiple exposures on the same plate. In this paper we describe an image analysis method to automatically detect and locate the irradiation field of a digital radiograph. The aim is to prepare the radiographic image for contrast-enhancement processing which is driven only by exposure data from the diagnostically useful part of the radiograph. The algorithm is broadly edge-based, and comprises four stages: (1) edge detection, culminating in a set of straight-line edge segments as non-iconic data structures; (2) clustering (where possible) these segments into longer lines in accordance with colinearity constraints and the like; (3) model matching by a trainable rule- based network to identify irradiation field boundaries; (4) masking, which acts upon the input image to blank out the regions not in the irradiation field. The network has been trained on a database of 2100 images, and has been tested in clinical use, delivering an accuracy for recognition of the irradiation field of better than 99%.

A scanned-beam digital mammography system in which the x-ray detector employs a phosphor coupled to a light sensitive CCD by a fiber optic taper is under development. The CCD is a narrow slot-shaped detector that is operated in time delay integration (TDI) mode to acquire a complete mammogram. The temporal response of the phosphor (persistence) can introduce blurring in the scan direction which is also dependent on the scanning velocity. A model based on a simple exponential decay of the luminescence was developed to predict the degradation due to phosphor persistence. The temporal response of several phosphor materials, including Gd₂O₂S:Tb, CsI:Tl, and BaFCl:Eu was measured under pulsed x-ray excitation. The luminescence lifetimes of the phosphors were obtained using an iterative deconvolution technique to correct for arbitrary x-ray pulse shapes. Using the luminescence decay model, the modulation transfer function (MTF) degradation due to phosphor persistence was predicted. The model was verified by comparing the predicted MTF for Gd₂O₂S:Tb with that measured experimentally in a prototype scanning system after correction for other sources of MTF degradation. The lifetime of Gd₂O₂S:Tb was found to be approximately 0.5 ms. Using the lifetime in the exponential decay model, the MTF was found to be markedly reduced at high scanning speeds. Other phosphors, such as CsI:Tl had much shorter lifetimes and would consequently cause less MTF degradation. Although Gd₂O₂S:Tb is widely used for film-screen mammography, its decay characteristics suggest that it may not be the best phosphor for a scanning digital system. Slower scan rates or alternative phosphors with shorter lifetimes are necessary to provide optimal image quality.

Grayscale accuracy and consistency, along with spatial frequency performance at high and low modulation levels were characterized for two types of 14 by 17-inch medical laser printers; conventional silver-film printers and a dry-process, pigment transfer printer. A total of 180 images of various test targets were created on a silver-film system operating in a clinical environment over a 30 day period. A similar number of films were produced on both a pigment transfer imager and a silver film printer under controlled laboratory conditions. Every care was taken to ensure that all systems were operating within the manufacturer's specifications. Evaluations consisted of objective densitometric measurements from uniform fields, grayscale step tablets, and spatial frequency gratings. Results comparing the tone and spatial performance of these systems are presented.

Phosphor based detectors proposed for digital mammography require multiple energy conversion and coupling stages which degrade imaging performance. We are investigating the potential of a hybrid photodiode array-CCD which will overcome this problem. The photodiode converts the x-ray energy directly into charge which is stored in the CCD. The detector is used in a slot-scan configuration to image the entire breast. Due to the large amount of charge generated in the photodiode, a charge divider is incorporated on the CCD to prevent detector saturation. Feasible values of design parameters such as photodiode thickness, operating voltages, and charge divider ratio were determined by modeling the signal and noise propagation through the detector. The quantum accounting diagram (QAD) and detective quantum efficiency (DQE) were used to measure imaging performance. A photodiode thickness of 1.5 mm ensures acceptable quantum efficiency. This thickness, however, causes significant geometric blur due to obliquely incident x-rays at edges of the detector unless the detector is curved by using a number of hybrid sections less than 2 cm long. No more than 90% of the photodiode charge can be discarded by the divider without causing a significant reduction in DQE due to direct x-ray interaction in the CCD. Using the design parameter values determined by the model, a high DQE up to the Nyquist frequency is achieved.

Digital x-ray imaging detectors for soft x-rays (1 to 50 keV) are significant for medical mammography, dental radiography, microradiography, x-ray diffraction, and x-ray telescope applications. For systems in which a scintillator is optically coupled to a photodetector, a small number of electrons are produced for each interaction in the scintillator (more than 10 e^- in well designed systems). However, x-rays interacting directly in the photodetector produce a large charge for each detected event (about 4,700 e^- for a 17 keV x-ray absorbed in silicon. These direct interactions can significantly increase image noise even when their interaction probability and contribution to image signal are low. In this paper we report on analytic investigations of the direct and indirect contributions to noise for a system consisting of a CsI scintillator, fiber optic coupler, and CCD photo-detector. The fiber optic coupler is shown to be essential to shield the CCD detector from direct interactions. A 3 mm thick fiber optic coupler is sufficient to eliminate most of the events that can noticeably increase noise.

We propose an empirical linear model for the scattered radiation distribution in chest radiographs in order to compensate digital x-ray images for the presence of radiation scatter. The total detected intensity distribution is convolved with two different convolution kernels and radiation scatter is modelled as the weighted sum of the two convolutions plus a constant factor. The weighting factors and the constant factor are optimized for a specific chest radiograph. This can be done by sampling the scattered radiation distribution in a limited area of the radiographic image. The scatter values are then used to determine the factors with a standard linear least squares technique. Another way to determine the factors is by utilizing lookup tables (LUTs), giving the relationship between the factors and the acquisition parameters. These LUTs are constructed with data from previous cases. We give some preliminary results and compare the model with some other convolution based models.

A prototype scanning-beam digital x-ray system for cardiac fluoroscopy has been constructed. Source-to-detector distance is 94 cm with the subject positioned near the source. The 4-kW source operates at 70-110 kVp and has an electromagnetically-scanned 25-cm-diameter transmission target. The target is at ground potential and is directly liquid cooled for continuous full-power operation. The source collimator has 22,000 holes whose axes are aligned with the center of the detector array. Beam divergence through the 0.38-mm-diameter collimator holes is matched to the 1.8-cm diameter of the detector array. The detector is a 96- element scintillator array optically coupled to a 96-channel photomultiplier tube. A narrow (0.6 degree half-angle) x-ray beam scans the 19-cm-diameter field of view at 30 frames/sec. A two-dimensional shift-and-add reconstruction algorithm produces a narrow-angle classical tomographic view of the subject in real time. The small detector area and large patient- detector distance result in negligible detected x-ray scatter. Signal-to-noise ratio is calculated to be equal to conventional fluoroscopic systems with ten times less patient skin exposure and better than four times less patient integral dose. Exposure reduction is due to the elimination of x-ray scatter and the anti-scatter grid, increased detector DQE, and geometric considerations.

Convolution-filtering methods and direct beam stop techniques have previously been investigated for estimation of scatter-glare distribution in images acquired using digital fluoroscopic systems. The purpose of this study is to compare the error in scatter-glare estimation for a direct beam stop technique and a convolution technique. Also, the change in scatter-glare intensity due to contrast material injection for ventriculography and coronary arteriography was quantified. Three different methods were used to estimate scatter-glare intensity in humanoid chest phantom and animal model images. A lead strip was used to scan the x-ray field for direct measurement of scatter-glare intensity in every pixel in the image. An array of lead beam-stops were used in order to sample scatter-glare intensity. An interpolation technique was used to estimate scatter-glare intensity for the remaining pixels in the image. Scatter-glare intensity was also estimated using a convolution filtering technique. This technique utilizes exposure parameters and image gray levels to assign equivalent Lucite thicknesses for every pixel in the image. The thickness information is then used to estimate scatter-glare intensity on a pixel-by-pixel basis. Finally, an array of lead beam-stops were used to measure the change in scatter-glare intensity due to contrast material injection. This was done by leaving the array of lead beam-stops in the x-ray beam for ventriculography and coronary arteriography procedures. The results indicate that the array of lead beam-stops produces significant errors in estimating scatter-glare intensity. This is primarily due to the fact that contrast material injection significantly changes scatter-glare intensity. The scatter- glare intensity was changed by up to 19% and 88% during coronary arteriography and ventriculography, respectively. In conclusion, contrast material injection significantly changes the scatter-glare intensity during coronary arteriography and ventriculography procedures. Therefore, convolution-filtering techniques are more suited for cardiac imaging, where scatter- glare intensity significantly changes during image acquisition.

Radiotherapy portal imaging is the process of producing images using a radiation treatment linear accelerator or cobalt 60 unit. Portal images are used to evaluate the position of the radiation treatment beam and placement of shielding blocks with respect to the patient's anatomy. This paper discusses types, characteristics and clinical use of radiation therapy portal imaging systems.

The concept of radiographic equalization has previously been investigated. However, a suitable technique for digital fluoroscopic applications has not been developed. The previously reported scanning equalization techniques cannot be applied to fluoroscopic applications due to their exposure time limitations. On the other hand, area beam equalization techniques are more suited for digital fluoroscopic applications. The purpose of this study is to develop an x- ray beam equalization technique for digital fluoroscopic applications that will produce an equalized radiograph with minimal image artifacts and tube loading. Preliminary unequalized images of a humanoid chest phantom were acquired using a digital fluoroscopic system. Using this preliminary image as a guide, an 8 by 8 array of square pistons were used to generate masks in a mold with CeO². The CeO² attenuator thicknesses were calculated using the gray level information from the unequalized image. The generated mask was positioned close to the focal spot (magnification of 8.0) in order to minimize edge artifacts from the mask. The masks were generated manually in order to investigate the piston and matrix size requirements. The development of an automated version of mask generation and positioning is in progress. The results of manual mask generation and positioning show that it is possible to generate equalized radiographs with minimal perceptible artifacts. The equalization of x-ray transmission across the field exiting from the object significantly improved the image quality by preserving local contrast throughout the image. Furthermore, the reduction in dynamic range significantly reduced the effect of x-ray scatter and veiling glare from high transmission to low transmission areas. Also, the x-ray tube loading due to the mask assembly itself was negligible. In conclusion it is possible to produce area beam compensation that will be compatible with digital fluoroscopy with minimal compensation artifacts. The compensation process produces an image with equalized signal to noise ratio in all parts of the image.

A recently introduced method for quantitative analysis of images of the American College of Radiology (ACR) mammography accreditation phantom has been extended to include signal- to-noise-ratio (SNR) measurements, and has been applied to survey the image quality of 54 mammography machines from 17 hospitals. Participants sent us phantom images to be evaluated for each mammography machine at their hospital. Each phantom was loaned to us for obtaining images of the wax insert plate on a reference machine at our institution. The images were digitized and analyzed to yield indices that quantified the image quality of the machines precisely. We have developed methods for normalizing for the variation of the individual speck sizes between different ACR phantoms, for the variation of the speck sizes within a microcalcification group, and for variations in overall speeds of the mammography systems. In terms of the microcalcification SNR, the variability of the x-ray machines was 40.5% when no allowance was made for phantom or mAs variations. This dropped to 17.1% when phantom variability was accounted for, and to 12.7% when mAs variability was also allowed for. Our work shows the feasibility of practical, low-cost, objective and accurate evaluations, as a useful adjunct to the present ACR method.

In order to determine an objective measure of a system's image quality, we developed a simple, non-invasive measurement procedure to determine the detective quantum efficiency of digital radiographic systems, especially image intensifier-tv systems. Therefore we set up measurement procedures for the quantities intensity transfer function (ITF) (also called characteristic curve), modulation transfer function (MTF), noise power spectrum (NPS) and the low frequency drop (LFD). The quantities ITF, MTF, NPS and LFD are determined by the analysis of images of simple, standardized test objects (a slit, Al-filters of different thickness and a lead disk). The images are automatically evaluated by means of an Apple Macintosh workstation and the program NIH image with some special extensions. The resulting quantities MTF, NPS, LFD are combined to determine the noise equivalent quanta (NEQ) and the detective quantum efficiency (DQE). By means of this measurement procedure quantities, that describe the objective image quality like NEQ and DQE, can be determined in a simple way. Only a set of 45 images is needed for diagnosis of a system. This method provides a powerful analysis tool for image quality, that is applicable in the field and can be done from a remote location. It may be used in a clinical environment (e.g. in constancy testing).

The goal of computed radiography (CR) acceptance testing is to establish the storage phosphor reader and phosphor screen baseline performance. Quality control testing then detects changes in the CR system that could affect radiographic image quality. These tests currently require multiple exposures that are laser printed on film, and evaluated subjectively and using optical density measurements. Soft copy digital images have rarely been used to characterize CR systems. This paper presents an automated test target for the acceptance and quality control testing of Kodak storage phosphor readers. By using digital images for testing, the CR system can be evaluated independent of laser printer, film, and film processing effects.

We are examining the feasibility of performing digital mammography by combining a storage- phosphor image receptor with a highly efficient x-ray system. The image receptor consists of Fuji series HR-V high resolution imaging plates and a Fuji 9000 reader. The x-ray system was developed using multiparameter optimization techniques, with the goal of reducing patient dose as much as possible while retaining acceptable imaging performance. We have measured sensitometric properties, modulation transfer function (MTF), and noise power spectrum (NPS) of the Fuji plates with low-energy x-ray spectra. We have used the measurements, along with information about the x-ray system, to estimate signal-to-noise ratios (SNRs) for objects in a contrast-detail (C-D) phantom. We present the results of our measurements on the Fuji plates, comparisons of calculated and observed C-D diagrams for this system and a conventional system, and comparisons of phantom images and doses for this system to images and doses for a conventional system. We conclude that digital mammography with the system studied is at least feasible since phantom image quality is comparable to that of a conventional system at dose levels that are somewhat lower.

The characteristics of a novel, very high resolution film/screen system for mammography are compared with those of a conventional film/screen system for mammography. The image quality characteristics of these systems are measured in terms of Wiener spectra, MTF and sensitometry. These parameters then determine the signal-to-noise (SNR) transfer properties of these systems in terms of the contrast transfer function (CTF), the detective quantum efficiency (DQE) and the noise equivalent quantum efficiency (NEQ). Furthermore, the SNR transfer characteristics are modeled and rationalized by means of the light scattering properties of the intensifying screens and the scintillation spectra of the phosphor. The new system displays superior SNR properties ascribable to its x-ray absorption attributes and the light scattering properties of the phosphor coating.

The basic performance characteristics of a wide variety of mammographic screen-film combinations have been experimentally determined. The performance characteristics studied include absolutely calibrated sensitometry, modulation transfer functions, noise power spectra, noise equivalent quanta, and detective quantum efficiency. Specific advantages of various systems are discussed as well as their potential limitations.

A design concept of a mammography system is presented that is expected to reduce the background radiation without the penalty of the increased dosage or time. The monochromatic radiation from Mo target is Bragg reflected from a crystal mirror fabricated as per the Johansson geometry. Radiation then comes to a focus on the opposite side of the source point, and a pinhole located at this focal point will filter out the scattering from the bulk of the breast tissue. The focal spot size should be identical to the target spot size under an ideal circumstance. LiF crystal was assumed for the study, but any crystal with appropriate lattice constant and reflectivity can be adopted, depending on the difficulty of fabrication. Expected technical difficulties are presented and discussed.

The purpose of this work is to examine the feasibility of dual-energy techniques to enhance the detection of microcalcifications in digital mammography. The digital mammography system used in this study consists of two different mammography systems; one is the conventional mammography system with molybdenum target and Mo filtration and the other is the clinical version of a low dose x-ray system with tungsten target and aluminum filtration. The low dose system is optimized for screen-film mammography with a highly efficient scatter rejection device built by Fischer Imaging Systems for evaluation at NIH. The system was designed by the University of Southern California based on multiparameter optimization techniques. Prototypes of this system have been constructed and evaluated at the Center for Devices and Radiological Health. The digital radiography system is based on the Fuji 9000 computed radiography (CR) system which uses a storage phosphor imaging plate as the receptor. High resolution plates (HR-V) are used in this study. Dual-energy is one technique to reduce the structured noise associated with the complexity of the background of normal anatomy surrounding a lesion. This can be done by taking the advantage of the x-ray attenuation characteristics of two different structures such as soft tissue and bone in chest radiography. We have applied this technique to the detection of microcalcifications in mammography. The overall system performance based on this technique is evaluated. Results presented are based on the evaluation of phantom images.

Three-dimensional ultrasound (3DUS) offers a valuable approach to many diagnostic imaging problems and provides a relatively low cost alternative to 3D imaging modalities such as MRI and CT. To obtain a 3DUS image, a clinical ultrasound scanner is used to produce a sequence of adjacent 'slices' of the region of interest, which are recorded and used to reconstruct a volume. The required location of each slice may be found either by a priori knowledge of the scan path (mechanized scanning) or by tracking the path of the transducer at scan time (freehand scanning). For some imaging applications the scan path is irregular, making mechanized scanning difficult or impossible. Some examples are: imaging vascular disease in the carotid arteries, joint imaging and obstetrics. We have developed a freehand 3DUS system that is capable of such imaging tasks. We used phantoms containing a wire and a 50% stenosed cylinder phantom to test our system for vascular applications. We found the mean error in locating any image plane to be less than 0.1 mm ((sigma) equals 0.05 mm), with minimal geometric distortion. The degree of stenosis of the phantom was measured from a reconstructed volume to be 52.6% ((sigma) equals 2%). We have started a clinical stenosis study and a pilot joint imaging study. Our clinical results indicate that measurement of stenosis in patients should be possible.

This paper presents a new method of two-dimensional phase correction for ultrasound imagery, in which a separate source and receiver are employed. By varying the phase and amplitude across in input wavefront and studying the output image it was possible to improve image quality. It is demonstrated that adaptive optics methods can improve ultrasonic imaging.

Photoacoustic signals generated by breasts irradiated with short microwave, infrared or optical pulses could be used to detect breast cancer. Since radiation at this spectrum is non-ionizing, the photoacoustic approach provides a special safety feature. The purpose of the paper is to present a means to predict photoacoustic pressure signals for different breast phantoms and irradiation conditions. The photoacoustic wave equation was derived for linear, non-viscous liquid media. The equation was solved assuming uniform acoustic properties in an infinite medium. Compressed breast phantoms were used as the objects of simulation. The spatial dependence of electromagnetic energy absorption was given by another research paper of this conference. The time dependence of the absorption was assumed to be either uniform or bell- shaped. Photoacoustic pressure signals received by transducers at different locations were calculated numerically.

The aim of the investigation reported here is to clarify the way in which spectral-modifying artefacts, such as tissue attenuation, compromise pulse-wave Doppler measurements, and to accurately measure the magnitude of the corrupting influence of attenuation under controlled laboratory conditions. A theoretical description of the structure of the pulse-echo sequence from a moving scatterer field is constructed from first principles by utilizing a time-domain description of the Doppler process. It is demonstrated that the essential features of the echo signal may be rather more accurately described by a wavelet-, rather than by a Fourier-, transform, and that the power spectrum of the Doppler signal does not necessarily encode the range of the scatterer velocities present in the (pulse-echo) sampling volumes. The analysis provides a better understanding of the origins of the significant levels of noise present in pulse-wave Doppler signals, and allows a novel approach towards noise-reduction -- by zero manipulation in complex frequency space -- to be developed.

An image intensifier-based rotational volume tomographic angiography imaging system has been constructed. The system consists of an x-ray tube and an image intensifier that are separately mounted on a gantry. This system uses an image intensifier coupled to a charge coupled device (CCD) camera as a two-dimensional detector so that a set of two-dimensional projections can be acquired for a direct three-dimensional reconstruction (3D). A computer simulation has been performed to quantify the reconstruction error due to using 'a single circle cone beam geometry.' The results from the computer simulation indicate that the reconstruction error is relatively small compared to object contrast encountered in intra- arterial angiography. In this study, a scatter and veiling glare correction technique is reported. This technique put a beam stop array behind a phantom and in front of the image intensifier to estimate the scatter and veiling glare distribution. Then the primary beam distribution was estimated by subtracting the scatter and veiling glare distribution from the total detected radiation distribution. After obtaining the primary projections, a direct 3D reconstruction was obtained. The experimental results indicate that image contrast can be substantially improved by applying our scatter and veiling glare correction technique.

Three-dimensional, conebeam image reconstruction from planar, transmitted x-ray projections is the most dose-efficient method for computed tomography. A number of factors, including the number of projections acquired, the cone angle subtended by the detector, the detector bit- depth, and the amount of noise in the projections, influence the quality of the reconstruction. Using simulations, we investigate the effects of these parameters on the signal-to-noise level attainable using the Feldkamp algorithm for conebeam reconstruction.

Treatment of subarachnoid aneurysms with endovascular techniques (e.g. placement of Guglielmi coils) is currently limited by the inability to visualize the neck of the aneurysm after the initial coils have been placed. Coils projecting into parent vessels may cause thrombosis, while incomplete filling leads to regrowth of the aneurysm. Since the procedure is performed using a gantry-mounted x-ray image intensifier (XRII), we have used this system to obtain 2- dimensional (2-D) images over approximately 200 degrees and reconstructed a 3-D image of the embolizing coil, residual aneurysm, and the parent vessel. The required data can be acquired, reconstructed and presented to the neuroradiologist during the interventional procedure. We have characterized an existing clinical C-arm radiographic system, and have developed correction procedures which provide a consistent and adequate data set for standard CT reconstruction using convolution-backprojection techniques. The angle of acquisition for each image is recorded using a hub-mounted angle encoder accurate to within plus or minus 0.3 degrees. We have characterized the XRII distortion over the rotation, and can correct images acquired at any known angle to within plus or minus 0.06 pixels. We have also measured the motion of the center of rotation (reproducible to within plus or minus 0.13 pixels) and correct for the displacement using an image shift and interpolation algorithm. Finally, we have investigated the effects on CT reconstruction of variable dilutions of contrast agent during the cardiac cycle, and have shown the contrast-to-noise ratio (CNR), in the absence of photon noise, to be 28. This is larger than the CNR which we achieve due to photon noise alone.

Recently there has been increasing interest in obtaining three-dimensional reconstructions of arterial vessels from multiple planar radiographs (obtained at angles around the object). Interventional angiography is the motivating application behind this research. Different methods have been proposed to acquire the planar data such as a gantry mounted x-ray image intensifier (XRII) or a C-arm mounted XRII. In order to obtain a three-dimensional reconstruction from a C-arm mounted XRII the trajectory of the source and detector system must be characterized. We have designed a calibration system that provides the necessary trajectory information using uniquely identifiable markers positioned on a cylinder. This calibration ring is to be placed around the patient's head, and consists of steel balls positioned in a predefined arrangement on a cylindrical acrylic support. The balls are arranged such that calibration can be done from almost any partial view, allowing reconstruction of a region of interest (ROI). Steel balls are placed around an acrylic cylinder, restricted to a band approximately 8.5 mm wide, thereby obscuring only a small fraction of the image. In this case the radiograph includes the region of interest (ROI) as well as a partial view of the calibration ring. This enables us to recover the geometry of X-ray imaging system from each individual frame. We call this process 'dynamic calibration' as opposed to 'off-line' calibration procedures which try to characterize the motion of C-arm before introducing the patient.

We have developed a new, simple technique for accurate determination of the imaging geometry for calibration of projection imaging systems. Images of a phantom containing lead beads were obtained after it was placed at arbitrary, unknown locations which were near the isocenter of a radiation therapy simulator. The location of the lead beads were identified in the images. A computer model of the phantom was generated and positioned at an arbitrary location. Projection images of the model were generated and compared with the actual image locations. A modified projection-Procrustes algorithm was then used to translate and rotate the phantom model until it aligned optimally with the set of lines connecting the focal spot with the corresponding image points. The rotation matrix and translation vector relating the multiple projection geometries were then determined using a Procrustes algorithm and the calculated positions of the phantom for each projection. For a wide range of starting points, the technique converged to 3D locations with an approximate precision of 0.1 cm in position and approximately 0.2 degrees in orientation. The calculated angles and distances agreed with the readouts on the imaging systems to within approximately 0.25 degrees.

A method has been developed to allow mammographic differential diagnosis based upon the 3- D orientation and morphology of mammary calcifications. Two to seven digital radiographs of a cluster of calcifications are acquired on a prone stereotactic breast biopsy system. The images are segmented using a recursive region-growing algorithm. Inclusion of calcific material is dependent upon local statistics calculated over a region surrounding each calcification. Segmentation is aided by correlation of the calcifications in two or more views through analysis of their positions, sizes and shapes. The location of each calcification is determined geometrically; the shape of each calcification is calculated using the segmented images and a simulated annealing reconstruction method. Image segmentation and reconstruction can be reproducibly performed with an accuracy of 0.1 mm, which is sufficient to perform 3-D morphologic analysis. Biopsy specimens and in vivo calcifications have been examined. In instances where calcifications are associated with amass, we can distinguish preferentially peripherally distributed calcifications from homogeneously distributed calcifications. We have also been able to elucidate the linear distribution of calcifications contained within the ductal system. In a preliminary ROC study involving 3 radiologists and 26 lesions (5 malignant), specificity increased when 3-D images were included in the diagnostic evaluation, resulting in an increase of A_z from 0.66 to 0.88 (p equals 0.0039).

Conventional magnetic resonance images have uniform spatial resolution across the entire field of view. We recently described a method of creating MR images with user-specified spatial resolution along one dimension of the field of view. This paper presents the 2D generalization of this technique which allows the user to specify arbitrary spatial resolution in arbitrary 2D regions. These images are reconstructed from signals which sparsely sample the k-space representation of the image. Therefore, locally focused images can be acquired in less time than that required by Fourier imaging with uniformly high resolution. In this paper we show how to increase the temporal resolution of dynamic imaging (e.g. interventional imaging) by using high resolution in areas of expected change and lower resolution elsewhere. Alternatively, by matching the local spatial resolution to the expected edge content of the image, it is possible to avoid the localized truncation artifacts which mark Fourier images reconstructed from the same number of signals. For example, we show how proton spectroscopic images of the head may be improved by using high resolution in the neighborhood of scalp lipids which might otherwise cause truncation artifacts.

Conventional tomographic imaging techniques are nonlocal: to reconstruct an unknown function f at a point x, one needs to know its radon transform (RT) f((theta) ,p) for all ((theta) ,p). Suppose that one is interested in the recovery of f only for x in some set U. We call U the region of interest (ROI). Define the local data as the integrals of f along the lines that intersect the ROI. We propose algorithms for finding locations and values of jumps (sharp variations) of f from only the local data. In case of transmission tomography, this results in a reduction of the x-ray dose to a patient. The proposed algorithms can also be used in emission tomographies. They allow one: to image jumps of f with better resolution than conventional techniques; to take into account variable attenuation (if it is known); and to obtain meaningful images even if the attenuation is not known. Results of testing the proposed algorithms on the simulated and real data are presented.

The mechanical properties of bone are directly linked to its trabecular structure. A computed microtomography (CMT) system allowing high resolution three-dimensional imaging of bone samples is reported. The CMT technique benefits from the outstanding properties of synchrotron radiation (i.e. energy tunability, monochromaticity, high photon flux). The detector is based on a two-dimensional digital CCD camera coupled to a fluorescent screen through light optics. A spatial resolution of 9 micrometers (55 line pairs/mm) at 10% contrast was obtained. CMT, which is a non-destructive imaging technique, seems promising for the investigation of bone trabecular structure.

To investigate the tradeoffs that occur in helical CT scanning due to the different combinations of pitch and collimation available. These tradeoffs affect patient dose, image noise and slice sensitivity profiles. For both a single and dual detector helical CT scanners, different combinations of pitch and collimation were investigated. For each combination, patient dose was estimated using calculations from TLD measurements made with a standard CTDI phantom. Noise was measured using a homogeneous water phantom. Slice sensitivity profiles were measured using phantoms approximating impulse functions. For each scanner, the effects on noise, slice thickness and radiation dose of variations in pitch and collimation were reported. In addition, the tradeoffs between combinations of pitch and collimation that utilize the same table speed were examined. As an example, the single helical scanner can obtain a table speed of 10 mm/s using either: (1) a collimation of 10 mm and a pitch of 1, or by (2) collimating to 5 mm and using a pitch of 2. Assuming a constant mA, the radiation dose decreases from 1.2 rad at the center to .71 rad at the center, when going from the first to the second technique. The image noise increases from a standard deviation of 4.1 to 5.7; and the slice sensitivity profile decreases from 10.46 mm to 6.55 mm. Results for other pitch and collimation combinations are reported as well as the results for the dual helical scanner. For a fixed table movement speed, thinner slices with higher pitches give better z-axis resolution and lower patient dose while increasing image noise.

The purpose of this work is to investigate the potential image quality of scintillator/CCD-based direct digital medical x-ray imaging systems. The x-ray detector is composed of a scintillating screen to convert x-ray photons into lower energy radiation (UV/visible/NIR) that is then collected by a lens or a fiberoptic taper and converted to an electrical signal by a CCD. The DQE (on-axis geometry) was modeled by extending the analysis used for storage phosphor systems. The effect of various system parameters on the system DQE has been investigated. Two coupling approaches, an array of lenses, and an array of fiberoptic tapers, have been studied. For each coupling approach, two applications, chest radiography and mammography, were examined and the DQE was modeled to be comparable to screen/film systems.

In this contribution we propose an alternative x-ray detector based upon multiple screen-CCD sensor combinations. The impinging x-ray quanta are detected by a scintillator screen (e.g. CsI) and converted to light photons (typ. 1200 photons per absorbed x-ray quantum). We propose a number of lens-CCD sensors for standard video performance to detect the light photons coming out of an x-ray intensifying screen. Due to the smaller demagnification the coupling efficiency is better even with moderate quality (F number) lenses. We thus obtain a matrix of subimages, the system is constructed such that the subimages partially overlap. With digital image processing we construct from the subimages a single high quality image. Special hardware (incl. ASICS) has been developed for imaging at video rates, enabling (almost) fluoroscopy with this new detector. We show a viable digital x-ray imaging detector concept by means of our 2 by 2 CCD camera prototype and real-time processing engine. The image quality, MTF and noise properties are satisfactory and well in the diagnostic application range.

A kinestatic charge detector (KCD) with fourteen segmented signal collectors was built to evaluate the imaging performance of the single-scan dual energy x-ray KCD system. Those front segments of the KCD collectors will produce a low energy signal and those rear segments will produce a high energy signal. The middle segments between the front and rear work as an inactive filter to increase the separation of mean energies absorbed in the front and rear segments. In this study, the segmented KCD output signals were measured and compared with the Monte Carlo simulated result. An optimal KCD collector configuration is gotten by mapping the low energy and high energy signal amplitude to the predetermined thickness of the lucite-aluminum wedge-phantom (0-228.6 mm lucite and 0-25.4 mm aluminum), and then performing error analysis between the calculated thickness and known thickness.

The objective of this effort is to improve x-ray absorption and light production while maintaining high spatial resolution in x-ray imaging phosphor screens. Our current target is to improve screen absorption efficiency and screen brightness by factors of 2 or greater over existing screens that have 10-1p/mm resolution. In this program, commercial phosphor screens are combined with highly absorbing, high-resolution scintillating fiber-optic (SFO) face plates to provide a hybrid sensor that exhibits superior spatial resolution, x-ray absorption, and brightness values over the phosphor material alone. These characteristics of hybrid scintillators can be adjusted to meet specific x-ray imaging requirements over a wide range of x-ray energy. This paper discusses the design, fabrication, and testing of a new series of hybrid scintillators.

Flat-panel imaging arrays fabricated from hydrogenated amorphous silicon (a-Si:H) are being developed for application in diagnostic x-ray imaging. The potential performance of such imagers is analyzed by means of modeling the imaging system as a cascade of distinct gain and spreading stages. The relative number of image quanta at each stage in the system is represented in a quantum accounting diagram (QAD), and the detective quantum efficiency (DQE) of the imaging system under radiographic, fluoroscopic, and mammographic conditions is calculated. Such an analysis provides a means of quantitatively evaluating and optimizing imager performance for a given application.

This paper describes a new prototype a-Si (amorphous silicon) x-ray image sensor developed at EG&G Reticon. The sensor consists of a 512 by 512 array of a-Si photodiodes placed on 100 micrometer centers. The active area measures 5 cm on a side. Each pixel is addressed by a pair of switching diodes rather than the more common TFT (thin film transistor) switch. This approach leads to a simplified all-diode design with excellent fill factor, large dynamic range and good performance. The sensor is intended primarily for high-resolution x-ray imaging applications. A scintillator in direct contact with the diode array is used to convert incident x-ray photons to visible light detectable by the a-Si photodiodes. Since the conversion takes place directly at the sensor surface, no bulky intermediate optics are needed. The entire sensor plus support electronics can be mounted in a package less than 3 cm thick. The paper describes the sensor performance in terms of dark current, image lag, sensitivity and dynamic range. The MTF of the sensor and attached scintillator is measured by exposing the device through a narrow slit illuminated by a 50 kV x-ray source. Good contrast and sensitivity are achieved even at the Nyquist limit of the sensor resolution. Several images demonstrating the resolution and sensitivity of the sensor are presented.

The design and the performance of a 20 cm by 20 cm flat panel x-ray detector for digital radiography and fluoroscopy is described: Thin film amorphous silicon (aSi) technology has been used to build a 1024 by 1024 photodetector matrix, each pixel including both a photodiode and a switching diode; the pixel size is 196 by 196 micrometers². A high resolution and high absorption CsI(Tl) scintillator layer covers the top of the photodetector matrix in order to provide for x ray to light conversion. For low electronic noise and 30 fr/s operating rate we developed a custom design charge readout integrated circuit. The detector delivers a 12 bit digital output. The image quality, signal to noise ratio, and DQE are presented and discussed. The flat panel detector provides a MTF in excess of 30% at 2 lp/mm and a high contrast ratio without any distortion on the whole imaging area. The x-ray absorption is 70% for 50 KeV photons. The readout amplifier is optimized to reduce the electronic noise down to 1000 e-. This low noise level, combined with high sensitivity (1150 e-/incident x-ray quantum) provides the capability for fluoroscopic applications. The digital flat panel detector has been integrated in a C-arm system for cardiology and has been used on a regular basis in a European hospital since February 1995. The results are discussed for several operating modes: radiography and fluoroscopy. Conclusions on present detector performances, as well as further improvements, are presented.

The imaging performance of a new direct digital radiographic detector based on amorphous selenium and amorphous silicon TFT array which is under development is discussed. Progress has been made on the development of a multilayer digital x-ray detector panel with a structure consisting of a thin-film transistor pixel array, selenium x-ray photoconductor, dielectric layer and top electrode. An electronic system allows the rapid readout of image data which produces high resolution and wide dynamic range images. Using a straight edge, small wires and low contrast small holes targets, we have studied the spatial resolution, contrast detectability, and dynamic range of this new detector. Digital signals obtained from each pixel of this detector are almost linear with the total x-ray energy absorbed within the pixel area over a wide range of x-ray exposures. The resultant wide dynamic range allows extended latitude of exposure conditions and the enhancement or emphasis of different gray level regions from a single set of image data. For example, from one single exposure of the head, the soft tissue of the nose, detail structure of the teeth, as well as the bone structure of the neck can be examined by displaying and emphasizing selective gray levels of the image data. Image information obtained from this detector appears to be more evenly distributed over a wide dynamic range which is different from digital data obtained from other digital modalities such as the electrometer sensing of discharged potentials on photoconductors or from film digitization. Examples of images are shown. The discrete pixel structure of this detector and the higher intrinsic spatial resolution of selenium combine to produce image sharpness greater than those produced from digital detectors of similar pixel pitch using indirect conversion method or from digitizing film-screen images. The applicability of mathematical tools, such as the MTF which was developed primarily for analog images on a continuous imaging medium, is discussed with respect to our new discrete element detector.

Our goal is to develop a large area, flat panel, solid-state detector for both digital radiography and fluoroscopy. The proposed detector employs a photoconductive layer of amorphous selenium (a-Se) to convert x rays into charge. The charge image formed on the a-Se is electronically read out in situ using a two dimensional array of thin film transistors (TFTs), or 'active matrix.' Since the active matrix readout is capable of producing x-ray images in real- time, it can potentially be applied in both radiography and fluoroscopy. Investigation of the practical feasibility of this concept with a prototype x-ray imaging detector is described. Measurement of x-ray imaging properties, e.g. modulation transfer function (MTF) and x-ray sensitivity of the prototype are investigated, and the results compared with our earlier theoretical analysis.

Many different types of detectors are used for dual energy x-ray radiography. We developed an objective, technology independent, experimental method to compare the performance of these detectors. Our method uses a quality factor defined as the signal to noise ratio squared divided by the dose. We measure this quality factor by making images of a calibration phantom containing steps with known aluminum and plastic thickness. From the energy selective data, we compute an aluminum equivalent thickness image. The signal is the difference in the average value between two steps of the phantom with different aluminum thickness. The noise is the standard deviation of the computed values over the steps. The dose is the incident x-ray exposure. We applied this method to compare three energy selective detectors: (1) a single exposure, double screen, 'sandwich' storage phosphor detector, (2) an active storage phosphor detector using two exposures with different x-ray tube voltages with a light pulse erasing the front screen signal between x-ray exposures, and (3) a single exposure, double screen film detector. We found that the quality factor of the active detector is much larger than the single exposure film detector which in turn is substantially larger than the single exposure, storage phosphor detector.

Kinestatic charge detector (KCD) systems have potential uses in all fields of digital radiography including chest, abdominal, vascular, peripheral, dual energy subtraction and mammography. The purpose of this study was to investigate the development of a chest imaging protocol for a full field of view KCD system with regard to the effect of significant variations in patient chest thickness and its possible influence on kinestatic electric field settings. This was investigated by optimizing the kinestatic setting for a 5 lp/mm bar pattern imaged with 0, 5, 10, 15, and 20 cm of added Lucite to simulate varying patient chest thickness. Preliminary results demonstrate that variation in kinestatic settings due to variations in patient thickness affects the kinestatic electric field setting. Since only one field setting is available per image scan it becomes necessary to either optimize kinestatic settings for an estimated average patient thickness or for the density function of the expected region of interest (ROI).

Inappropriate or poorly designed x-ray imaging systems can lead to inadequate image quality and/or excessive patient absorbed dose. To study this dilemma, we have developed a computer program, based on Monte Carlo methods, to model the imaging chain and to optimize the design of individual imaging system components. The Monte Carlo method is used to simulate the transport of x-ray photons through the patient and anti-scatter device and into the image receptor. Image quality is calculated in terms of contrast and signal-to-noise ratio (SNR) and patient risk in terms of mean absorbed dose. The model of the radiographic screen includes the statistics of x-ray to light conversion and diffusion of light photons and the spatial distribution of absorbed x rays. The optimal system was defined as that which, for a constant measure of image quality, results in the lowest mean absorbed dose in the patient. The results of the model agree with measured data. Examples are given for different measures of reducing the absorbed dose for constant SNR in a simulated lumbar spine examination. At the optimal tube potential, the dose reductions associated with using a thicker screen, a fiber-interspaced grid and an additional Cu-filtration are 53%, 42%, and 31%, respectively.

Diagnostic medical imaging systems consist of multiple subsystems. Between the image acquisition and diagnostic image display there can be more than six subsystems. These systems can utilize both analog and digital processing, and can be from several different manufacturers. Procedures have been developed to monitor total imaging systems with multiple input devices and multiple vendors. These procedures utilize 'standard' test images and apply objective measurement methods with the aid of a desktop computer. The devices being monitored include film digitizers, video frame grabbers, clinical display stations, film laser printers, vinyl paper printers, CT scanners, MR scanners, and ultrasound scanners. Vendor calibration test patterns provided internally on different devices were evaluated and found to be necessary for image quality monitoring. Vendor provided image analysis software tools, transmission and reflective densitometers, and photometers are utilized to evaluate the image quality. A commercially available image program is used to measure the bit values of the resultant image files. A comprehensive yet simple test procedure has been developed. A 'standard' phantom scanned by a CT scanner and a vendor-provided SMPTE pattern is used to evaluate the total system. The accuracy of data acquisition, image display and printing are all verified with the test images. The SMPTE pattern is used to evaluate the frame grabber A/D conversion and laser printer/processor performance. A SMPTE film printed on a properly calibrated laser printer is used to monitor the performance of film digitizers and display stations. These evaluations are performed visually in the clinical setting and offline with a desktop computer.

We have obtained the density or x-ray tube voltage dependence of the Wiener spectra of the three factors of radiographic mottle, i.e., quantum mottle, structure mottle, and film granularity, for the double emulsions of the x-ray film. For the purpose, we proposed a method for obtaining the Wiener spectrum of each of the three factors for the double emulsions. To investigate the relation between the Wiener spectra of the quantum mottle for the double and single emulsions, we compared the Wiener spectra of the quantum mottle for the double emulsions with the sum of those for the front and back emulsions on the same radiographs. The Wiener spectral values of the quantum mottle for the double emulsions at 0 mm^-1 were greater than the sums of the Wiener spectral values of the quantum mottle for the front and back emulsions at tube voltages of 50 to 100 kV. We found that this was the result which originated from the following phenomena: (1) the noise-equivalent numbers of quanta (NEQ) of the dual screens were roughly equal to the sum of the NEQ of the front and back screens; (2) the squares of the gradients for the double emulsions were more than four times on an average as great as those for the front and back emulsions.

Because of beam hardening in the target, the energy distribution of the x-radiation emitted by a conventional x-ray tube differs from different directions of emission. This fact is generally neglected in dual-energy subtraction imaging. In this paper we study the influence of the non- uniformity of the energy spectrum of emitted radiation on dual-energy subtraction. Experiments with synthetic (aluminum) phantoms and with dual-energy images of humanoid chests (made with a computed radiography system) give more insight in the degree to which dual-energy subtraction is affected by the described effects.