HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Figures Only Full Text (PDF) Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Gruden, J. F. Articles by McGuinness, G. Search for Related Content PubMed PubMed Citation Articles by Gruden, J. F. Articles by McGuinness, G.

Multinodular Disease: Anatomic Localization at Thin-Section CT—Multireader Evaluation of a Simple Algorithm

¹ Department of Radiology, Chest Section, New York University Hospitals System and Bellevue Hospital Center, 560 First Ave, New York, NY 10003 (J.F.G., D.P.N., G.M.)
² Thoracic Imaging Section, University of California San Francisco (W.R.W.).

	Abstract

TOP Abstract Introduction MATERIALS AND METHODS RESULTS DISCUSSION References

 
PURPOSE: To evaluate the interobserver variability and accuracy of an algorithm for anatomic localization of small nodules evident on thin-section computed tomographic (CT) images of the lungs.

MATERIALS AND METHODS: Four experienced chest radiologists independently evaluated thin-section CT images in 58 patients by using an algorithm and a standard score sheet. Nodules were placed into four possible anatomic locations or categories: perilymphatic, random, associated with small airways disease, or centrilobular. Algorithm accuracy was assessed by comparing the localization by the observers to that expected for each specific disease in the study group on the basis of reports in the literature. Interobserver variability was assessed by placing cases into one of three groups: (a) complete concordance, (b) triple concordance, and (c) discordant.

RESULTS: All observers agreed in 79% (46 of 58) of the cases with regard to nodule localization; three of the four concurred in an additional 17% (10 of 58). The observers were correct in 218 (94%) of 232 localizations in the 58 cases. There were no apparent differences in the number of either discordant or incorrect localizations between the observers. The most noteworthy source of error and of disagreement between observers was the confusion of perilymphatic and small airways disease–associated nodules in a small number of cases.

CONCLUSION: The proposed algorithm is reproducible and accurate in the majority of cases and facilitates nodule localization at thin-section CT.

Index terms: Computed tomography (CT), thin-section, 60.12118 • Lung, CT, 60.12118 • Lung, infection, 60.20, 60.23, 60.2031 • Lung, interstitial disease, 60.28, 60.79 • Lung, nodule, 60.28 • Lung neoplasm, secondary, 60.33 • Mycobacteria, 60.2031 • Sarcoidosis, 60.22 • Tuberculosis, 60.23

	Introduction

TOP Abstract Introduction MATERIALS AND METHODS RESULTS DISCUSSION References

 
There are many causes of multiple small nodular opacities evident on thin-section computed tomographic (CT) images of the lung parenchyma. Accurate thin-section CT interpretation and differential diagnosis in these patients requires detailed anatomic nodule localization (1–4). Although specific thin-section CT findings in individual multinodular diseases are well known, previous descriptions of nodule localization are anecdotal or are based on thin-section CT–histopathologic correlation in specific disease states (5–7). To our knowledge, a structured approach to nodule localization and image interpretation with proved accuracy in the individual patient does not exist; in addition, the effects of interobserver variability in this setting are unknown, as are potential sources of observer or technical error.

We evaluated an algorithm for anatomic nodule localization at thin-section CT in an effort to (a) determine the reproducibility of findings with the algorithm between multiple observers, (b) evaluate the accuracy of the algorithm result in comparison with the expected anatomic nodule location in specific disease states, and (c) identify and explain sources of observer disagreement and potential sources of error in algorithm application.

	MATERIALS AND METHODS

TOP Abstract Introduction MATERIALS AND METHODS RESULTS DISCUSSION References

 
Patients who underwent thin-section CT of the thorax and in whom the dominant pattern of disease was multinodular were identified through review of the CT case records (logbooks, databases) and corresponding interpretative reports at two large teaching institutions (New York University Medical Center and Bellevue Hospital Center, New York) for 1 year; one author (J.F.G.) reviewed all images to corroborate the initial clinical interpretation of multinodular disease. Image-based criteria for study entry included (a) the presence of multiple small nodular opacities (diameter, <1 cm) involving both lungs and (b) in the presence of other abnormalities, small nodules representing the dominant disease pattern. All examinations were performed with the thin-section CT technique standard at both institutions: 1.0- or 1.5-mm collimation, 10- or 20-mm intersection gap, 170 kVp, 140 mA, 2-second scanning time.

All patients had a definitive final diagnosis based on (a) sputum smear or culture results (n = 15), (b) bronchoalveolar lavage or transbronchial biopsy results (n = 33), (c) video-assisted thoracoscopic surgery findings (n = 2), or (d) a diagnostic thin-section CT appearance in the appropriate clinical setting (n = 8) (8). All patients in whom the final diagnosis was based solely on sputum smear or culture results, bronchoalveolar lavage results, or a diagnostic thin-section CT appearance had clinical and radiographic follow-up findings documenting either disease resolution after appropriate therapy or a clinical and radiographic course otherwise consistent with the initial diagnosis. The final study population consisted of 58 patients (58 cases; 32 men, 26 women; age range, 35–76 years).

Localization Algorithm
For the purposes of algorithm application, we classified nodules as belonging to one of four possible anatomic types similar to those described by Colby and Swensen (7): (a) hematogenous or random, (b) interstitial or perilymphatic, (c) centrilobular, and (d) associated with small airways disease. Small airways disease–associated nodules, technically centrilobular in location, have previously been grouped with other causes of centrilobular disease (9,10). For the purposes of this study, we considered small airways disease–associated nodules as a separate entity because of the distinctive imaging appearance of impacted bronchioles, which resemble a "tree-in-bud" or "jacks," and air-space nodules in relation to other centrilobular nodules (11–13). This nodule morphology indicates the presence of infectious bronchiolitis; ill-defined, sublobular nodules of early or mild peribronchiolar air-space consolidation (so-called air-space nodules), also common in patients with infection, have a distinctive appearance and may occur with or without the presence of bronchiolar impaction. Small airways disease–associated nodules (bronchiolar impaction or air-space nodules) were also given an independent designation because, as will be discussed later, the diagnostic and therapeutic management in this group of patients is unique with respect to that in patients with other centrilobular diseases.

Using these four nodule categories, the authors developed and refined a standard localization algorithm; this has been used clinically and for teaching purposes at our institutions for nearly 2 years.

The first step in the algorithm (Fig 1) requires determining whether nodules abut pleural or fissural surfaces, with profusion scored visually as either 0 for none, 1 for a few (ie, <10% of the total number of nodules present), or 2 for many (ie, 10% of all nodules). Only the small (<1-cm) nodules are assessed, and areas of confluent and coalescent disease are omitted, if present.

View larger version (15K): [in this window] [in a new window]   Figure 1. Flowchart of the first step in the algorithm shows how the algorithm is applied to the number and distribution of nodules. CEN = centrilobular, PL = perilymphatic, RAND = random, SAD = small airways disease associated.

View larger version (15K): [in this window] [in a new window]   Figure 2. Flowchart shows perilymphatic-random nodule arm. PL = perilymphatic, RAND = random.

View larger version (15K): [in this window] [in a new window]   Figure 3. Flowchart shows the small airways disease–associated–centrilobular nodule arm. CEN = centrilobular, SAD = small airways disease associated.

View larger version (20K): [in this window] [in a new window]   Figure 4. Flowchart shows indeterminate nodule arm. CEN = centrilobular, PL = perilymphatic, RAND = random, SAD = small airways disease associated.

View larger version (96K): [in this window] [in a new window]   Figure 5a. (a) Thin-section CT image shows perilymphatic nodules in the lungs of a patient with sarcoidosis. There are many nodules, with a profusion score of 2, along the right major fissure and lateral pleural surface; the patchy distribution distinguishes perilymphatic from random nodules. Note the beading of visible peripheral arterial structures (arrows) that is typical of axial interstitial disease. (b) Thin-section CT image shows random nodules in the lungs of a patient with metastases. Many nodules, with a profusion score of 2, reach the pleural surfaces and fissures (arrows); the distribution is diffuse. The axial and peripheral interstitia appear normal. (c) Thin-section CT image shows small airways disease–associated nodules in the left lung of a patient with bacterial bronchiolitis. Nodular opacities spare pleural and fissural surfaces, with a profusion score of 0; the patchy distribution is diagnostic of small airways disease–associated nodules. Note the typical appearance of bronchiolar impaction (arrows). (d) Thin-section CT image shows centrilobular nodules in the left lung of a patient with hypersensitivity pneumonitis. The pleural and fissural surfaces are spared, with a profusion score of 0; the ground-glass attenuation nodules are diffuse, and impacted airways are absent. Note the relationship of the nodules to the visible arterial branches (arrows), which are themselves centrilobular structures.

View larger version (107K): [in this window] [in a new window]   Figure 5b. (a) Thin-section CT image shows perilymphatic nodules in the lungs of a patient with sarcoidosis. There are many nodules, with a profusion score of 2, along the right major fissure and lateral pleural surface; the patchy distribution distinguishes perilymphatic from random nodules. Note the beading of visible peripheral arterial structures (arrows) that is typical of axial interstitial disease. (b) Thin-section CT image shows random nodules in the lungs of a patient with metastases. Many nodules, with a profusion score of 2, reach the pleural surfaces and fissures (arrows); the distribution is diffuse. The axial and peripheral interstitia appear normal. (c) Thin-section CT image shows small airways disease–associated nodules in the left lung of a patient with bacterial bronchiolitis. Nodular opacities spare pleural and fissural surfaces, with a profusion score of 0; the patchy distribution is diagnostic of small airways disease–associated nodules. Note the typical appearance of bronchiolar impaction (arrows). (d) Thin-section CT image shows centrilobular nodules in the left lung of a patient with hypersensitivity pneumonitis. The pleural and fissural surfaces are spared, with a profusion score of 0; the ground-glass attenuation nodules are diffuse, and impacted airways are absent. Note the relationship of the nodules to the visible arterial branches (arrows), which are themselves centrilobular structures.

View larger version (136K): [in this window] [in a new window]   Figure 5c. (a) Thin-section CT image shows perilymphatic nodules in the lungs of a patient with sarcoidosis. There are many nodules, with a profusion score of 2, along the right major fissure and lateral pleural surface; the patchy distribution distinguishes perilymphatic from random nodules. Note the beading of visible peripheral arterial structures (arrows) that is typical of axial interstitial disease. (b) Thin-section CT image shows random nodules in the lungs of a patient with metastases. Many nodules, with a profusion score of 2, reach the pleural surfaces and fissures (arrows); the distribution is diffuse. The axial and peripheral interstitia appear normal. (c) Thin-section CT image shows small airways disease–associated nodules in the left lung of a patient with bacterial bronchiolitis. Nodular opacities spare pleural and fissural surfaces, with a profusion score of 0; the patchy distribution is diagnostic of small airways disease–associated nodules. Note the typical appearance of bronchiolar impaction (arrows). (d) Thin-section CT image shows centrilobular nodules in the left lung of a patient with hypersensitivity pneumonitis. The pleural and fissural surfaces are spared, with a profusion score of 0; the ground-glass attenuation nodules are diffuse, and impacted airways are absent. Note the relationship of the nodules to the visible arterial branches (arrows), which are themselves centrilobular structures.

View larger version (128K): [in this window] [in a new window]   Figure 5d. (a) Thin-section CT image shows perilymphatic nodules in the lungs of a patient with sarcoidosis. There are many nodules, with a profusion score of 2, along the right major fissure and lateral pleural surface; the patchy distribution distinguishes perilymphatic from random nodules. Note the beading of visible peripheral arterial structures (arrows) that is typical of axial interstitial disease. (b) Thin-section CT image shows random nodules in the lungs of a patient with metastases. Many nodules, with a profusion score of 2, reach the pleural surfaces and fissures (arrows); the distribution is diffuse. The axial and peripheral interstitia appear normal. (c) Thin-section CT image shows small airways disease–associated nodules in the left lung of a patient with bacterial bronchiolitis. Nodular opacities spare pleural and fissural surfaces, with a profusion score of 0; the patchy distribution is diagnostic of small airways disease–associated nodules. Note the typical appearance of bronchiolar impaction (arrows). (d) Thin-section CT image shows centrilobular nodules in the left lung of a patient with hypersensitivity pneumonitis. The pleural and fissural surfaces are spared, with a profusion score of 0; the ground-glass attenuation nodules are diffuse, and impacted airways are absent. Note the relationship of the nodules to the visible arterial branches (arrows), which are themselves centrilobular structures.

Second (Fig 2), further categorization involves assessment of the interstitium, both axially (peribronchovascular and peribronchiolovascular) and peripherally (interlobular septal), scored visually as either normal or abnormal (thickened, and either smooth or nodular). By definition, patients with interstitial abnormalities have a diffuse form of perilymphatic disease, while those with a normal interstitium have random nodules (Fig 5b).

Algorithm Arm Comprising Centrilobular and Small Airways Disease–associated Nodules
Similar to nodules in the perilymphatic-random nodule arm of the algorithm, nodules in the small airways disease–associated–centrilobular nodule arm are separated by using a two-step approach (Fig 3). The first step again requires assessment of nodule distribution by using definitions identical to those used in the perilymphatic-random nodule arm. By definition, a patchy distribution is diagnostic of small airways disease–associated nodules (Fig 5c). A diffuse distribution is in itself nonspecific and may represent either a centrilobular or a diffuse form of small airways disease–associated nodules. Differentiation simply requires a second step; identification of impacted bronchioles or typical air-space nodules is diagnostic of small airways disease. In the absence of such nodule morphology, diffuse nodules are considered centrilobular (Fig 5d).

Algorithm Arm Comprising Indeterminate Nodules
This arm (Fig 4) includes those cases in which the initial pleural surface or fissural nodule profusion is scored as 1 (indeterminate); in such cases, differentiation between the perilymphatic-random nodule arm and the small airways disease–associated–centrilobular nodule arm is problematic. Assessment of nodule distribution (patchy or diffuse) is again the initial step in the indeterminate nodule arm. Patchy nodules indicate the presence of either perilymphatic or small airways disease–associated nodules (see earlier). As before, the presence of small airway impactions or typical air-space nodules indicates small airways disease; interstitial abnormalities are diagnostic of perilymphatic disease.

Diffusely distributed nodules are also evaluated as previously described, with bronchiolar impaction or air-space nodules diagnostic of small airways disease and interstitial abnormalities diagnostic of perilymphatic disease. In the absence of either of these abnormalities, centrilobular and random nodules are distinguished on the basis of nodule morphology. By definition, nodules of uniform ground-glass attenuation indicate centrilobular disease, while all other nodules, including those that are well-defined or of uniform or mixed attenuation, are considered indeterminate and are designated centrilobular-random (both nodules can have these characteristics). By following this approach, categorization of nodules as random can never occur through application of the indeterminate nodule arm, as they cannot be accurately distinguished from centrilobular nodules in cases of indeterminate pleural or fissural profusion (Fig 4).

Data Assessment
All cases were evaluated by using only lung windows (level, -700 HU; width, -1,000 HU to -1,500 HU) and were reviewed in random order by each of four independent observers (J.F.G., W.R.W., D.P.N., G.M.), all experienced chest radiologists proficient in thin-section CT image interpretation. Observers were blinded to patient name, institution of origin, and clinical information. The cases were reviewed in random order, differing for each observer. A standard score sheet was filled out for each case on the basis of the application of the localization algorithm.

To assess algorithm reproducibility, we separated cases into three major groups on the basis of the final localization by each observer in each case: (a) complete concordance (all observers agreed), (b) triple concordance (three of the four agreed), and (c) discordant (two or none agreed).

The accuracy of the algorithm was assessed by comparing the responses of the observers to a standard answer key developed by the authors; this answer key was a list of all final diagnoses in the study group with the expected anatomic nodule locations for each disease state categorized on the basis of previous reports of radiologic-histopathologic correlation, the known histopathologic distribution of nodules in each disorder, and the clinical experience of the authors (5,7). The authors developed the answer key (Table 1) at the conclusion of the study and had no knowledge of the case distribution at the time of the interpretative sessions. All interpretations were considered correct (agreement with the answer key), incorrect (disagreement), or indeterminate (centrilobular-random nodule interpretations).

	RESULTS

TOP Abstract Introduction MATERIALS AND METHODS RESULTS DISCUSSION References

View larger version (93K): [in this window] [in a new window]   Figure 6. Thin-section CT image shows nodules with discordant localization in the lungs of a patient with lymphohematogenous metastases. Many pleural and fissural nodules are present in a diffuse distribution; interstitial disease with septal lines (arrows) was noted by three observers, which indicates perilymphatic disease. The final observer did not note interstitial abnormalities and considered the nodules random. A right pleural effusion (E) is present. Either perilymphatic or random nodules could be considered diagnostic of metastases in this patient with known malignancy.

Third, the distinction between centrilobular, random, and centrilobular-random nodule localizations led to single-observer discordance in two cases (one case of hypersensitivity pneumonitis and the single case of lymphocytic interstitial pneumonitis). These disagreements related to interobserver variability in the assessment of nodule attenuation alone; all observers applied the indeterminate nodule arm in these two cases, but the discordant observer interpreted nodule attenuation as mixed rather than as ground-glass alone, leading to centrilobular-random rather than centrilobular nodule localizations (Table 3).

Fourth, disagreements occurred in both cases of Langerhans cell histiocytosis (Tables 3, 4). In one case, three observers applied the perilymphatic-random nodule arm but then disagreed in the evaluation of the interstitium; the final observer applied the indeterminate nodule arm and localized the nodules as centrilobular-random. In the other case, disagreement reflected interobserver variation in the assessment of pleural or fissural nodule profusion. The concordant observers applied the indeterminate nodule arm and localized the nodule as centrilobular-random; the discordant observer applied the perilymphatic-random nodule arm and localized the nodule as random.

No single reader accounted for more discordant interpretations than the others. At least two discordant localizations were rendered by each of the four readers. None of the major disagreements outlined earlier was problematic for only one reader.

Accuracy
Overall, 218 (94%) of 232 individual interpretations were correct with respect to the answer key (Table 1). Nodules were correctly localized in 45 (98%) of 46 cases of complete localization concordance; the concordant observers were correct in all cases of triple concordance. The discordant observer was also correct (more than one correct answer was possible in some cases) in three of the latter cases. All observers were also correct in one (50%) of the two cases of limited concordance; in the other case (Langerhans cell histiocytosis), only one observer was correct.

There were three major error types. The first was case-specific. In one case of sarcoidosis, all readers agreed on a centrilobular rather than the expected perilymphatic nodule distribution (Fig 7). Both cases of Langerhans cell histiocytosis were problematic; four of the eight interpretations in these two cases were incorrect (Fig 8). Thus, these three cases accounted for eight (57%) of 14 errors.

View larger version (118K): [in this window] [in a new window]   Figure 7. Thin-section CT image shows nodules in the lungs of a patient with sarcoidosis with incorrect, completely concordant localization. All observers applied the indeterminate nodule arm because of the presence of a few pleural and subfissural nodules (curved arrow). The profusion score is 1. The diffuse distribution, absence of impacted airways or interstitial abnormality, and uniform ground-glass attenuation of the nodules (straight arrows) led all observers to a centrilobular nodule localization. This is an atypical manifestation of sarcoidosis.

View larger version (102K): [in this window] [in a new window]   Figure 8. Thin-section CT image shows nodules in the lungs of a patient with Langerhans cell histiocytosis with incorrect localization. Two observers applied the perilymphatic-random nodule arm and considered the patchy nodules perilymphatic; the other observers applied the indeterminate nodule arm but disagreed as to nodule attenuation and diagnosed centrilobular and centrilobular-random nodules. A few nodules are in fact highly attenuating (white arrows). Note the presence of cavitation or cysts (black arrows) in some of the nodules, which is typical of this disease but is not considered in the algorithm application.

The third error type comprised two errors due to subjective differences in the assessment of nodule attenuation in the small airways disease–associated–centrilobular nodule arm (Fig 3); discordant observers considered the nodules to be of mixed attenuation (both ground-glass and high attenuation) rather than ground-glass attenuation alone in single cases of hypersensitivity pneumonitis and lymphocytic interstitial pneumonitis, leading to centrilobular-random rather than centrilobular nodule localizations.

Each of the individual readers accounted for three to five incorrect interpretations; three of the four readers confused perilymphatic and small airways disease–associated nodules at least once.

	DISCUSSION

TOP Abstract Introduction MATERIALS AND METHODS RESULTS DISCUSSION References

 
Accurate differential diagnosis of multinodular disease at thin-section CT requires accurate and specific anatomic nodule localization. It was neither our purpose nor our interest to attempt specific thin-section CT diagnoses or to integrate other imaging findings into the algorithm to enable such diagnoses; our aim was simply to evaluate the reproducibility and anatomic accuracy of a localization algorithm. Differential diagnoses for each nodule type can be found in an authoritative textbook (5,8).

Irrespective of specific or differential diagnoses, however, immediate patient treatment hinges on accurate nodule localization. Patients with perilymphatic or random nodules often undergo transbronchial biopsy for diagnosis. However, sarcoidosis and metastatic disease, the most common causes of these nodules, often have diagnostic thin-section CT appearances when the imaging findings are correlated with clinical information. The majority of patients with centrilobular nodules have hypersensitivity pneumonitis; again, in the proper clinical context, the thin-section CT appearance is diagnostic if the nodules are correctly assessed. The few patients in this group who do not have a relevant exposure history may require transbronchial biopsy, but empiric steroid therapy may be administered in some patients without an immediate tissue biopsy. Patients who have small airways disease–associated nodules and who have infectious bronchiolitis or bronchopneumonia are treated with empiric antibiotics and sputum analysis; diagnostic procedures are otherwise not immediately required (29,30).

Reproducibility of the Algorithm
All observers agreed on localization in 79% (46 of 58) of all cases; at least three agreed in 97% (56 of 58). These results are impressive given the independent and often subjective nature of the interpretations and the varied nature of the case mix. Observer agreement was high for all nodule types and for all of the diseases represented, with the exception of lymphocytic interstitial pneumonitis (one case) and Langerhans cell histiocytosis (two cases).

The distinction between perilymphatic and random nodules accounted for interpretative disagreement in four (33%) of the 12 cases without complete concordance. This reflected the variable subjective assessment of the interstitium in the perilymphatic-random nodule arm in cases of miliary infection or metastatic disease (Fig 2), which is essential to the distinction between these nodules in cases in which nodule distribution is diffuse. The clinical history is of paramount importance in such cases; in patients with a known malignancy, diffuse nodules, whether perilymphatic or random, are usually diagnostic of metastatic disease. Either lymphangitic carcinoma or hematogenous metastases may result in a perilymphatic distribution; interstitial abnormalities can occur even with experimentally induced hematogenous metastases (31). In such cases, the term "lymphohematogenous" metastases may be appropriate; precise nodule localization as random or perilymphatic may not be possible or necessary. Similarly, miliary infection, particularly tuberculosis, can cause interstitial abnormalities, typically septal lines (23,24). Regardless of the perilymphatic or random nodule localization in these scenarios, patients with diffuse nodules typically undergo transbronchial biopsy if a tissue diagnosis is required; the initial entry into the perilymphatic-random nodule arm (assessment of pleural and fissural profusion) is the key diagnostic determination.

These interpretative subtleties did not result in observer error; either perilymphatic or random distributions were considered correct in patients with metastases or miliary infection for the reasons discussed earlier (answer key [Table 1]). The eight remaining cases in which disagreement occurred (67% of 12) involved discordant interpretations that were also incorrect and are detailed later.

Accuracy of the Algorithm
Individual interpretations were correct in 94% (218 of 232) of all observations with respect to the answer key (Table 1). The 14 errors occurred in eight specific cases and varied in type and importance.

The first error type comprised three cases that accounted for more than half of all errors. One case of sarcoidosis was incorrectly considered centrilobular by all observers because of the presence of diffuse nodules of ground-glass attenuation without pleural or fissural involvement. This is a rare but known appearance of this disease not reflected in the answer key; the key included only the most commonly expected nodule locations. Given the lack of an exposure history to suggest hypersensitivity pneumonitis, transbronchial biopsy was performed, and patient outcome was not affected by the thin-section CT interpretation. Errors also occurred in both cases of Langerhans cell histiocytosis; the nodules in this disease are known to be variable in anatomic location (7,26,27). Cystic changes characteristic of the disease were present in each instance, but the algorithm does not include such ancillary observations.

The second error type related to the distinction between perilymphatic and small airways disease–associated nodules. These errors occurred only in the application of the indeterminate nodule arm; in most instances, these nodules are effectively separated because pleural or fissural nodules are present in perilymphatic disease but not in infectious bronchiolitis or bronchopneumonia. The algorithm is actually of great benefit in this distinction despite the superficial resemblance of the two nodules (32). Nodularity in the peribronchiolovascular axial interstitium, particularly common in sarcoidosis but possible in other perilymphatic diseases as well, causes apparent beading of the visible arterial branches in the lung periphery. These branching opacities superficially resemble impacted small airways (Fig 9). In cases of indeterminate pleural or fissural profusion, close morphologic assessment of the nodules may be important to avoid confusion. In general, impacted airways are smooth, tubular opacities (impacted bronchioles) with nodularity peripherally located. Perilymphatic disease results in nodularity distributed along the entire course of the tubular structure (the artery). Ancillary findings such as lymph node enlargement in sarcoidosis and bronchiolar dilatation or air trapping in cases of bronchiolitis may also afford distinction. The clinical presentations in patients with these disorders are also usually different. The four errors that occurred in the differentiation of perilymphatic and small airways disease–associated nodules (in four different cases) represented a small fraction of the total observations in the 29 patients with either sarcoidosis or bronchiolitis (Fig 10).

View larger version (123K): [in this window] [in a new window]   Figure 9a. Utility of the algorithm in distinguishing perilymphatic from small airways disease–associated nodules. (a) Thin-section CT image shows nodules in the lungs of a patient with bacterial bronchiolitis. The pleural and fissural surfaces are spared; the patchy distribution is diagnostic of small airways disease–associated nodules. Note the morphology of bronchiolar impactions (arrows) in the lingula and left lower lobe. Mild bronchiectasis is also present in the left lower lobe. (b) Thin-section CT image shows nodules in the lungs of a patient with sarcoidosis. Although the nodules are superficially similar to those in a, the profusion of pleural and fissural nodules (arrowheads along the right major fissure) leads to application of the perilymphatic-random nodule arm; the patchy distribution is diagnostic of perilymphatic disease. Note that peripheral axial interstitial disease, seen as beading of the visible arterial branches (arrows), closely mimics the appearance of bronchiolar impaction.

View larger version (127K): [in this window] [in a new window]   Figure 9b. Utility of the algorithm in distinguishing perilymphatic from small airways disease–associated nodules. (a) Thin-section CT image shows nodules in the lungs of a patient with bacterial bronchiolitis. The pleural and fissural surfaces are spared; the patchy distribution is diagnostic of small airways disease–associated nodules. Note the morphology of bronchiolar impactions (arrows) in the lingula and left lower lobe. Mild bronchiectasis is also present in the left lower lobe. (b) Thin-section CT image shows nodules in the lungs of a patient with sarcoidosis. Although the nodules are superficially similar to those in a, the profusion of pleural and fissural nodules (arrowheads along the right major fissure) leads to application of the perilymphatic-random nodule arm; the patchy distribution is diagnostic of perilymphatic disease. Note that peripheral axial interstitial disease, seen as beading of the visible arterial branches (arrows), closely mimics the appearance of bronchiolar impaction.

View larger version (102K): [in this window] [in a new window]   Figure 10a. Errors in the distinction between perilymphatic and small airways disease–associated nodules. (a) Thin-section CT image shows nodules in the right lung of a patient with sarcoidosis. Note the branching configuration of the patchy nodular opacities in the periphery of the lung (arrows); pleural and fissural nodules are present but are predominantly in the region of the major fissure (arrowheads) rather than along the pleural surfaces. One observer failed to notice the fissural involvement and applied the indeterminate nodule arm; the peripheral branching structures are thought to represent impacted bronchioles. (b) Thin-section CT image shows nodules in the right lung of a patient with infectious bronchiolitis. Small, branching, nodular structures are present in the right lower lobe. One observer applied the indeterminate nodule arm (a few pleural and fissural nodules are present but are not shown) and mistakenly diagnosed perilymphatic disease. Note that the nodular opacities (arrows) are confined to the ends of the branching structures and are not distributed along their entire course. This may allow distinction between axial interstitial nodularity and impacted bronchioles in difficult cases.

View larger version (103K): [in this window] [in a new window]   Figure 10b. Errors in the distinction between perilymphatic and small airways disease–associated nodules. (a) Thin-section CT image shows nodules in the right lung of a patient with sarcoidosis. Note the branching configuration of the patchy nodular opacities in the periphery of the lung (arrows); pleural and fissural nodules are present but are predominantly in the region of the major fissure (arrowheads) rather than along the pleural surfaces. One observer failed to notice the fissural involvement and applied the indeterminate nodule arm; the peripheral branching structures are thought to represent impacted bronchioles. (b) Thin-section CT image shows nodules in the right lung of a patient with infectious bronchiolitis. Small, branching, nodular structures are present in the right lower lobe. One observer applied the indeterminate nodule arm (a few pleural and fissural nodules are present but are not shown) and mistakenly diagnosed perilymphatic disease. Note that the nodular opacities (arrows) are confined to the ends of the branching structures and are not distributed along their entire course. This may allow distinction between axial interstitial nodularity and impacted bronchioles in difficult cases.

The third error type, which occurred in only two cases, involved observer variation in the subjective assessment of nodule attenuation. This may be problematic only in the indeterminate nodule arm (Fig 4). The algorithm usually allows distinction between centrilobular and random nodules on the basis of the initial assessment of pleural and fissural nodule profusion.

Anatomic Basis for Nodule Localization and the Algorithm
The four nodule locations described in this study are superficially unique with respect to prior descriptions only in that small airways disease–associated nodules are considered a distinct entity rather than a type of centrilobular disease. This classification closely follows that proposed by Colby and Swensen (7). As discussed, the characteristic appearance of small airways disease–associated nodules and the unique clinical treatment of these patients warrant this unique classification. However, there are other substantive differences between the current approach and the traditional scheme used to describe the anatomic location of small nodules at thin-section CT; prior reports have focused on observations in relation to the individual secondary pulmonary lobule (2,4).

At the level of the secondary lobule, the only visible structure in the normal lung is the arterial branch; the accompanying bronchiole cannot be identified in the normal state. Nodules in the interlobular septa or along pleural or fissural surfaces at the level of the individual lobule have been described as perilobular or interstitial (perilymphatic) (2). Those that appear to lie along the arterial branch, whether of high attenuation, of ground-glass attenuation, or of a branching nature, have been described as "intralobular" or "centrilobular" (9,10). Nodules distributed both along the course of the visible arteries and along the lobular borders (septa or pleural or fissural surfaces) have been classified as "panlobular" or "random" (hematogenous) (2).

There are limitations to this approach. For example, each of the four nodule types in our classification scheme can have branching or rounded nodules that appear anatomically centrilobular at the lobular level. In addition to truly centrilobular nodules, random nodules along the arterial branch, perilymphatic nodules in the peribronchiolovascular axial interstitium, and impacted small airways all technically appear centrilobular at the lobular level.

The algorithm changes the initial focus from the individual lobule to the lung as a whole; by focusing on an initial global assessment of pleural or fissural nodule profusion, the apparent branching centrilobular opacities of perilymphatic and those of small airways disease are readily distinguished, as are the centrilobular nodules of random or truly centrilobular disorders (Fig 11). To our knowledge, Remy-Jardin and colleagues (15) were the first to report on the potential importance of apparent pleural or fissural nodularity in distinguishing various causes of nodular lung disease.

View larger version (114K): [in this window] [in a new window]   Figure 11a. Algorithm utility in comparison with lobular-level nodule localization. (a) Thin-section CT image shows innumerable diffuse, highly attenuating, well-defined nodules with apparent pleural nodularity in this patient with metastatic adenocarcinoma of the breast. Some nodules (arrows) lie along the course of vessels and therefore could be considered intralobular or centrilobular at the level of the secondary lobule. (b) Thin-section CT image shows small airways disease–associated nodules in the right lung of a patient with tuberculosis. The nodules are well defined and patchy without pleural or fissural nodularity; the algorithm enables localization of small airways disease–associated nodules. However, at the lobular level, nodules (arrows) lie along the course of arterial branches identical to the lobular distribution in a. The nodules along the vessels in a and b are specifically and accurately localized and are distinguished from one another by using the algorithm.

View larger version (97K): [in this window] [in a new window]   Figure 11b. Algorithm utility in comparison with lobular-level nodule localization. (a) Thin-section CT image shows innumerable diffuse, highly attenuating, well-defined nodules with apparent pleural nodularity in this patient with metastatic adenocarcinoma of the breast. Some nodules (arrows) lie along the course of vessels and therefore could be considered intralobular or centrilobular at the level of the secondary lobule. (b) Thin-section CT image shows small airways disease–associated nodules in the right lung of a patient with tuberculosis. The nodules are well defined and patchy without pleural or fissural nodularity; the algorithm enables localization of small airways disease–associated nodules. However, at the lobular level, nodules (arrows) lie along the course of arterial branches identical to the lobular distribution in a. The nodules along the vessels in a and b are specifically and accurately localized and are distinguished from one another by using the algorithm.

The only errors of importance that were potentially avoidable in our series involved confusion between perilymphatic (specifically, sarcoidosis) and small airways disease-associated nodules in cases of indeterminate pleural or fissural profusion. We recommend close assessment of nodule morphology (see earlier) and correlation with clinical parameters and other imaging findings to avoid this potential problem.

All of our observers were experienced in thin-section CT interpretation. We chose to assess the algorithm itself by using experienced thin-section CT observers rather than to introduce the variable effects of observer experience into the current study; given the simplicity of thin-section CT interpretation with the algorithmic approach compared with thin-section CT interpretation in its absence, we expect that the algorithm will be beneficial.

Correct nodule localization can lead to a specific diagnosis when integrated with pertinent clinical information and associated imaging findings (cysts, lymph node enlargement, pleural disease, etc); localization also immediately affects patient treatment. Continued application and evaluation of the algorithm in clinical practice appears warranted.

	Footnotes

 
Address reprint requests to J.F.G.

Author contributions: Guarantor of integrity of entire study, J.F.G.; study concepts, J.F.G., W.R.W., D.P.N.; study design, J.F.G.; definition of intellectual content, J.F.G., W.R.W., D.P.N.; literature research, J.F.G., D.P.N.; clinical studies, J.F.G., W.R.W., D.P.N., G.M.; data acquisition, J.F.G., W.R.W., D.P.N., G.M.; data analysis, J.F.G.; manuscript preparation, J.F.G.; manuscript editing, W.R.W., D.P.N.; manuscript review, J.F.G., W.R.W., D.P.N., G.M.

Received June 26, 1998; revision requested August 5, 1998; revision received September 3, 1998; accepted October 6, 1998.

	References

TOP Abstract Introduction MATERIALS AND METHODS RESULTS DISCUSSION References

 

1. Heitzman ER, Markarian B, Berger I, Dailey E. The secondary pulmonary lobule: a practical concept for interpretation of radiographs. I. Roentgen anatomy of the normal secondary pulmonary lobule. Radiology 1969; 93:508-513.
2. Murata K, Khan A, Herman PG. Pulmonary parenchymal disease: evaluation with high-resolution CT. Radiology 1989; 170:629-635.[Abstract/Free Full Text]
3. Weibel ER, Taylor CR. Design and structure of the human lung. In: Weibel ER, Taylor CR, eds. Pulmonary diseases and disorders. New York, NY: McGraw-Hill, 1988; 11-60.
4. Bergin C, Roggli V, Coblentz C, Chiles C. Secondary pulmonary lobule: normal and abnormal CT appearances. AJR 1988; 151:21-25.[Abstract/Free Full Text]
5. Webb WR, Müller NL, Naidich DP. High-resolution CT of the lung 2nd ed. Philadelphia, Pa: Lippincott-Raven, 1996.
6. Gruden JF, McGuinness G. Subjective pitfalls in HRCT interpretation. Crit Rev Diagn Imaging 1996; 37:349-434.[Medline]
7. Colby TV, Swensen SJ. Anatomic distribution and histopathologic patterns in diffuse lung disease: correlation with HRCT. J Thorac Imaging 1996; 11:1-26.[Medline]
8. Gruden JF, Naidich DP. High-resolution CT: can it obviate lung biopsy?. Clin Pulm Med 1998; 5:23-35.
9. Murata K, Itoh H, Todo G, et al. Centrilobular lesions of the lung: demonstration by high-resolution CT and pathologic correlation. Radiology 1986; 161:641-645.[Abstract/Free Full Text]
10. Gruden JF, Webb WR. Centrilobular opacities on HRCT: diagnosis and differential diagnosis with pathologic correlation. AJR 1994; 162:569-574.[Abstract/Free Full Text]
11. Akira M, Kitatani F, Lee YS, et al. Diffuse panbronchiolitis: evaluation with high-resolution CT. Radiology 1988; 168:433-438.[Abstract/Free Full Text]
12. Aquino SL, Gamsu G, Webb WR, Kee ST. Tree-in-bud pattern: frequency and significance on thin-section CT. J Comput Assist Tomogr 1996; 20:594-599.[Medline]
13. Gruden JF, Webb WR. Identification and evaluation of centrilobular opacities on high-resolution CT. Semin Ultrasound CT MR 1995; 16:435-449.[Medline]
14. Itoh H, Tokunaga S, Asamoto H, et al. Radiologic-pathologic correlations of small lung nodules with special reference to peribronchiolar nodules. AJR 1978; 130:223-231.[Abstract]
15. Remy-Jardin M, Beuscart R, Sault MC, Marquette CH, Remy J. Subpleural micronodules in diffuse infiltrative lung diseases: evaluation with thin-section CT scans. Radiology 1990; 177:133-139.[Abstract/Free Full Text]
16. Nishimura K, Itoh H, Kitaichi M, Nagai S, Izumi T. Pulmonary sarcoidosis: correlation of CT and histopathologic findings. Radiology 1993; 189:105-109.[Abstract/Free Full Text]
17. Brauner MW, Grenier P, Mompoint D, Lenoir S, de Crémoux H. Pulmonary sarcoidosis: evaluation with high-resolution CT. Radiology 1989; 172:467-471.[Abstract/Free Full Text]
18. Müller NL, Kullnig P, Miller RR. The CT findings of pulmonary sarcoidosis: analysis of 25 patients. AJR 1989; 152:1179-1182.[Abstract/Free Full Text]
19. Murata K, Takahashi M, Mori M, et al. Pulmonary metastatic nodules: CT-pathologic correlation. Radiology 1992; 182:331-335.[Abstract/Free Full Text]
20. Hirakata K, Nakata H, Haratake J. Appearance of pulmonary metastases on high-resolution CT scans: comparison with histopathologic findings from autopsy specimens. AJR 1993; 161:37-43.[Abstract/Free Full Text]
21. Ren HR, Hruban RH, Kuhlman JE, et al. Computed tomography of inflation-fixed lungs: the beaded septum sign of pulmonary metastases. J Comput Assist Tomogr 1989; 13:411-416.[Medline]
22. Munk PL, Müller NL, Miller RR, Ostrow DN. Pulmonary lymphangitic carcinomatosis: CT and pathologic findings. Radiology 1988; 166:705-709.[Abstract/Free Full Text]
23. McGuinness G, Naidich DP, Jagardar J, Leitman B, McCauley DI. High resolution CT findings in miliary lung disease. J Comput Assist Tomogr 1992; 16:384-390.[Medline]
24. Oh YW, Kim YH, Lee NJ, et al. High-resolution CT appearance of miliary tuberculosis. J Comput Assist Tomogr 1994; 18:862-866.[Medline]
25. Lynch DA, Rose CS, Way D, et al. Hypersensitivity pneumonitis: sensitivity of high-resolution CT in a population-based study. AJR 1992; 159:469-472.[Abstract/Free Full Text]
26. Moore AD, Godwin JD, Müller NL, et al. Pulmonary histiocytosis X: comparison of radiographic and CT findings. Radiology 1989; 172:249-254.[Abstract/Free Full Text]
27. Brauner MW, Grenier P, Moelhi M, et al. Pulmonary histiocytosis X: evaluation with high-resolution CT. Radiology 1989; 172:255-258.[Abstract/Free Full Text]
28. Gruden JF, Webb WR. CT findings in a proved case of respiratory bronchiolitis. AJR 1993; 161:44-46.[Free Full Text]
29. McGuinness G, Gruden JF, Naidich DP, Jagardar J, Harkin T, Bhalla M. AIDS-related airways disease. AJR 1997; 168:67-77.[Abstract/Free Full Text]
30. Gruden JF, Huang L, Turner J, et al. High-resolution CT in the evaluation of clinically suspected Pneumocytis carinii pneumonia in AIDS patients with normal, equivocal, or nonspecific radiographic findings. AJR 1997; 169:967-975.[Abstract/Free Full Text]
31. Kim WS, Im JG, Chung EC, et al. Hematogenous pulmonary metastases: an experimental study using VX-2 carcinoma in rabbits. Acta Radiol 1993; 34:581-585.[Medline]
32. Collins J, Blankenbaker D, Stern EJ. CT patterns of bronchiolar disease: what is "tree-in-bud"?. AJR 1998; 171:365-370.[Free Full Text]

This article has been cited by other articles:

				D. M. Systrom and C. Wittram Case 9-2005 - A 67-Year-Old Man with Acute Respiratory Failure N. Engl. J. Med., March 24, 2005; 352(12): 1238 - 1246. [Full Text] [PDF]

				J.-N. Heo, Y. W. Choi, S. C. Jeon, and C. K. Park Pulmonary Tuberculosis: Another Disease Showing Clusters of Small Nodules Am. J. Roentgenol., February 1, 2005; 184(2): 639 - 642. [Abstract] [Full Text] [PDF]

				E. L. Gasparetto, D. L. Escuissato, E. Marchiori, S. Ono, R. L. F. e Silva, and N. L. Muller High-Resolution CT Findings of Respiratory Syncytial Virus Pneumonia After Bone Marrow Transplantation Am. J. Roentgenol., May 1, 2004; 182(5): 1133 - 1137. [Abstract] [Full Text] [PDF]

				N. L. Muller, G. C. Ooi, P. L. Khong, L. J. Zhou, K. W. T. Tsang, and S. Nicolaou High-Resolution CT Findings of Severe Acute Respiratory Syndrome at Presentation and After Admission Am. J. Roentgenol., January 1, 2004; 182(1): 39 - 44. [Abstract] [Full Text] [PDF]

				A. Oikonomou, N. L. Muller, and S. Nantel Radiographic and High-Resolution CT Findings of Influenza Virus Pneumonia in Patients with Hematologic Malignancies Am. J. Roentgenol., August 1, 2003; 181(2): 507 - 511. [Abstract] [Full Text] [PDF]

				D. Tack, M.-C. Nollevaux, and P. A. Gevenois Tree-in-Bud Pattern in Neoplastic Pulmonary Emboli Am. J. Roentgenol., June 1, 2001; 176(6): 1421 - 1422. [Full Text] [PDF]

				P. Reittner, N. L. Muller, L. Heyneman, T. Johkoh, J. S. Park, K. S. Lee, O. Honda, and N. Tomiyama Mycoplasma pneumoniae Pneumonia: Radiographic and High-Resolution CT Features in 28 Patients Am. J. Roentgenol., January 1, 2000; 174(1): 37 - 41. [Abstract] [Full Text] [PDF]

This Article