Blindness is nature is fatal. In biology and physiology one finds many situations where nature has obtained neat solutions to problems, solutions that ar every nearly the best possible. Many of the design parameters for the eye are not arbitrarily selected, but are constrained to a narrow range of values by physics and information theory considerations. As Helmholtz mentioned more than a century ago 'the eye has every possible defect that can be found in an optical instrument and even some which are peculiar to itself; but they are all so interacted, that the inexactness of the image which results from their presence very little exceeds, under ordinary conditions of illumination, the limits which are set to the delicacy of sensation by the dimensions of the retinal cones.' Helmholtz was particularly prescient in his reference to cone dimension because, as we will see, many eye properties are completely determined once cone diameter is selected. The ideas presented in this paper are based on the working assumption that the eye does the best possible job within physical limits. This idea originated with Horace Barlow more than 40 years ago. Once excellent reference is the proceedings of a conference organized to honor Barlow's retirement with presentations by his many collaborators over the years. The list includes practically everyone referenced in this paper, which explores the design and optimization of the optics of the eye, retinal transduction and coding of visual data.

Model observers have been compared to human performance detecting low contrast signals in a variety of computer generated background including white noise, correlated noise, lumpy backgrounds, and two component noise. The purpose of the present paper is to extend this work by comparing a cumber of previously proposed model observers to human visual detection performance in real anatomic backgrounds. Human and model observer performance are compared as a function of increasing added white noise. Our results show that three of the four models are good predictors of human performance.

This paper addresses the question of how to determine the performance of the optimum linear or Hotelling observer when only sample images are available. This observer is specified by a template from which a scalar test statistic is computed for each image. It is argued that estimation of the Hotelling template is analogous to problems in image reconstruction , where many difficulties can be avoided through judicious use of prior information. In the present problem, prior information is enforced by choice of the representation used for the template. We consider specifically a representation based on Laguerre-Gauss functions, and we discuss ways of estimating the coefficients in this expansion from sample images for the problem of detection of a known signal. The method is illustrated by two experiments, one based on simulated nonuniform fields called lumpy backgrounds, the other on real coronary angiograms.

Anatomical noise in chest radiography, created by the projection of anatomical features in the thorax such as ribs and pulmonary vessels, greatly influences the detection of subtle lung nodules in chest radiographs. Detection may be hindered by 1) the 'global' statistical characteristics of the background in relation to the signal associated withthe nodule, and/or 2) the interference of the 'local' background pattern with the nodule signal. This investigation aimed at assessing the influence of the latter process in the detection of subtle lung nodules. Six 8 X 8 cm images were extracted from the lung regions o six digital chest radiographs of normal patients from our clinic. Simulated nodules emulating the radiographic characteristics of subtle tissue-equivalent lesions ranging in size from 3.2 to 6.4 mm were numerically superimposed on the images. For each of the six lung images, a set of thirty-one processed images were produced, six containing no nodule, and the remaining 25 containing single nodules of five different sizes placed at five different locations within 6 mm of the center. The variation in location allowed different local background patterns to overlay the nodules. An observer detection study was then performed using 14 experienced radiologists. The observer data were analyzed to determine the variation in detectability with nodule location for all five sizes of the nodules. The preliminary results indicate that the variation in detectability of a nodule due to the influence of its local background surroundings is equivalent to that caused by changing its CD product by a factor of 4.45.

Previous experiments using highpass nose have either suggested that humans cannot compensate for anti-correlated noise in images or were inconclusive. These results may have been misleading because of the use of a single noise component. For large exponents of fⁿ, image noise within the bandwidth of the signal amplitude for detection. This situation does not correspond to CT or SPECT imaging cases where patient structure with a lowpass spectrum is also present and limits detection accuracy. In addition, humans have two forms of internal noise that limit detection and this may have been the source of poor human performance. So, in this work, experiments were done with two noise components - one broadband to ensure that task performance was always limited by external noise. The experiments were designed to be more precise test of compensation for anti- correlated noise and to provide a more sensitive test of existing observer models. In all cases, separate experiments were done to estimate observer internal noise. The new results show a marked asymmetry between lowpass and highpass noise effects and are consistent withthe view that internal noise is the cause of poor highpass noise performance.

It is well known that the optimum way to perform a signal- detection or discrimination task is to compute the likelihood ratio and compare it to a threshold. Varying the threshold generates the receiver operating characteristic (ROC) curve, and the area under this curve (AUC) is a common figure of merit for task performance. AUC can be converted to a signal-to-noise ratio, often known as d_a, using a well-known formula involving an error function. The ROC curve can also be determined by psychophysical studies for humans performing the same task, and again figures of merit such as AUC and d_z can be derived. Since the likelihood ratio is optimal, however, the d_a values for the human must necessarily be less than those for the ideal observer, and the square of the ratio of d_a (human)/d_a(ideal) is frequently taken as a measure of the perceptual efficiency of the human. The applicability of this efficiency measure is limited, however, since there are very few problems for which we can actually compute d_a or AUC for the ideal observer. In this paper we examine some basic mathematical properties of the likelihood ratio and its logarithm. We demonstrate that there are strong constraints on the form of the probability density functions for these test statistics. In fact, if one knows, say, the density on the logarithm of the likelihood ratio under the null hypothesis, the densities of both the likelihood and the log-likelihood under both hypotheses are specified in terms of a likelihood-generating function. From this single function one can obtain all moments of both the likelihood and the log-likelihood under both hypotheses. Moreover, a AUC is expressed to an excellent approximation by a single point on the function. We illustrate these mathematical properties by considering the problem of signal detection with uncertain signal location.

Mixture distribution analysis (MDA) is proposed as a statistical methodology for comparing observer readings on different imaging modalities when the image findings cannot be independently verified. The study utilized a data set consisting of independent, blinded readings by 4 radiologists of a stratified sample of 95 bedside chest images obtained using computed radiography. Each case was rad on hard and soft copy. The area under the ROC curve (AUC) was calculated using ROCFIT and the relative percent correct (RPC) was calculated from point distributions estimated by the MDA. The expectation maximization algorithm was used to perform a maximum likelihood estimation of the fit to either 3, 4 or 5 point distributions. There was agreement between the AUC and the RPC based upon 3 point distributions representing easy normals, hard normals and abnormals, easy abnormals, hard normals, hard abnormals and easy abnormals. We conclude that the MDA may be a viable alternative to the ROC for evaluating observer performance on imaging modalities in clinical settings where image verification is either difficult or impossible.

The goal of this study was to determine what the influence of image processing functions was on decisions and decision changes made while reading chest radiographs displayed on a monitor. Six radiologists read 168 computed radiography chest images first without them with the use of six image processing functions. Diagnostic performance was measured using receiver operating characteristic analysis, and decision changes made without and with processing use were analyzed. Diagnostic performance did not differ statistically for readings without and with image processing. The decision change analysis showed that readers were just as likely to change decisions from true-positive to false-negative as they were from false-negative to true- positive. With image processing, there were significantly more changes from true-negative to false-positive than from false-positive to true-negative. 93 percent of all decisions did not change with the use of image processing. No significant correlations were found between the type of lesions present on the radiography and the type of image processing function sued. Positive decision changes made with the use of image processing are offset by equivalent numbers of negative decision changes. The use of image processing does not affect significantly diagnostic performance in chest radiography.

The goal of this study was to determine the influence of adjusting monitor display parameters such as brightness and tone scale on observer diagnostic performance in order to determine what display settings are best for radiographic soft copy image display. Six radiologists viewed a series of 50 pairs of mammograms on CRT monitors in two conditions. The images contained either a single subtle mass, a single subtle cluster of microcalcifications, or no lesion. In the first condition, the characteristic curve of the monitor was studied. Two curves were tested, one which was related to human perception and one which was typical for an unmodified CRT. In the second condition, display luminance was manipulated. Readers had to report on the presence of microcalcification clusters or masses and had to report their decision confidence using a 6- level rating scale. Monitor brightness and tone scale can influence diagnostic performance to some degree. Whether the magnitude of these differences is important clinically needs to be further studied.

The display of a 12-bit MR image on a common 8-bit computer monitor is usually achieved by linearly mapping the image values through a display window, which is determined by the width and center values. The adjustment of the display window for a variety of MR images involves considerable user interaction. In this paper, we present an advanced algorithm with the hierarchical neural network structure for robust and automatic adjustment of display window width and center for a wide range of MR images. This algorithm consists of a feature generator utilizing both histogram and spatial information computed from a MR image, a wavelet transform for compressing the feature vector, a competitive layer neural network for clustering MR images into different subclasses, a bi-modal linear estimator and an RBF (radial basis function) network based estimator for each subclass, as well as a data fusion process to integrate estimates from both estimators of different subclasses to compute the final display parameters. Both estimators can adapt a new types of MR images simply by training them with those images, thus making the algorithm adaptive and extendable. This trainability makes also possible for advanced future developments such as adaptation of the display parameters to user's personal preference. While the RBF neural network based estimators perform very well for images similar to those in the training data set, the bi-modal linear estimators provide reasonable estimation for a wide range of images that may not be included in the training data set. The data fusion step makes the final estimation of the display parameters accurate for trained images and robust for the unknown images. The algorithm has been tested on a wide range of MR images and shown satisfactory results. Although the proposed algorithm is very comprehensive, its execution time is kept within a reasonable range.

Digital mosaic imaging techniques provide a cost effective means to acquiring high resolution images. Constrained mosaic imaging techniques make use of special purpose fiducial patterns in order to define a priori the relation between images on each tile. This 'inter-tile' relation is applied to any images acquired subsequently. A simulation study was carried out where a model of the digital mosaic imager was used. By doing so, it was possible to compare the original data to that reconstructed using different techniques. The effects of these technique so the quality of the final digital mosaic image wee investigated. The techniques were applied towards reconstructing mammogram images. In order to evaluate performance of the approach, a set of features of interest were selected to measure image quality. Features that are important to visual perception include micro-calcifications and other fine details on the image, as per a radiologist's suggestion. Features important to the computerized diagnostic software include, edge maps and other common features used in existing computerized mammogram analysis approaches. Results of this experimental study provide a better understanding of how mosaic reconstruction approaches affect the quality of the final image. The study is also helpful in defining the role that features of interest, be it from a visual perception or computer software point of view, play towards selecting the image reconstruction scheme better suited for digital mosaic mammography.

Many receiver operating characteristic (ROC) studies rely on establishing 'truth' about lesion absence/presence on the agreement of a panel of experts. In addition, in the consensus committee methodology, images where the members of the committee did not reach any agreement about the lesion absence/presence are discarded from the ROC study. But how reliable are 'gold standards' established by these expert committees. And does discarding images where no agreement was reached bias the spectrum of difficulty of the test image set for the ROC study. Computer simulated lesions of different strengths were embedded in real x-ray coronary angiogram background in order to measure the agreement among the decisions of members of the committee as a function of signal strength, to establish the accuracy of the decisions of the consensus expert committee and to compare it to individual more inexperienced readers.

In physicians' interpretation, morphologic characteristics of pulmonary nodules are not only important signs for the discrimination, but also important features for the diagnosis with a reasonable degree of confidence. This paper describes about the computerized interpretation system which is developed to analyze the relation between the measuring values and the morphologic characteristics, and to make clear the logic of physicians' diagnosis. We think that the four basic morphologic characteristics of the discriminative diagnosis between benign and malignant nodules exist which are: (1) the density; (2) the homogeneity; (3) the definition; and (4) the convergence. To obtain each grade of the parameters, we developed an interpretation system. On the other hand, to obtain digital feature values, we used our computer aided diagnosis system. Interpretation experiments were performed by using 15 benign and 19 malignant cases of chest x-ray CT images. As the result of a statistical analysis, some digital features have the significant differences between benign and malignant nodules, and the morphological characteristics have also differences. Therefore the computerized system is feasible to help physicians' interpretation to distinct between malignant and benign nodules by showing digital feature values as some references.

Image quality assessment in medical imaging requires realistic textured background that can be statistically characterized for the computation of model observers' performance. We present a modeling framework for the synthesis of texture as well as a statistical analysis of both sample and synthesized textures. The model employs a two-component image-decomposition consisting of a slowly, spatially varying mean-background and a residual texture image. Each component is synthesized independently. The technique is demonstrated using radiological breast tissue. For statistical characterization, we compute the two-point probability density functions for the real and synthesized breast tissue textures in order to provide a complete characterization and comparison of their second-order statistics. Similar computations for other textures yield further insight into the statistical properties of these types of random fields.