CT numbers of the spleen, liver, and trachea air were measured from non-contrast images obtained from 4-channel and
64-channel scanners from the same vendor. Image sections of 1 mm and 5 mm were reconstructed using smooth and
sharp kernels. For spleen and liver, no significant differences associated with the variations in kernels or slice thickness
could be demonstrated. The increase of the number of channels from 4 to 64 lowered the spleen CT numbers from 53
HU to 43 HU (p <0.00001). The 4-channel spleen CT numbers slightly increased as function of patient size, while the
64-channel CT numbers decreased as function of patient size. Linear regressions predicted for 40-cm patients the spleen
64-channel CT values were 23 HU lower than 4-channel CT numbers. The smooth kernel, 4-channel trachea air CT
numbers had mean of -1004 +/-4.8 HU and the 64-channel trachea air CT numbers had a mean of -989+/-4.5 HU. The
patient-size dependencies suggest that the CT attenuation variation is associated with increased scatter in 64-channel
MSCT. Using CT number to distinguish solid lesions from cysts or quantitative evaluation of COPD disease using CT
images may be complicated by inconsistencies between CT scanners.
The high-frequency of suspicious non-malignant image features limit the use of CT lung-cancer screening of
asymptomatic individuals. A reference database of 6 radiologists' localizations of suspicious image features was created.
The frequency, sizes, shapes, margins and degree of calcification of the images features were determined. The
radiologist exam report identified 50% findings reported on 5 or 6 occasions, while the CAD system identified 40%. The
radiologist exam report missed 80% to 50% of suspicious, retrospectively identified in CT lung cancer images. Many
were less than 4 mm. Radiologists can use a lenient criterion in experiments of previously read screening cases.
If the locations of abnormalities (targets) in an image are unknown, the evaluation of human observers' detection
performance can be complex. Richard Swensson in 1996 developed a model that unified the various analysis
approaches to this problem. For the LROC experiment, the model assumed that a false-positive report-arises from the
latent decision variable of the most suspicious non-target location of the target stimuli. The localization scoring was
based on the same latent decision variable, i.e., when the latent decision variable at the non-target location was greater
than latent decision variable at the target location the response was scored as a miss. Human observer reports vary, i.e.,
different locations have been identified during replications. A Monte Carlo model was developed to investigate this
variation and identified a non-intuitive aspect of Swensson's LROC model. When the number of potentially suspicious
locations was 1, the model performance was greater than apparently possible. For example, assume that target expected
latent decision variable is 1.0. Both target and non-target standard deviations were assumed to be 1.0. The model
predicts the area-under-the-ROC is 0.815, which implies da=1.27. If the target latent decision variable was 0.0, then
da=0.61. The reason was the number latent decision variables in the model for the non-target stimuli is one, while the
number latent decision variables for the target stimuli is the maximum of 2. The simulation indicated that the
parameters of a LROC fit, when the number of suspicious locations is small or the observer performance is low, does
not have the same intuitive meaning as ROC parameters of a SKE task.
The aim of this study was to determine whether radiologists are more likely to report as calcified the small nodules detected during CT lung-cancer screening, if sharper reconstruction filters are utilized. Images were reconstructed with the 2 filters used at our institution for the lung (B50f) and for the mediastinum (B30f). The 4 lung-cancer screening cases were reconstructed with 1.25-mm section thickness at 0.6-mm section increments. Using a lax criterion, 2 radiologists identified the locations of nodular features and rated the likelihood that the features were calcified. There were 302 nodules reports. More of these (57%) were reported on images reconstructed using the smooth filter. Sixty (60) reports were definitely or possibly calcified. Seventy-three percent (73%) calcification reports were from images reconstructed using B50f. There were 27 calcification reports of one of the radiologist that were classified as non-calcified by the other radiologist. Most of calcification reports (81%) of 27 reports on which radiologists disagree regarding the likelihood of calcification were from images reconstructed using B50f. Radiologists are more likely to report small nodules detected during lung-cancer screening as calcified using the sharper reconstruction filter. Whether these nodules are actually calcified or not remains a question.
Normal mammographic image backgrounds have approximately isotropic power spectra of the form, P(f) =K/fe, where f is radial frequency. The values ofthe exponent, 3, range from 1.5 to 3.5 with an average of about 2.8. The ideal observer model predicts that, for signals with certain properties, the log-log contrast-detail (CD) diagram slope, m, is given by: m = O.5(3-2). Previously, we reported results for detection of a model mass (designer nodule) in filtered noise with an exponent of 3. The model and human observer CD slopes were 0.5 and 0.45 respectively. Here, we report preliminary results for human and model observer 2AFC detection of a simple signal in filtered noise with exponents from 1.5 to 3.5. Our results are in good agreement with the prediction of the above equation. We will also describe results of 2AFC detection experiments done using "twin" noise backgrounds with identical noise realizations in the two backgrounds. We could not replicate the results ofJohnson et al. For '1/f3' noise, they found a CD slope of—O.59 while we found +0.37.
Kundel et al. Suggested the use of circle cues to assist human observers during signal-known-exactly (SKE) detection experiments. The circles were bipolar (with concentric black and white rings) and centered on potential locations of simulated masses added to mammographic backgrounds. They used a large circle cue (diameter 6.4 cm) and a background size of 7.7 cm (referred to the initial mammogram). They found significant detection performance improvement compared to the no cue conditions. In our previous experiments, we use mammographic background sizes of 6.1 cm and smaller circles with sizes dependent on lesion size. Our circle sizes were selected to subjectively optimize utility but choices may not have been the best. Also, detectability may also depend on background size. In this work, we present human observer results for detecting a realist mass added to mammographic backgrounds using 30 conditions (all combinations of the mass scaled to 3 sizes, 2 background sizes and 5 circle sizes). Performance did not depend on background size. For the smallest mass size (1 mm, 8 pixels), detectability decreased as circle size increased. There may be an optimum near a circle/mass size ratio of 4. The optimum size ratio for the 4 mm mass was 3. For the 16 mm mass, detectability decreased as steadily as circle size increased. The smallest size ratio used was 1.2.
Segmentation of chest CT images has several purposes. In lung-cancer screening programs, for nodules below 5mm, growth measured from sequential CT scans is the primary indication of malignancy. Automatic segmentation procedures have been used as a means to insure a reliable measurement of lung nodule size. A lung nodule phantom was developed to evaluate the validity and reliability of size measurements using CT images. Thirty acrylic spheres and cubes (2-8 mm) were placed in a 15cm diameter disk of uniform-material that simulated the lung. To demonstrate the use of the phantom, it was scanned using out hospital's lung-cancer screening protocol. A simple, yet objective threshold technique was used to segment all of the images in which the objects were visible. All the pixels above a common threshold (the mean of the lung material and the acrylic CT numbers) were considered within the nodule. The relative bias did not depend on the shape of the objects and ranged from -18% for the 2 mm objects to -2.5% for 8-mm objects. DICOM image files of the phantom are available for investigators with an interest in using the images to evaluate and compare segmentation procedures.
Detection of mass lesions in mammograms is essentially limited by image variation due to normal patient structure, which has an average power-spectrum of the form `1/f3'. Image noise plays little role in limiting mass detection. The contrast-detail (CD) diagram for lesion detection in mammographic structure is novel, for both human and model observers. Contrast thresholds increase with increasing signal size for signals larger than about 1 mm, with CD slopes of about 0.3 for humans and 0.4 for model observers. Similar results were obtained in search experiments. The work was done using hybrid images, with of tumor masses (extracted from specimen radiographs) added to digitized mammographic backgrounds. We have been able to explain the results using a number of observer models. These results demonstrate that CD diagrams based on image noise-limited detection with simple phantoms are not useful for evaluation of mass detection in mammograms--so more realistic approaches are necessary in order to model mammographic imaging systems for optimization.
This paper is a report on very surprising results from recent work on detection of real lesions in digitized mammograms. The experiments were done using a novel experimental procedure with hybrid images. The lesions (signals) were real tumor masses extracted from breast tissue specimen radiographs. In the detection experiments, the tumors were added to digitized normal mammographic backgrounds. The results of this new work have been both novel and very surprising. Contrast thresholds increased with increasing lesion size for lesions larger than approximately 1 mm in diameter. Earlier work with white noise, radiographic image noise, computed tomography (CT) noise and some types of patient structure have accustomed us to a particular relationship between lesion size and contrast for constant detectability. All previous contrast/detail (CD) diagrams have been similar, the contrast threshold decreases as lesion size increases and flattens at large lesion sizes. The CD diagram for lesion detection in mammographic structure is completely different. It will be shown that this is a consequence of the power-law dependence of the projected breast tissue structure spectral density on spatial frequency. Mammographic tissue structure power spectra have the form P(f) equals B/f(beta ), with an average exponent of approximately 3 (range from 2 to 4), and are approximately isotropic (small angular dependence). Results for two-alternative forced-choice (2AFC) signal detection experiments using 4 tumor lesions and one mathematically generated signal will be presented. These results are for an unbiased selection of mammographic backgrounds. It is possible that an additional understanding of the effects of breast structure on lesion detectability can be obtained by investigating detectability in various classes of mammographic backgrounds. This will be the subject of future research.
We make a preliminary study with a ROC experiment to determine the acceptable levels of image compression that may be utilized for PACS archives. CR images of 1760 X 2140 pixel size and 10 bit depth are studied. The experiment uses wavelet algorithm for image compression and printed films for image viewing. The 'internal standard' experiment results in an acceptable value of compression ratio of 6 for imperceptible difference (da' equals 1) between the compressed and the uncompressed image. Such ratio would lead a storage reduction factor of 9.6 for these images. The information capacity of the CR images may be extrapolated to be 40 bits per millimeter of viewing area.
Women with mammograms that radiologists classify as dense have been found to have an increased risk of breast cancer. The purpose to this investigation was to determine whether human readers are willing and able to make reliable comparisons of five attributes of pairs of mammograms matched by a quantitative estimate of the fraction of dense tissue (FDT). Forty pairs of CC projections were digitized and presented using a computer workstation. The 40 pairs of mammograms had the same FDT as measured by a visual threshold procedure. Each breast image was from a different woman. The difference in the following 5 attributes were rated: (1) fraction of dense tissue, (2) fraction of homogeneous of the dense tissue, (3) fraction of ductal dense tissue, (4) prominence of scalloping of dense tissue, and (5) prominence of subareolar structures. The rating were replicated to evaluate their reliability. Spearman rank-order correlations of replicated measurements ranged from 0.89 to 0.65 (p was less than 0.0001). Homogeneous dense tissue ratings were negatively correlated with ductal dense tissue ratings (-0.59, p equals 0.0001). The prominence of scalloping rating was not significantly correlated with other attributes. The ratings of the attributes, except scalloping, were significantly correlated to differences mean gray level of breast parenchyma. Readers can make reliable judgments regarding the differences in attributes of mammograms that are matched by FDT. The negative correlation between the homogeneous dense and the ductal dense tissue ratings suggest that homogeneous dense and ductal dense tissues contend for perceived dense breast area. The absence of correlation between scalloping and other image attributes suggests further investigation of scalloping as an independent, breast-cancer risk factor is warranted.
Performance accuracy for detecting and localizing small nodules on liver CT images depends on whether an observer is required to find dark nodules or bright nodules on those images. We investigated these asymmetric polarity effects using simulated nodules of varying sizes placed on spiral CT scans of clinical patients acquired with intravenous contrast material, which made blood vessels appear brighter than liver background on the displayed CT images. A concurrent analysis of each observer's detection-rating and scored-localization data estimated separate perceptual effects for the nodules of different sizes, and for locations of the dark or bright hepatic findings that observers regarded as most suspicious on the CT images. The results were consistent with equal visibility for dark and bright nodules of identical size and CT-contrast, and a linear increase in visibility with nodule signal-to-noise ratio for a non-prewhitening matched-filter calculation (NPW-SNR). The substantial lower accuracy for detecting and localizing the bright nodules, compared to the dark nodules, was a polarity effect apparently produced by the non- stationary liver CT backgrounds -- i.e., the presence of stronger confusing signals from the bright hepatic findings on these (contrast-enhanced) CT images than from the dark hepatic findings.
The effect of target size and target-size uncertainty on human observers' ability to detect Gaussian and disk targets in spatially uncorrelated and correlated noise was measured. Disk and Gaussian targets were centered in circular areas (diameter, 128 pixels) of uncorrelated noise and uncorrelated noise filtered to resemble CT noise. Size uncertainty was introduced in the target stimuli by presenting targets with effective areas that ranging from 15 to 10,000 pixels. A constant non-prewhitening, matched- filter signal-to-noise ratio (NPW-SNR) was maintained for all target sizes within each trial by adjusting target contrast. Stimulus sets were rendered on the gray-scale monitor of the computer workstation used to collect observer responses. Observers rated for each stimulus the likelihood that a target was present. The observer ratings were analyzed using a multiple-distributing extension of the bi- normal ROC curve fitting procedure. A control experiment evaluated the influence of the circular noise area on performance with size-specified targets. Observer detection performance loss, the ratio of d' to NPW-SNR, decreased for small and large targets. Variations of target shape and noise correlation had no significant effect on performance loss. Observer performance when target was uncertain was the same as observer performance when the target size was specified. In the control trials the investigator specifies the target size to the observer, yet the observer cannot exploit that information. The observer apparently uses same perceptual resources in both the experimental and control trials to render a rating and consequently performs similarly in size-uncertain and size-specified trials. These results suggest a substantial role of higher order mechanisms in the detection of compact targets in noisy backgrounds.
The effect of displayed CT image size and observer viewing protocol on human observer ability to detect nodules was measured. Synthetic nodules (3.0 to 5.0 mm) were added to random locations within the lungs of 80 CT images from spiral CT scans of 13 patients. Each test set consisted of 160 images. Each CT image was presented twice in a trial, once with nodule present and once without nodule present. The images were rendered as film transparencies using 6 pixel sizes (0.074 to 0.259 mm). Four observers read the films using two viewing protocols. In one protocol, the observers could vary their viewing distance and were provided with a magnification lens. In the second protocol, the observers viewing distance was fixed at 55 cm. Observers rated the likelihood that a nodule was present in the image and indicated the lung most likely to contain a nodule. The ratings were used to estimate an ROC curve for each trial. Detectability was better using the variable distance viewing protocol compared to the fixed viewing distance protocol. The area under the ROC curve was constant as a function of pixel size for the variable viewing distance protocol (0.881 plus or minus 0.007) and decreased (0.894 plus or minus 0.013 to 0.668 plus or minus 0.053) as a function of pixel size for the fixed viewing distance protocol. Radiologists should be encouraged to vary their viewing distance when reading CT images rendered as films. In order to reduce costs, some radiology departments may be tempted to reduce the CT image size on the film and there by increase the number of CT images on each film. Our study suggests that this manipulation could impair the radiologist's ability to detect lung nodules on CT images of the chest.
An investigation was performed to evaluate objectively observer's ability to find lung nodules on compressed spiral computerized tomographic (CT) images of the chest. A set of 80 images from 13 patients served as backdrops. One simulated nodule of either 3.0, 3.4, 4.0, or 5.0 mm in diameter was inserted into each image. These 80 images were viewed on a computer screen in two formats: compressed with a wavelet transform coder at a compression rate of 40:1, and in the uncompressed 8 bit-per-pixel format, windowed down from the 12 bit-per- pixel originals. The images were presented one at a time in random order, as two conditions of 80 images each. Six observers searched for lung nodules on both the original and compressed formats. The tasks were to locate the nodule in each image, and, using a five category rating scale, to indicate the confidence that the indicated location contained a nodule. The results indicate that all observers detected a higher fraction of nodules in the original images than in the compressed images. Even though the compressed images were described by the observers as unacceptable for clinical use because they contained numerous artifacts, the percentage of 4 and 5 mm nodules found in the compressed images was high. Directions of further research include measurement of detection performance at lower compression rates, identification of compression artifacts that get confused with nodules, and analysis of the confidence ratings.
The effect of the lump amplitude of lumpy backgrounds on human observers' ability to find clusters of simulated calcifications was measured. Clusters of simulated calcifications were randomly located in 56-pixel-radius circular areas of lumpy backgrounds to which Gaussian noise was added. The clusters of simulated calcifications were projections of five 2.5 pixel- radius spheres randomly located less than 10 pixels of its center. The lumpy backgrounds were produced by adding to the circular areas at random locations, 50 Gaussian lumps of a standard deviation of 7 pixels. Stimulus sets were rendered on the gray-scale monitor of a computer workstation. Three observers searched each area for the cluster. They indicated the location most likely to contain a cluster with a mouse pointer and rated the likelihood that the indicated location actually contained a cluster. The Gaussian noise was adjusted so that the fraction of clusters found was approximately 0.5 for the three levels of amplitude of the lumpy backgrounds and the uniform backgrounds. The detectability, d', was calculated from the fraction of clusters found using an M-alternative forced choice model. A cluster was scored as found (a hit) if the center of the cluster was contained within a specified area surrounding the observer indicated location. The observer detection loss ratio, the ratio of d' to SNR-SKE, decreased from 0.6 for the uniform noise background to 0.25 for the lumpy backgrounds. Observers' ability to find clusters of simulated calcifications is significantly decreased by lumpy backgrounds.
The effect of target size and size uncertainty on human observer ability to see disk targets in uncorrelated noise was measured. disk targets were centered in 64-pixel-radius areas of uncorrelated Gaussian noise. Human observers rated the likelihood that a target was present Size uncertainty was introduced in the target-present stimuli by using disk targets with radii ranging 2 to 33.2 pixels. A constant matched-filter signal-to-noise ration was maintained across the range of disk sizes by adjusting the disk contrast. For this mixed size experiment the observer ratings were analyzed using a multiple-distribution extension of the binormal ROC curve fitting procedure. A control experiment measured observer performance in conditions with target-present stimuli of known-size disks. A third experiment evaluated the influence of noise-area size on performance with known-size disks. An observer detection efficiency index, the square of the ration of d' to SNR, decreased at small and large disk radii. The efficiency index decrease for small disks was less in the control experiment (size- known). Observer efficiency indexes for medium and large disks were not significantly difference for the mixed size experiment and the control experiment. Reducing the noise-area size increased the efficiency for small disks and produced an approximately constant efficiency for the small to medium sized disks. Size uncertainty decreased observer detection performance relative to known-size performance for small disk targets. the observer efficiency index for the small targets was increased when small noise areas are used. This finding suggests that the decreased efficiency index for small targets on large noise areas was caused by increased observer uncertainty of target location.