Radiology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online before print May 15, 2003, 10.1148/radiol.2273020509
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2273020509v1
228/1/76    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Samei, E.
Right arrow Articles by Eyler, W. R.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Samei, E.
Right arrow Articles by Eyler, W. R.
(Radiology 2003;228:76-84.)
© RSNA, 2003

Thoracic Imaging

Subtle Lung Nodules: Influence of Local Anatomic Variations on Detection1

Ehsan Samei, PhD, Michael J. Flynn, PhD, Edward Peterson, PhD and William R. Eyler, MD

1 From the Departments of Radiology, Physics, and Biomedical Engineering, Duke University Medical Center, DUMC 3302, Durham, NC 27710 (E.S.); and Departments of Radiology (E.S., M.J.F., W.R.E.) and Biostatistics and Research Epidemiology (E.P.), Henry Ford Health System, Detroit, Mich. Received May 15, 2002; revision requested July 12; revision received August 7; accepted September 26. Supported in part by a small project grant from the Henry Ford Health Sciences Center, Detroit, Mich, and by the NIH, R21 CA91806. Address correspondence to E.S. (e-mail: samei@duke.edu).


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 REFERENCES
 
PURPOSE: To assess the influence of local anatomic noise on the detection of subtle lung nodules depicted on chest radiographs.

MATERIALS AND METHODS: Six 8 x 8-cm lung regions were extracted from digital chest radiographs obtained in healthy subjects. Simulated nodules emulating the radiographic characteristics of subtle tissue-equivalent lesions 3.2–6.4 mm in diameter (equivalent to 0.1–0.4 mm in contrast-diameter product [CD]) were added to the images. On multiple renditions of each image, nodules were inserted at slightly different locations within 6 mm of the center; this process allowed different local background patterns to overlie the nodules. An observer detection study involving 15 experienced radiologists was performed. The authors performed analysis of variance and pairwise t test analyses to determine variations in nodule detectability related to nodule location and size on each image.

RESULTS: Results indicated a strong correlation between nodule size and observer detection score and significant variation in nodule detectability as a function of location. Changes in nodule position caused observer score variations that were equivalent to the variation caused by an up to 185% change in nodule CD (78% average over all six images), an up to 68% change in diameter (32% average), and an up to 28% change in area under the receiver operating characteristic curve (Az) (14% average).

CONCLUSION: Local anatomic variations surrounding and overlying a subtle lung nodule on a chest radiograph that are created by the projection of anatomic features in the thorax, such as ribs and pulmonary vessels, can greatly influence the detection of nodules, altering the Az by as much as 28%.

© RSNA, 2003

Index terms: Diagnostic radiology, observer performance • Lung, nodule, 60.1215, 60.281 • Lung neoplasms, diagnosis, 60.1215, 60.31 • Radiography, digital, 60.1215 • Thorax, radiography, 60.1215


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
RESULTS
 DISCUSSION
 REFERENCES
 
Lung cancer is one of the leading causes of death in the United States; as a cause of death, it is second only to cardiovascular diseases. Notwithstanding the relatively recent initiative for research of lung cancer screening with computed tomography (CT) (1), chest radiography is currently the most common, most cost-effective, and most dose-effective diagnostic tool for the detection of lung cancer (2,3). There is no national screening program currently in place for lung cancer, and, thus, early-stage lung cancer is often discovered in the form of solitary lung nodules when a chest radiograph is obtained in a patient for another purpose. Approximately 40% of lung nodules are manifestations of either primary lung cancer or metastatic disease (4).

American Cancer Society statistics indicate that the 5-year survival rate for patients with lung cancer can be improved from an average of 14% to up to 49% if the disease is diagnosed early, when it is still in a localized stage (3). However, only 15% of lung cancers are discovered in the early stage (5). Failure to detect these lesions may result in costly delays in appropriate treatment, not only for patients with primary cancer but also for those with metastatic cancer in the lungs in whom the stage or evidence of malignancy is being determined or followed up.

The detection of subtle abnormalities on medical images is limited by noise. The noise on chest radiographs can be categorized into two major types: radiographic noise (ie, mottle) (6,7) and anatomic noise (8). In chest radiography, anatomic noise has by far the greatest influence on the detection of pulmonary nodules (9–11). In a previous study (12), the detectability of tissue-equivalent nodules on background images with only radiographic mottle and on background images with radiographic mottle and anatomic noise was measured by means of observer performance experiments. To achieve equivalent detection, the nodules in anatomic noise backgrounds had to have contrast-diameter products at least one order of magnitude larger than those for the nodules in radiographic noise backgrounds. This finding suggests that anatomic noise influences the detection of subtle lung nodules more substantially than does radiographic noise and that if the detection of subtle lung nodules depicted on chest radiographs is to be improved, then a better understanding of the mechanism of this influence is warranted.

The purpose of this study was to assess the influence of anatomic noise on the detectability of subtle lung nodules depicted on chest radiographs. The study was focused on the influence of local anatomic noise—that is, variations immediately surrounding or overlying suspected lesions—as opposed to general anatomic variations within the image (13). Another aim of this study was to substantiate the magnitude of the influence of local anatomic noise on the detection of subtle lung nodules in typical anatomic backgrounds on chest radiographs in terms of variations in nodule size and in area under the receiver operating characteristic curve (Az).


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 REFERENCES
 
Chest Background Images
Six normal posteroanterior chest radiographs were selected from a research database. The database, which was compiled as part of another investigation (12), contained anonymous digital chest radiographs that were obtained in individuals who did not have acute pulmonary abnormalities. Institutional review board approval and informed consent were not required for use of the anonymous radiographs. No specific criteria, other than being normal, were applied for the selection of cases. The selected cases were those of four women and two men (age range, 28–71 years; mean age, 50 years). The normalcy of the image findings and the appropriateness of their use in this study were assessed by one of the authors (W.R.E.).

All radiographs in the database were acquired as 2k images (matrix size, 1,760 x 2,140) with an upright computed radiography chest unit (FCR-9501-HQ; Fuji Medical Systems, Tokyo, Japan) by using 35 x 43-cm2 standard-resolution imaging plates (ST-Va; Fuji Medical Systems) and a moving 12:1 antiscatter grid (Gilardoni, Milan, Italy) (15). All image acquisitions were performed with a 115-kVp x-ray beam emitted from a high-frequency x-ray generator (1050 HF; Acoma, Wheeling, Ill); the exposures were phototimed by using an automatic exposure control that was calibrated to operate the computed radiography system as a 200-speed-equivalent system (22.7 mR entrance exposure for a typical 22-cm-thick patient). The computed radiography sensitivity values of the radiographs used in the study ranged from 196 to 348 (mean, 271).

From the selected chest images, six 8 x 8-cm regions (400 x 400 array) were extracted from random positions within the lung areas on the radiographs—one region per image. The selected regions were used in the nodule simulations and subsequent observer performance experiments, which are detailed in the following text.

Nodule Simulation
Simulated nodules emulating the radiographic characteristics of tissue-equivalent lesions were computationally superimposed into the central area on the six lung images. Superpositioning was performed by subtracting the nodule contrast profiles from the log-signal data on the images. Contrast was defined in terms of relative change in exposure-dependent linear detector signal ({Delta}E/E). The subtraction was performed by means of transforming the desired contrast into the log-signal units associated with the latitude of the acquired computed radiographic image. The simulated nodules were circular and had subject-contrast profiles based on a mathematical function deduced from a database of real lung nodules (16), as follows:

where c(r) is the contrast profile as a function of radial distance (r), C is the peak contrast value of the nodule, and D is the diameter of the nodule at the specified imaging plane expressed as the full-width-at-fifth-maximum of the profile. The results of an earlier study showed that the radiographic appearance of such simulated nodules is indistinguishable from that of real lesions (17).

Five different sizes of simulated nodules were used in the study. These nodules had peak contrast-diameter products that resulted in a range of average detectability values—Az values of approximately 0.60 to approximately 0.95—as determined in a previous study (12). All nodules had contrast-diameter ratios that would correspond to the ratios expected from spherical uniform tissue-equivalent lesions within the lung (16). The characteristics of the simulated nodules that were used in the study are listed in Table 1.


View this table:
[in this window]
[in a new window]

  		TABLE 1. Characteristics of Simulated Nodules Used in Observer Performance Experiment

 
For each of the six lung images, a set of 31 processed images was produced: six images containing no nodule and the remaining 25 images containing single nodules of five sizes placed at five locations—one at the center and four at 6-mm distances from the center. Figure 1 shows the locations of the nodules on the six lung images used in the study. The location of the nodules on the images was varied as a precaution to avoid any visual indication of the locations to the observers while allowing different local background patterns to overlie the nodules. This scheme yielded a total of 150 lung images that contained superimposed simulated nodules and 36 control images (ie, with no nodule superimposed); thus, a total of 186 images were used in the observer experiments. Figure 2 shows examples of the images, with the simulated nodules inserted at or near the center.



View larger version (25K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1. Five locations (L0-L4) where simulated nodules were superimposed near the center of the lung images. To prevent the observers from learning the exact locations of the nodules, the locations were sequentially rotated by 15° on the images. As examples, black circles represent locations in image 1; dark gray circles, locations in image 2; and light gray circles, locations in image 3.

 


View larger version (139K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2a. (a-f) Each of the six images used in the study, with a 6.4-mm-diameter (ie, 0.4-mm contrast-diameter-product) nodule (arrow) inserted at the center.

 


View larger version (158K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2b. (a-f) Each of the six images used in the study, with a 6.4-mm-diameter (ie, 0.4-mm contrast-diameter-product) nodule (arrow) inserted at the center.

 


View larger version (150K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2c. (a-f) Each of the six images used in the study, with a 6.4-mm-diameter (ie, 0.4-mm contrast-diameter-product) nodule (arrow) inserted at the center.

 


View larger version (167K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2d. (a-f) Each of the six images used in the study, with a 6.4-mm-diameter (ie, 0.4-mm contrast-diameter-product) nodule (arrow) inserted at the center.

 


View larger version (148K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2e. (a-f) Each of the six images used in the study, with a 6.4-mm-diameter (ie, 0.4-mm contrast-diameter-product) nodule (arrow) inserted at the center.

 


View larger version (156K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2f. (a-f) Each of the six images used in the study, with a 6.4-mm-diameter (ie, 0.4-mm contrast-diameter-product) nodule (arrow) inserted at the center.

 
After insertion of the nodules, the log-signal values of each image were converted into film optical density values by using a density-versus–log-signal transformation (12). This operation yields a Hurter and Driffield characteristic curve by using two Gaussian gradient functions of contrast versus log-signal. For each Gaussian component, the maximum contrast, SD, and mean contrast are specified. The cumulative integral of the Gaussian components is then calculated as the Hurter and Driffield characteristic curve. For this study, a Hurter and Driffield characteristic curve similar to that generated from a conventional screen-film system for chest radiography was constructed.

For each image, the average pixel value from a 128 x 128 central block was obtained prior to insertion of the nodules and used as the mean value for the higher contrast Gaussian component. This process ensured that the average contrast and the average optical density at the center of all images, where the nodules were inserted, were constant. We "padded" the images with 2.6-cm horizontal and 4.1-cm vertical margins of 1.8 constant optical density to create visual glare characteristics during the reading sessions that were similar to the visual glare characteristics that are present at actual clinical image interpretations. The 186 images were then randomized by using a random-number generator and printed on 31 sheets of film (six images on one sheet) by using a laser printer (model XLP; Eastman-Kodak, Rochester, NY). The printer was calibrated, and its linearity was verified to be between film optical densities of 0.1 and 3.0 before the images were printed.

Observer Performance Experiments
Fifteen experienced radiologists participated in the study. Each observer read the images independently in three separate sessions. The sessions were scheduled at least 1 day apart. At the beginning of each reading session, the observer protocol was explained. The observers were told that there was an 80% chance that a 3–6-mm-diameter nodule was located within 1/2 inch of the center of each image. The observer was instructed to discern whether a nodule was present on each image and to grade his or her level of confidence in this interpretation by using a five-point scale: -2 meant most likely not present; -1, probably not present; 0, equivocal; +1, probably present; and +2, most likely present. A priori knowledge of the approximate location of the nodules was expected to result in higher detection rates than those observed in actual clinical practice. However, this methodology was implemented because the goal of this study was to examine the rate of nodule detection when the approximate location was known.

Before each reading session, the observer viewed two graded example films, each containing six images of a lung background with six superimposed simulated nodules of increasing contrast-diameter product values along with their expected confidence gradings. The viewing of example films was followed by the reading of two practice films. The observer was subsequently provided with the expected answers so that he or she could assess the visibility level of the nodules and adjust his or her confidence grading. All of the readings were performed on the same view box with optimal environmental surroundings, including low ambient lighting. The observers read the images in different orders to average out any possible memory effect. No time constraints were imposed on the observers. The average reading time for each session (10–11 films, 60–66 images) was about 15 minutes, for a total of about 45 minutes of reading per observer.

Data Analysis
We analyzed the observer scores to determine the detectability of the nodules on each image, as averaged among all observers as a function of the nodule contrast-diameter product (or corresponding nodule size) and the nodule location. For each image, the variation in nodule detectability due to nodule size was characterized by a least-squares linear regression fit to the mean observer scores versus the nodule contrast-diameter product. To investigate how the detectability of a nodule may vary from one location to another, paired t test comparisons of the mean scores from any two locations were performed; thus, a total of ten pairwise comparisons per image and per nodule size were performed. The Holm sequential procedure was used to adjust and evaluate the P values that indicated the rejection of the null hypothesis (ie, a definite difference in nodule detectability between any two of the locations being compared) (18). The overall influence of location on nodule detectability among all five locations within an image was also assessed by using analysis of variance of the mean observer scores, with the P values adjusted and evaluated again by using the Holm sequential procedure (18).

To understand the magnitude of the interference of local background tissue in the detection task, we attempted to compare the variation in mean observer score due to nodule location with the variation in mean score due to change in nodule size. The minimally detectable significant difference in mean observer score due to nodule location was equated by using contrast-diameter product–dependent changes in mean observer scores (characterized in terms of the slope of the linear regression fit). This task was performed by using analysis of variance for repeated measures for each combination of image and nodule size (ie, 30 combinations).

With knowledge of the relationship between the contrast-diameter product and the diameter of the tissue-equivalent nodules used in the study, the variation in nodule detectability as a function of nodule location was also expressed in terms of change in nodule diameter. In addition, we derived an empirical relationship from data reported in a previous study (12) to relate the contrast-diameter product of a subtle lung nodule as follows: Az = 0.615log(CD) + 1.231 and R2 = 0.994, where CD is the contrast-diameter product and the Az is between 0.7 and 0.9. The data in the previous study (12) were obtained from very similar experiments in which the exact location was known. With use of this relationship, the magnitude of the nodule location effect was further expressed in terms of Az.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
 REFERENCES
 
The summary descriptive statistics of the observer experiment results are depicted in Figure 3. The x-axis values for different images are slightly shifted for better visualization. For reference, the means and SDs of the scores that corresponded to the control images (ie, with no nodules inserted) are shown at the point where the nodule contrast-diameter product is equal to 0.06 mm (logCD = -1.2 mm). The error bars indicate ±1 SD. In general, there is a gradual increase in mean score as a function of the contrast-diameter product of the nodule.



View larger version (40K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3. Graph illustrates mean observer scores as a function of nodule contrast-diameter products for the six images used in the study. The error bars represent ±1 SD of the mean score. For reference, the mean and SD of the scores for the control images (ie, with no nodules inserted) are shown at the point where the nodule contrast-diameter product equals 0.06 mm (logCD = -1.2). The data points for the different images were slightly shifted for better visualization of the data.

 
The results for images 1 and 6 are notable. On image 1 (Fig 2a), an opacity at the center of the image led to falsely elevated scores when the contrast-diameter product of the nodule was low. Consequently, the mean scores for the nodules with low contrast-diameter products were shifted to higher values. On image 6 (Fig 2f), a calcified costal cartilage at the center of the image greatly obscured the simulated nodules. Consequently, the mean scores generally were low for that image and the relative increase in observer score with increase in contrast-diameter product was modest. The complexities of backgrounds on images 4 and 5 also had an obvious effect on the shape of the observer nodule detection responses. However, for all images, linear fits to the mean score/contrast-diameter product ratio data indicated a statistically significant interdependence between observer score and nodule contrast-diameter product (Table 2).


View this table:
[in this window]
[in a new window]

  		TABLE 2. Linear Regression Relationships between Observer Scores and Nodule Contrast-Diameter Products, Regardless of Nodule Location

 
The contour plots in Figure 4 depict the influence of nodule location on nodule detection in terms of the mean observer scores for each image as a function of nodule size (ie, contrast-diameter product) and nodule location. There is a notable variation in observer score as a function of nodule location for the different images. Within each image, certain locations had more camouflaging effects than others—in other words, some locations were more favorable and some less favorable for the detection of a nodule. For example, the following locations resulted in higher observer scores and better nodule detectability than the other locations on the images: locations 0 and 1 on image 1 for nodules with a contrast-diameter product of 0.4 mm, locations 0–2 and 4 on image 2 for nodules with a contrast-diameter product of 0.2–0.4 mm, locations 1 and 4 on image 3 for nodules with a contrast-diameter product of 0.20–0.28 mm, locations 2–4 on image 4 for nodules with a contrast-diameter product of 0.4 mm, locations 0 and 4 on image 5 for nodules with a contrast-diameter product of 0.14–0.40 mm, and location 2 on image 6 for nodules with a contrast-diameter product of 0.28–0.40 mm. It should be remembered that the location designation on each image was unique (Fig 1), and, thus, the results for specific locations on different images were not related.



View larger version (67K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 4. Contour plots depicting observer scores as a function of nodule location and contrast-diameter product for each of the six images used in the study. The location scale (L0-L4) indicates a path from the center (ie, location 0 [L0]) around the circumference of a 12-mm-diameter central region, as illustrated in Figure 1.

 
The data regarding nodule detection according to location also suggest that the location effect may vary with different nodule sizes. For example, on image 2, with regard to the three largest nodule sizes, location 3 rendered a notably lower mean observer score than the other locations; in this location, nodules with contrast-diameter products in the 0.2–0.4-mm range were masked more substantially than nodules in the other locations. For nodules with a contrast-diameter product of 0.14 mm, however, location 1 appeared to be a less favorable location.

The results of the paired t test for the significance of differences in mean observer score for any two locations per image and per nodule size (ie, contrast-diameter product) are summarized in Figure 5. The results confirm the qualitative indications illustrated in Figure 4, as just summarized: Any two locations for which the observer scores were notably different in Figure 4 are noted to be significantly different in terms of P values in Figure 5.



View larger version (42K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 5. Plotted t test results show significant pairwise differences in nodule detectability between any two locations on each image (see Fig 1). Each solid circle represents a statistically significant difference in the detectability of nodules between the two locations according to Holm ordered rejection P values: solid circles in the 0,1 location column represent P < .005; solid circles in the 0,2 column, P < .010; solid circles in the 0,3 column, P < .015; solid circles in the 0,4 column, P < .020; solid circles in the 1,2 column, P < .025; solid circles in the 1,3 column, P < .030; solid circles in the 1,4 column, P < .035; solid circles in the 2,3 column, P < .040; solid circles in the 2,4 column, P < .045; and solid circles in the 3,4 column, P < .050. All open circles represent marginally significant differences at P < .05. Blank areas indicate no significant difference in nodule detectability between the two locations.

 
The overall significance of the location effect as characterized by using analysis of variance is summarized in Table 3. The results cited in Table 3 are complementary to the t test results illustrated in Figure 5: For the nodule sizes for which analysis of variance results indicated a significant location effect in Table 3, there were more pairwise significant differences indicated in Figure 5. Overall, there were differences in mean scores for the three largest nodule sizes on image 1, for all nodule sizes on image 2, for the four smallest nodule sizes on image 3, for the two largest nodule sizes on image 4, and for the four largest nodule sizes on images 5 and 6.


View this table:
[in this window]
[in a new window]

  		TABLE 3. Significance of Effect of Nodule Location on Nodule Detection

 
The minimally detectable significant differences in mean observer score due to changes in nodule location, for each image and each nodule size, are given in Table 4. Corresponding data, relating variations in observer scores due to nodule location to those due to nodule contrast-diameter product (Fig 3), show the levels of significance of the location effect in terms of the contrast-diameter product and diameter of the nodules. Additional data in Table 4 further relate location-dependent variation to expected changes in Az. The results indicate that, on average, a slight change in a nodule’s position (by about 6–12 mm) causes a variation in observer score that is equivalent to that caused by changing the contrast-diameter product of a nodule by a factor of 1.78 (or 78%) or to that caused by changing the diameter of a nodule by 32%, which corresponds to a 14% change in Az. Averaging across the nodule contrast-diameter products for the six images in this study, location-dependent variation corresponded to a 22%–185% change in contrast-diameter product, a 10%–68% change in diameter, and a 5%–28% change in Az. When the values for all of the images were averaged, there were relatively less variations in location effect for nodules with contrast-diameter products ranging from 0.1 to 0.4 mm: a 63%–92% change in contrast-diameter product, a 27%–36% change in diameter, and a 13%–15% change in Az.


View this table:
[in this window]
[in a new window]

  		TABLE 4. Minimum Detectable Differences in Mean Observer Scores Due to Location and Corresponding Changes in Nodule Contrast-Diameter Product, Diameter, and Detectability in Terms of Az

 

   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
REFERENCES
 
The influence of anatomic noise on the detection of lung nodules may be considered from two aspects: from a global aspect, in which the influence can be characterized in terms of the global statistical measures (ie, for noise power spectrum or variance in fluctuations) of the background tissue in relation to the signal associated with the nodule; and from a local aspect, in which the background tissue pattern immediately surrounding the nodule is the focus.

At the global level, the detectability of a nodule is determined by the degree of the nodule’s distinctiveness from false-positive findings created by the overall noise characteristics of the background tissue. At this level, nodule detection is independent of nodule location. With a global treatment of anatomic noise, one does not take into account the spatial characteristics of the background structure or the possible local correlation between the background and the signal. In the past, many investigators (19–23) have used the global approach to model the detection of lesions on medical images, and this method has proved to be valid for some detection tasks—from dental radiography to CT. In chest radiography, the global treatment of anatomic variations has also resulted in outcomes that correlated closely with observer results (12). However, there have also been indications that slight changes in the location of a nodule can affect nodule detection dramatically (17); these findings suggest that local interference has an important role in the detection of lung nodules on chest radiographs.

Some investigators have tried to assess the degrading influence of local background lung structure on lesion detection. Most noteworthy are the investigations of Revesz et al (11), Revesz and Kundel (24,25), and Kundel et al (26) of the effect of anatomic noise (described as structured noise in the studies) on the detection of lung nodules. Their formulation of the concept of conspicuity in the detection process takes into account the spatial characteristics of the immediate background structure and the possible local correlation between the background noise and the signal. However, the results of some later studies (27–29) indicate that this approach may be inadequate for complete quantification of the influence of background structures.

This study aimed to quantify the influence of local anatomic variations on the detection of subtle lung nodules on chest radiographs; findings indicate that small changes in a nodule’s location can vary the detection of the nodule by an Az of up to 28% (14% average for all images used in this study). This variation is equivalent to that caused by changing the contrast-diameter product of a nodule at a fixed location by a factor of 185% (78% average for all images). With the assumption that nodules are spherical and tissue equivalent, this change would correspond to a 68% change in nodule diameter (32% average for all images).

Several promising methods to reduce the influence of anatomic noise have been developed. The rib pattern is a prominent component of anatomic noise on chest radiographs because about two-thirds of the lung region on posteroanterior chest radiographs is covered by bone (30). Dual-energy digital chest radiography enables one to eliminate bone structures from a chest radiograph by using the difference in spectral absorption characteristics of bone and soft tissue (31–33). To our knowledge, the image quality characteristics of this imaging method, particularly the noise characteristics, have not been fully evaluated. Nevertheless, improved detection of lung nodules with use of this technique has been reported (34–36). In addition, a relatively new technology, dual-energy imaging, is under clinical evaluation.

The notable influence of local anatomic noise on lesion detection suggests that an imaging technique that involved the use of multiple projections of the chest would have the potential to improve the detection of lung nodules. The findings of this study imply that the potential improvement in nodule detection afforded by a multiprojection technique would be more than that expected from the increased chance of detecting a lesion due to an increase in the number of images read. Stereo chest radiography (28,37) and digital tomosynthesis (38–40) are two such techniques. Stereo chest radiography, once a popular imaging technique (41), has long been replaced by two-view posteroanterior or lateral chest radiography owing to economic considerations, additional anatomic features depicted in the lateral view, complications associated with the stereoscopic perception of translucent objects (42,43), and an unproven notion that if a lesion can be seen on a single radiograph, then the acquisition of a second radiograph may not be justified (44). Recent developments in biplane chest imaging suggest that stereo chest radiography should perhaps be revisited as a method for improved detection of lung nodules (45). The second technique, digital tomosynthesis, is currently under investigation and awaits further technical development, commercial implementation, and clinical evaluation.

CT represents an effective method of substantially reducing the influence of anatomic noise on lung nodule detection because it eliminates overlays of anatomic structures associated with projection imaging. Clinical trial results have demonstrated the usefulness of low-dose CT in screening for lung cancer (1). However, at present, the broad use of CT as a direct screening method for subtle lung nodules is controversial owing to economic (eg, cost and technology availability) and patient care (patient dose and overdiagnosis) considerations. If CT is to be used for early cancer detection, then the need for thin-section scans (ie, approximately 1-mm sections) will further increase the cost and dose associated with this imaging examination.


   FOOTNOTES
 
Abbreviation: Az = area under the receiver operating characteristic curve

Author contributions: Guarantor of integrity of entire study, E.S.; study concepts, E.S., M.J.F.; study design, all authors; literature research, E.S.; experimental studies, E.S., M.J.F., W.R.E.; data acquisition, E.S., W.R.E.; data analysis/interpretation, all authors; statistical analysis, E.S., E.P.; manuscript preparation, E.S.; manuscript definition of intellectual content and editing, all authors; manuscript revision/review, E.S.; manuscript final version approval, all authors.


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 REFERENCES
 

  1. Henschke CI, McCauley DI, Yankelevitz DF, et al. Early lung cancer action project: overall design and findings from baseline screening. Lancet 1999; 354:99-105.[CrossRef][Medline]
  2. Muhm JR, Miller WE, Fontana RS, Sanderson DR, Uhlenhopp MA. Lung cancer detected during a screening program using four-month chest radiographs. Radiology 1983; 148:561-565.[Abstract/Free Full Text]
  3. Murphy GP, Lawrence W, Jr, Lenhard RE, Jr. Clinical oncology Washington, DC: American Cancer Society, 1995.
  4. Zweirewich CV, Vedal S, Miller RR, Muller NL. Solitary pulmonary nodule: high-resolution CT and radiologic-pathologic correlation. Radiology 1991; 179:469-476.[Abstract/Free Full Text]
  5. Cancer facts and figures 2003 Atlanta, Ga: American Cancer Society, 2003.
  6. Rose A. The sensitivity performance of the human eye on an absolute scale. J Opt Soc Am 1948; 38:196-208.
  7. Burgess AE, Wagner RF, Jennings RJ. Human signal detection performance for noisy medical images. IEEE Computer Society International Workshop on Medical Imaging 1982; 99-105.
  8. Samei E, Eyler WR, Baron LF. Effects of anatomical structure on signal detection. In: Beutel J, Van Metter R, Kundel H, eds. Handbook of medical imaging. Vol 1, Physics and psychophysics. Bellingham, Wash: SPIE Press, 2000; 655-682.
  9. Greening RG, Pendergrass EP. Postmortem roentgenography with particular emphasis upon the lung. Radiology 1954; 62:720-724.
  10. Boynton RM, Bush WR. Recognition of forms against a complex background. J Opt Soc Am 1956; 46:758-764.[Medline]
  11. Revesz G, Kundel HL, Graber MA. The influence of structured noise on the detection of radiologic abnormalities. Invest Radiol 1974; 9:479-486.[CrossRef][Medline]
  12. Samei E, Flynn MJ, Eyler WR. Detection of subtle lung nodules: relative influence of quantum and anatomical noise on chest radiographs. Radiology 1999; 213:727-734.[Abstract/Free Full Text]
  13. Samei E, Flynn MJ, Eyler WR, Peterson E. The effect of local background anatomical patterns on the detection of subtle lung nodules in chest radiographs. Proc SPIE 1998; 3340:44-54.[CrossRef]
  14. Samei E, Flynn MJ. An experimental comparison of detector performance for computed radiography systems. Med Phys 2002; 29:447-459.[CrossRef][Medline]
  15. Flynn MJ, Samei E. Experimental comparison of noise and resolution for 2k and 4k storage phosphor radiography systems. Med Phys 1999; 26:1612-1623.[CrossRef][Medline]
  16. Samei E, Flynn MJ, Eyler W. Simulation of subtle lung nodules in projection chest radiography. Radiology 1997; 202:117- 124.[Abstract/Free Full Text]
  17. Samei E, Flynn MJ, Beute GH, Peterson E. Comparison of observer performance for real and simulated nodules in chest radiography. Proc SPIE 1996; 2712:60-70.[CrossRef]
  18. Holm S. A sample sequentially negative multiple test procedure. Scand J Stat 1979; 6:65-70.
  19. Seeley GW, Roehrig H, Hillman BJ. A computerized method for measurement of conspicuity: comparison of film and digital nephrograms. Invest Radiol 1984; 19:583-586.[CrossRef][Medline]
  20. Ruttimann UE, Webber RL. A simple model combining quantum noise and anatomical variation in radiographs. Med Phys 1984; 11:50-60.[CrossRef][Medline]
  21. Rolland JP, Barrett HH. Effect of random background inhomogeneity on observer detection performance. J Opt Soc Am A 1992; 9:649-658.[Medline]
  22. Burgess AE. Statistically defined backgrounds: performance of a modified nonprewhitening observer model. J Opt Soc Am A 1994; 11:1237-1242.[Medline]
  23. Judy PF, Swensson RG, Kijewsky MF. Observer detection efficiency of disk and Gaussian targets (abstr). Radiology 1995; 197:356.
  24. Revesz G, Kundel HL. Psychophysical studies of detection errors in chest radiology. Radiology 1977; 123:559-562.[Abstract]
  25. Kundel HL, Revesz G. Lesion conspicuity, structured noise, and film reader error. Am J Roentgenol 1976; 126:1233-1238.[Abstract]
  26. Kundel HL, Revesz G, Toto L. Contrast gradient and the detection of lung nodules. Invest Radiol 1979; 14:18-22.[CrossRef][Medline]
  27. Hallberg JR, Kelsey CA, Briscoe D. Some effects of method on the measured conspicuity of chest lesions. Invest Radiol 1978; 13:439-443.[CrossRef][Medline]
  28. Brogdon BG, Moseley RD, Kelsey CA, Hallberg JR. Perception of simulated lung lesions. Invest Radiol 1978; 13:12-15.[CrossRef][Medline]
  29. Revesz G. Conspicuity and uncertainty in the radiographic detection of lesions. Radiology 1985; 154:625-628.[Abstract/Free Full Text]
  30. Waring JJ, Wasson WW. The imperfections of the stereoscopic maneuver in radiography of the chest. Radiology 1926; 6:198-203.
  31. Lehmann LA, Alvarez RE, Macovski A, et al. Generalized image combinations in dual kVp digital radiography. Med Phys 1981; 8:659-667.[CrossRef][Medline]
  32. Stewart BK, Huang HK. Single-exposure dual-energy computed radiography. Med Phys 1990; 17:866-875.[CrossRef][Medline]
  33. Kamimura R, Takashima T. Clinical application of single dual-energy subtraction technique with digital storage phosphor radiography. J Digit Imaging 1995; 8(1 suppl 1):21-24.
  34. Ergun DL, Mistretta CA, Brown DE, et al. Single-exposure dual-energy computed radiography: improved detection and processing. Radiology 1990; 174:242-249.
  35. Kelcz F, Zink FE, Peppler WW, Kruger DG, Ergun DI, Mistretta CA. Conventional chest radiography vs dual-energy computed radiography in the detection and characterization of pulmonary nodules. AJR Am J Roentgenol 1994; 162:271-278.[Abstract/Free Full Text]
  36. Kido S, Ikezoe J, Naito H, et al. Clinical evaluation of pulmonary nodules with single-exposure dual-energy subtraction chest radiography with an iterative noise-reduction algorithm. Radiology 1995; 194:407-412.[Abstract/Free Full Text]
  37. Kelsey CA, Moseley RD, Mettler FA, Briscoe DE. Cost-effectiveness of stereoscopic radiographs in detection of lung nodules. Radiology 1982; 142:611-613.[Abstract/Free Full Text]
  38. Zwicker RD, Atari NA. Transverse tomosynthesis on a digital simulator. Med Phys 1997; 24:867-871.[CrossRef][Medline]
  39. Dobbins JT, Webber RL, Hames SM. Tomosynthesis for improved pulmonary nodule detection (abstr). Radiology 1998; 209(P):280.
  40. Godfrey DJ, Warp RJ, Dobbins JT, III. Optimization of matrix inverse tomosynthesis. Proc SPIE 2001; 4320:696-704.[CrossRef]
  41. Levitin J. Ten thousand chest examinations with the stereoscopic photoroentgen unit. AJR Am J Roentgenol 1943; 49:469-475.
  42. Batson OV, Carpentier VE. Stereoscopic depth perception. AJR Am J Roentgenol 1944; 51:202-204.
  43. Daves ML, Dalrymple GV. Objective stereoscopy. Radiology 1962; 78:802-805.
  44. Lewis I, Morgan RH. The value of stereoscopy in mass radiography of the chest. Radiology 1946; 46:171-172.
  45. Samei E, Catarious DM, Baydush AB, Floyd CE, Vargas-Voracek R. Bi-plane correlation imaging for improved detection of lung nodules. Proc SPIE. (in press).



This article has been cited by other articles:


Home page
 Br. J. Radiol.Home page
H HAYASHI, K ASHIZAWA, M UETANI, S FUTAGAWA, A FUKUSHIMA, K MINAMI, S HONDA, and K HAYASHI
Detectability of peripheral lung cancer on chest radiographs: effect of the size, location and extent of ground-glass opacity
Br. J. Radiol., April 1, 2009; 82(976): 272 - 278.
[Abstract] [Full Text] [PDF]

Home page
 Am. J. Roentgenol.Home page
F. Girvin and J. P. Ko
Pulmonary Nodules: Detection, Assessment, and CAD
Am. J. Roentgenol., October 1, 2008; 191(4): 1057 - 1069.
[Abstract] [Full Text] [PDF]

Home page
 Radiat Prot DosimetryHome page
M. J. Tapiovaara
Review of relationships between physical measurements and user evaluation of image quality
Radiat Prot Dosimetry, March 1, 2008; 129(1-3): 244 - 248.
[Abstract] [Full Text] [PDF]

Home page
 Br. J. Radiol.Home page
N Buls, W Shabana, P Verbeek, P Pevenage, and J De Mey
Influence of display quality on radiologists' performance in the detection of lung nodules on radiographs
Br. J. Radiol., September 1, 2007; 80(957): 738 - 743.
[Abstract] [Full Text] [PDF]

Home page
 Br. J. Radiol.Home page
D S Brettle, E Berry, and M A Smith
The effect of experience on detectability in local area anatomical noise
Br. J. Radiol., March 1, 2007; 80(951): 186 - 193.
[Abstract] [Full Text] [PDF]

Home page
 RadiologyHome page
H. P. McAdams, E. Samei, J. Dobbins III, G. D. Tourassi, and C. E. Ravin
Recent Advances in Chest Radiography
Radiology, December 1, 2006; 241(3): 663 - 683.
[Abstract] [Full Text] [PDF]

Home page
 RadiologyHome page
L. J. M. Kroft, W. J. H. Veldkamp, B. J. A. Mertens, J. P. A. van Delft, and J. Geleijns
Detection of simulated nodules on clinical radiographs: dose reduction at digital posteroanterior chest radiography.
Radiology, November 1, 2006; 241(2): 392 - 398.
[Abstract] [Full Text] [PDF]

Home page
 RadiologyHome page
H. W. Venema, M. van Straten, G. J. den Heeten, T. M. Bernhardt, H. Lenzen, and W. Heindel
Digital Radiography of the Chest: Reassessment of the High-Voltage Technique? * Dr Bernhardt and colleagues respond:
Radiology, April 1, 2005; 235(1): 336 - 338.
[Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2273020509v1
228/1/76    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Samei, E.
Right arrow Articles by Eyler, W. R.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Samei, E.
Right arrow Articles by Eyler, W. R.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
RADIOLOGY RADIOGRAPHICS RSNA JOURNALS ONLINE