|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
ABSTRACT |
|---|
TOP
ABSTRACT
INTRODUCTION
METHODS
RESULTS
DISCUSSION
REFERENCES
|
|---|
A texture-based adaptive multiple feature method (AMFM) for evaluating pulmonary parenchyma from computed tomography (CT) images is described. This method incorporates multiple statistical and fractal texture features. The AMFM was compared to two previously published methods, namely, mean lung density (MLD) and the lowest fifth percentile of the histogram (HIST). First, the ability of these methods to detect subtle differences in ventral-dorsal lung density gradient in the prone normal lung was studied. Second, their abilities to differentiate between normal and emphysematous whole lung slices were compared. Finally, regional analyses comparing normal and emphysematous regions were performed by dividing the lungs in the CT slices into six equal regions, ventral to dorsal, and analyzing each region separately. The results demonstrated that the AMFM could separate the ventral from the dorsal one-third of the normal prone lung with 89.8% accuracy, compared to an accuracy of 74.6% with the MLD and 64.4% with the HIST methods. The normal and emphysematous slices were separated on a global basis with 100.0% accuracy using the AMFM as compared to an accuracy of 94.7% and 97.4% using the MLD and HIST methods, respectively. The regional normal and emphysematous tissues were discriminated with an average accuracy of 97.9%, 89.9%, and 99.1% with the AMFM, MLD, and HIST methods, respectively. The three methods and the pulmonary function tests in the normal and emphysema groups were poorly correlated. Quantitative texture analysis using adaptive multiple features holds promise for the objective noninvasive evaluation of the pulmonary parenchyma.
| |
INTRODUCTION |
|---|
|
TOP
ABSTRACT
INTRODUCTION
METHODS
RESULTS
DISCUSSION
REFERENCES
|
|---|
Pulmonary emphysema is defined as a lung disease characterized by "abnormal enlargement of the air spaces distal to the terminal, nonrespiratory bronchiole, accompanied by destructive changes of the alveolar walls" (1). These lung parenchymal changes are pathognomonic for emphysema. Detection of such changes during life, however, is difficult. Early and accurate diagnosis of emphysema is important for smoking cessation advice, evaluating the natural history of the disease, and disease phenotyping as we improve our understanding of disease processes and therapeutic interventions. Generally, the diagnosis of emphysema is based on indirect features, such as clinical examination, pulmonary function tests, and subjective visual evaluation of computed tomography (CT) scans. These tests are of limited value in assessing mild to moderate emphysema.
Computed tomography demonstrates pathological changes
in the lung parenchyma and has shown to be more sensitive
than the chest radiograph in detecting alterations in lung parenchyma, including changes due to emphysema (2). Two
basic approaches to detecting and quantifying emphysema
from CT scans have been used. One approach involves assigning a grade or a rating to assess the presence of emphysema by
visual examination of the hard copy scans (2, 3, 6, 7). In this
approach, the visual assessments are compared with the subsequent pathological examination for emphysema, and the results have demonstrated a good correlation. However, such
visual evaluations are time-consuming and limited by a wide range of interobserver variability and lack of sensitivity to early disease. The second approach directly analyzes digital
data obtained from the CT scan (4, 5, 8). This approach is objective and, therefore, not subject to interpreter bias. There are at least two computerized methods of identifying emphysema currently in use. One technique identifies areas of low
attenuation (a characteristic of emphysema) based on a single
density index threshold or a range of density indices (X-ray attenuation coefficients reported as Hounsfield Units [HU] or
electromagnetic imaging [EMI] units). Here, all areas having
densities lower than the threshold (e.g., 
910 HU or the lowest fifth percentile of the histogram) or falling within a given
range of densities are considered to be emphysematous. Additionally, the lowest fifth percentile of the histograms of emphysematous subjects has been shown to correlate well with
the surface area of walls of distal air spaces per unit lung volume, or AWUV (4). The second method computes the mean lung density as a defining characteristic of emphysema. These studies have reported good correlation with some pulmonary
function tests, such as DLCO and lung pathology.
Because of partial volume effects, when lung parenchymal regions affected by the pathologic process are smaller and more dispersed amongst normal parenchyma, simple density-based features become insensitive. Thus, while the use of these lung density masks (5) has moved CT evaluation into an objective, quantitative arena, the utility of such approaches is limited. Improved methodology based upon the underlying pattern of disease is one possible way of improving quantitative CT evaluation. In this study, we examined the use of an automated texture-based adaptive multiple feature method (AMFM) and compared it to two previously published measures of emphysema, namely mean lung density (MLD) and the lowest fifth percentile of the density histogram (HIST).
| |
METHODS |
|---|
|
TOP
ABSTRACT
INTRODUCTION
METHODS
RESULTS
DISCUSSION
REFERENCES
|
|---|
Subjects
We studied nine normal volunteers and ten patients with emphysema. The normal volunteers had no history of pulmonary disease, were lifelong nonsmokers, were taking no medications, and had no evidence of lung disease. Normal data were acquired as part of a previous study to evaluate interstitial pulmonary fibrosis. The subjects with emphysema were from a group of patients who were being evaluated for potential lung reduction surgery. These subjects had severe emphysema, and the CT scans were performed as part of their preoperative evaluation for lung reduction. All subjects had a CT scan and pulmonary function tests in close temporal proximity.
Computed Tomography Data Acquisition
Computed tomography scans were acquired with an electron beam CT scanner (Imatron Fastrac C-150 XL, South San Francisco, CA) at the University of Iowa Hospitals and Clinics. Scans of the normal subjects were acquired in the prone position, and the scans of the subjects with emphysema were acquired in the supine position. It has long been recognized that the lung is more uniformly expanded in the prone posture, and thus this posture would be expected to allow for better characterization of the lung field. In the subjects with severe emphysema, however, it was considered too difficult to perform prone scans. All subjects were scanned during breath-holding at maximal inspiration. A standardized protocol of the same slice thickness and a similar field of view was used. The field of view was 300- 400 mm, and collimation was 3 mm. The images were reconstructed to 512 × 512 pixels. The grey level resolution of the images was 11 bit.
Pulmonary Function Tests
The pulmonary function tests (PFT) consisted of standard spirometry using the Medical Graphics 1070 system (Medical Graphics, St. Paul, MN) and lung volumes via plethysmography using a Medical Graphics 1085 system. Single-breath diffusing capacity (DLCO) was tested using the Medical Graphics 1070 system. The measurements of lung function were performed using standard protocols, and the American Thoracic Society guidelines were used to determine acceptability (12). The predicted normal values used were those of Morris and co-workers (13) for spirometry, Goldman and Becklake (14) for lung volumes, and Van Ganse and colleagues (15) for the diffusing capacity.
Computed Tomography Data Analysis
The AMFM for examining the lung parenchyma from CT scans uses a combination of statistical (16) and fractal approaches. The combination of all these features provides a description of the texture as well as structure in the regions of interest (ROI). This method was compared with the MLD and the HIST methods for evaluating the normal lung as well as differentiating between the normal and emphysematous parenchyma. Three experiments were designed for this purpose.
1. Adaptive Multiple Feature Method
The steps involved in this method are outlined in Figure 1. Each CT
slice was first segmented (17) to remove nonlung structures (Figure
2). Preprocessing on the original image was performed using edgmentation (18, 19) for computing some of the texture features. Edgmentation is a region-growing technique that merges adjacent pixels to form
regions where the difference between the grey levels of the adjacent
pixels is small. The grey level assigned to a merged region is the average of the grey levels of the pixels forming the region. Figures 3 and 4
show examples of the edgmented image of a normal and an emphysematous CT slice, respectively. The ROI was then defined on the original image as well as on the preprocessed image. Feature extraction
was performed on the ROIs. Five first-order texture features adapted
from Ferdeghini and colleagues (20) and the geometric fractal dimension (21) were computed on the ROI of the original image. The first-order features were mean, variance, skewness, kurtosis, and grey-level
entropy. Eleven second-order features adapted from Fleagle and colleagues (22) were computed on the ROI from the preprocessed image. Five of the second-order features were run-length features: short-run emphasis, long-run emphasis, grey-level nonuniformity, run-length nonuniformity, and run percentage. The remaining six features were
based on the co-occurrence matrix: angular second moment, entropy,
inertia, contrast, correlation, and inverse difference moment. All the
features were normalized for pixel sizes and for the size of the lung in
the CT image. The available ROIs from the CT slices were then split
randomly into two distinct sets
the training set and the test set. The
training set provided example regions for the available classes, and
the classes were chosen differently for each experiment. The optimal set
of features was selected using the "divergence" measure along with
correlation analysis (23). Optimal feature selection was performed using the ROIs in the training set. Finally, classification was performed using the Bayesian classifier (24). There are two phases to the classification process. In the training phase, classifier parameters are determined from the optimal features extracted on the ROIs in the training
set. Finally, in the testing phase, the ROIs in the test set are classified
based on the parameters determined in the training phase.
|
|
|
|
2. Mean Lung Density Method
The mean lung density extracted from the ROI of the original image, as a feature by itself, was used to study parenchymal changes in normal and emphysematous subjects. The MLD is equivalent to the mean extracted as a first-order feature. Figure 5 shows the steps involved in the MLD method. The classification step was similar to that in the AMFM.
|
3. Histogram Analysis
The lowest fifth percentile of the histogram of the original image ROI was computed. This percentile was used to classify the test ROIs (Figure 6). Again, the classification step was similar to that in the AMFM.
|
These three methods were compared in three different experiments. Four slices from each CT scan were included in the experiments. By an arbitrary convention, two of these slices were from the region at the carina and the remaining two were from the region halfway between the carina and the base of the lung.
The first experiment involved the evaluation of the lung parenchyma in nine normal subjects. It has long been recognized that there is lung density variation from dependent to nondependent regions of the lung (25). To examine this dependence, each lung in the CT image slice was divided into three equal parts, anterior to posterior (Figure 7A). The posterior one-third and the anterior one-third regions were defined as the ROIs and served as the available classes. Regions of interest passing directly through the horizontal fissure line or blurry due to movement were removed from the analysis. Classifier training used samples of ROIs from these two regions. The classifier was then applied to the test set and each test sample was identified as being either anterior one-third or posterior one-third. The rate of correct classification (percentage of samples correctly classified in each class) and accuracy (percentage of correctly classified samples in both classes) were calculated.
|
In the second experiment, we compared the normal lung images with the images obtained from subjects with emphysema. Global analysis to assess the overall normality and emphysema in the lung was performed with the entire image slice as the ROI to extract the texture features. Nine normal subjects and ten patients with emphysema were used in this experiment. The available samples were split randomly into the training set and test set. Training was performed on ROIs of normal and emphysematous lungs from the training set. Therefore, the two possible outputs of the classifier when analyzing the test set were normal and emphysematous. The sensitivity, specificity (26) and accuracy of the results were calculated. "Sensitivity" was defined as the percentage of regions with disease correctly classified. "Specificity" was defined as the percentage of normal samples correctly classified. "Accuracy" was defined as the percentage of correctly classified samples, taking both the classes into consideration.
Finally, in our third experiment, regional comparisons of normal and emphysematous lungs were performed. Each lung in the CT slice was split into six equal parts, anterior to posterior (Figure 7B). The samples were blindly reviewed by a trained pulmonologist experienced in CT scan assessment of parenchymal lung disease and only those CT regions being definitely normal from normal subjects and having definite emphysema from subjects with emphysema were retained. On two blinded readings, 4 mo apart, the accuracy of the pulmonologist in selecting or removing regions in normal and emphysematous lungs was 90.0% and 76.4%, respectively. In each of the six parts, the classifier was trained using samples from normal and emphysematous lungs. The classifier was then asked to identify samples from the test set as being normal or emphysematous. This process was repeated for all six regions. The sensitivity, specificity, and accuracy were calculated in each of the six regions.
Statistical Analysis
To compare with the PFTs, features obtained from the entire CT slice
were used. Regression analysis was performed between the average
features obtained from the four CT slices (global analysis) and the
PFTs, in the normal and emphysema groups separately. All three
methodsAMFM, MLD, and HISTwere analyzed. Multiple regression analysis was performed between the best features chosen by
the AMFM for the global analysis and the PFT parameters. Simple regression analysis was used when comparing the PFT parameters with
the MLD and HIST methods. StatView 4.1 (Abacus Concepts, Inc.,
Berkeley, CA) was used for the regression analyses. Analysis of variance (ANOVA) was used to test the significance of the linear regression using the same package. Multiple comparisons were not accounted for in ANOVA.
| |
RESULTS |
|---|
|
TOP
ABSTRACT
INTRODUCTION
METHODS
RESULTS
DISCUSSION
REFERENCES
|
|---|
In the first experiment, normal subjects were studied by evaluating the anterior and posterior thirds using the AMFM, MLD, and HIST methods. The optimal set of features for the AMFM were contrast, variance, geometric fractal dimension, mean, skewness, correlation, and grey level nonuniformity. Table 1 summarizes the ability of each of the three methods to detect differences in density between the anterior and posterior portions of the normal lung.
|
In the second experiment, global analysis was performed to discriminate entire CT slices obtained from normal and emphysematous lungs. The best combination of features selected automatically for the AMFM were short-run emphasis, mean, and grey level nonuniformity. Again, MLD and HIST methods were also applied and compared with AMFM. Table 2 shows the results of classification obtained using all three methods for global analysis.
|
In the last experiment, the scans obtained from the normal and emphysematous lungs were split into six regions anterior to posterior. In each of the six regions, normal and emphysematous regions were discriminated. As before, AMFM, MLD, and HIST methods were applied and compared. Table 3 summarizes the results of regional classification obtained for the classes, normal and emphysematous, for each of the six regions. The best combination of features that were automatically selected for each of the six regions for the AMFM are summarized in Table 4.
|
|
The means and one standard deviation of the PFT parameters are presented in Table 5. These PFT results (total lung capacity, or TLC [actual and percent predicted], FEV1 [actual and percent predicted], FEV1/FVC, PaO2 [actual], DLCO [actual and percent predicted], and residual volume, or RV [actual and percent predicted]) and the AMFM, MLD, and HIST methods were compared using regression analysis. The regression analyses were performed separately for the normal group and the group with emphysema. Multiple regression analysis was used to compare the PFT parameters with the optimal combination of features determined by the AMFM, i.e., short-run emphasis, mean, and grey level nonuniformity. Simple regression analysis was used to compare PFT parameters with MLD and HIST. The results are tabulated in Table 6. Total lung capacity (actual) and RV (actual) significantly correlated with the AMFM in the group with emphysema. The FEV1/ FVC ratio significantly correlated with MLD in the group with emphysema. None of the actual PFT parameters correlated with HIST in either group. In addition, AMFM and MLD did not show significant correlations with any of the actual PFT parameters in the normal group. For the percent-predicted PFT parameters, AMFM, MLD and HIST showed significant correlations with TLC in the normal group. No other significant correlations were observed.
|
|
| |
DISCUSSION |
|---|
|
TOP
ABSTRACT
INTRODUCTION
METHODS
RESULTS
DISCUSSION
REFERENCES
|
|---|
This study is a part of our long-term project of developing a generalizable, completely automated method that can identify specific regions of pulmonary parenchymal abnormality and objectively assess the severity of disease in different locations within the lung. The AMFM incorporates first-order and second-order statistical texture features and the geometric fractal dimension. The first-order features are derived from the grey-scale frequency histograms. The second-order features describe the relationships between pixels and their grey levels. In addition to statistical features, a feature to quantify texture is used: the fractal dimension. It has been shown previously (27) that images of natural surfaces can be modeled using the fractional Brownian motion model, and the fractal dimension can provide a feature of a surface's roughness or smoothness. The use of fractals to describe medical images is increasing (28).
The MLD has been previously shown to be significantly lower in subjects with emphysema than in normal volunteers (31). It also has been previously demonstrated that the histograms of emphysematous lungs are skewed more toward low densities than are normal lungs (4, 32). In this study, the utility of the AMFM is tested against these older methods for detecting emphysema.
As a step in validating our approach, particularly in view of difficulty in finding gold standards, we evaluated nine normal subjects and ten subjects with well-defined emphysema. Initially, to apply and understand the AMFM, we examined the lung textural changes associated with position in normal subjects. It has been shown previously that lung density changes due to gravity are minimized when the scanning is performed with the subject in prone position (25). We demonstrated that the AMFM can detect ventral-dorsal differences using an automatically selected set of seven features. The AMFM separated the ventral and dorsal regions with 89.8% accuracy. This suggests that, despite the fact that the lung is more uniformly expanded in the prone position (25), subtle gravity-oriented changes still occur, and the AMFM is able to discriminate this. The ventral-dorsal texture differences are not as well detected by MLD and HIST methods, which have accuracies of 74.6% and 64.4%, respectively. This finding indicates that the AMFM is better than single-feature methods in detecting this type of textural difference in the lung.
We next studied a group of subjects with emphysema, examining the texture of the entire CT slice to determine the presence or absence of emphysema. In this study, the AMFM was capable of this assessment with an accuracy of 100.0%, using an optimal set of three features. The sensitivity and specificity of AMFM were both found to be 100%. Using MLD alone, the detection of normal and emphysematous scans yielded an accuracy of 94.7%. The sensitivity of the MLD method was 95.0% and the specificity was 94.4%. The HIST method demonstrated an accuracy of 97.4% in determining the presence of emphysema, with sensitivity of 95.0% and specificity of 100.0%. The AMFM, therefore, compares very well to previous methods of discriminating normal from emphysematous lungs. It should be noted that the results of the AMFM are dependent only on the initial grouping of the subjects as normal or emphysematous. They are not dependent in any way on the CT scan assessment by the observer. The analysis includes all of the CT slice information.
We then examined focal regions of interest in the lung. Regional analysis of the normal and emphysema scans was performed by splitting each lung in each slice into six regions. This provided a reasonable compromise between too small a region, which would not capture texture, and too large a region. Because the normal subjects were scanned prone and subjects with emphysema were scanned supine, it was necessary to align the scans anatomically before splitting the lung into regions. Some regions were removed from the analysis either because the CT region contained artifacts, such as movement, or was not classifiable as either emphysematous or normal because of other disease processes. The emphysematous and normal samples were successfully discriminated in each of the six regions with an average accuracy of 97.9% using the AMFM. The average sensitivity and average specificity were 94.8% and 100.0%, respectively. The average sensitivity and specificity using the MLD method obtained over the same six regions was 88.9% and 90.5%, respectively. The MLD accuracy was 89.9%. Therefore, the AMFM performed better than the MLD method. Classification results for regional analysis using the HIST method demonstrated an average sensitivity of 97.7% and average specificity of 100.0% over all six regions. The average accuracy of 99.1% for the HIST method shows that is comparable to the AMFM for this kind of classification.
In this analysis, three to eight features were selected for classification by the AMFM in each region. Overall, thirteen different features were automatically selected. (The four features that did not contribute to the classification at any time were angular second moment, inverse difference moment, long-run emphasis, and entropy.) This adaptability demonstrates that the AMFM can recognize local textures in different regions of the lung and still retain discriminatory power.
Pulmonary function tests are currently used to indicate the severity of emphysema. Regression analysis between the PFT parameters and the AMFM, MLD, and HIST methods showed no correlation for the important PFT markers of emphysema such as DLCO, and FEV1. One explanation for this is that all three methods detect emphysema by measuring parameters that may be independent of airway function and gas exchange. Also, the lung function tests evaluate all of the lung, whereas the CT analysis is restricted to four slices per subject. Another explanation is that our data set included only normal subjects and patients with advanced emphysema; these may not, as groups, have enough variation to indicate significant correlation within the group.
The AMFM compares very favorably with previously used
methodsMLD and HISTin the detection of emphysema.
Several features in the AMFM show significant abnormalities
in emphysema when compared to the normal lung. These features may be useful for objectively describing the lung texture
associated with emphysema and may, in the future, be shown
to vary with time or therapy. It is also important that these
features can be measured in regional parts of the lung, as opposed to PFT's evaluation of the whole lung function. The
AMFM, with its automatic selection of different features, is
better able to discriminate positional changes in a normal lung
than either the MLD or the HIST method. This finding is encouraging and suggests that the AMFM is more sensitive to textural changes of increased radiopacity. It also suggests that the AMFM may be more generally applicable for other parenchymal lung diseases. The AMFM automatically picked different features to differentiate the ventral-dorsal lung density
gradient, to discriminate normal and emphysematous slices,
and to discriminate regions. This implies that the AMFM is
adaptable to different lung patterns.
We conclude that the AMFM can be used for tissue characterization of the lung. This method provides multiple objective features and therefore allows objective, consistent evaluation of parenchymal lung disease. In the future, a larger number of subjects with varying degrees of emphysema will be investigated. Furthermore, improvements in the classification scheme will be incorporated and this methodology will be applied to study other forms of parenchymal lung disease, such as idiopathic pulmonary fibrosis and sarcoidosis.
| |
Footnotes |
|---|
Correspondence and requests for reprints should be addressed to Geoffrey McLennan, Department of Internal Medicine, The University of Iowa, Iowa City, IA, 52242.
(Received in original form June 24, 1996 and in revised form March 14, 1997).
The work was supported in part by an award from the Whitaker Foundation, a contract from the National Library of Medicine (N01-LM-4-3511 US PHS), and a Career Investigator Award from the American Lung Association.Acknowledgments: The authors would like to thank Dr. Leon F. Burmeister, Professor, Department of Preventive Medicine and Environmental Health, The University of Iowa, for his help regarding statistical analysis of their results. They would also like to acknowledge the helpful suggestions in manuscript preparation by Dr. Gary W. Hunninghake, Professor, Department of Internal Medicine, The University of Iowa.
| |
References |
|---|
|
TOP
ABSTRACT
INTRODUCTION
METHODS
RESULTS
DISCUSSION
REFERENCES
|
|---|
1. American Thoracic Society. 1962. Chronic bronchitis, asthma, and pulmonary emphysema: a statement by the committee on diagnostic standards for nontuberculous respiratory diseases. Am. Rev. Respir. Dis. 85: 762-768 .
2.
Foster, W. L. Jr.,
P. C. Pratt,
V. L. Roggli,
J. D. Godwin,
R. A. Halvorsen Jr., and
C. E. Putman.
1986.
Centrilobular emphysema: CT-pathogenic correlation.
Radiology
159:
27-32
3. Bergin, C. J., N. L. Muller, and R. R. Miller. 1986. CT in the quantitative assessment of emphysema. J. Thorac. Imaging 1: 94-103 [Medline].
4.
Gould, G. A.,
W. MacNee,
A. McLean,
P. M. Warren,
A. Redpath,
J. J. Best,
D. Lamb, and
D. C. Flenley.
1988.
CT measurements of lung density in life can quantitate distal airspace enlargementan essential
defining feature of human emphysema.
Am. Rev. Respir. Dis.
137:
380-392
[Medline].
5.
Muller, N. L.,
C. A. Staples,
R. R. Miller, and
R. T. Abboud.
1988.
Density mask: an objective method to quantitate emphysema using computed tomography.
Chest
94:
782-787
6.
Kuwano, K.,
K. Matsuba,
T. Ikeda,
J. Murakami,
A. Araki,
H. Nishitani,
T. Ishida,
K. Yasumoto, and
N. Shigematsu.
1990.
The diagnosis of
mild emphysemacorrelation of computed tomography and pathology scores.
Am. Rev. Respir. Dis.
141:
169-178
[Medline].
7. Dillon, T. J., R. L. Walsh, R. Scicchitano, B. Eckert, E. G. Cleary, and G. McLennan. 1992. Plasma elastin derived peptide levels in normal adults, children and emphysematous subjects: physiological and CT scan correlates. Am. Rev. Respir. Dis. 146: 1143-1148 [Medline].
8.
Sakai, N.,
M. Mishima,
K. Nishimura,
H. Itoh, and
K. Kuno.
1994.
An automated method to assess the distribution of low attenuation areas
on chest CT scans in chronic pulmonary emphysema patients.
Chest
106:
1319-1325
9.
Kinsella, M.,
N. L. Muller,
R. T. Abboud,
N. J. Morrison, and
A. DyBuncio.
1990.
A quantitation of emphysema by computed tomography using a `density mask' program and correlation with pulmonary
function tests.
Chest
97:
315-321
10.
Archer, D. C.,
C. L. Coblentz,
R. A. deKemp,
C. Nahmias, and
G. Norman.
1993.
Automated in vivo quantification of emphysema.
Radiology
188:
835-838
11.
Knudson, R. J.,
J. R. Standen,
W. T. Kaltenborn,
D. E. Knudson,
K. Rehm,
M. P. Habib, and
J. D. Newell.
1991.
Expiratory computed tomography for assessment of suspected pulmonary emphysema.
Chest
99:
1357-1366
12. American Thoracic Society. 1979. American Thoracic Society statement: Snowbird Workshop on Standardization of Spirometry. Am. Rev. Respir. Dis. 119: 831-838 [Medline].
13. Morris, J. F., A. Koski, and L. C. Johnson. 1971. Spirometric standard for healthy nonsmoking adults. Am. Rev. Respir. Dis. 103: 57-67 [Medline].
14. Goldman, H. I., and M. R. Becklake. 1959. Respiratory function tests: normal values at median altitudes and the prediction of normal results. Am. Rev. Tuberc. 79: 457-467 . [Medline]
15. Van Ganse, W. F., B. G. Ferris, and J. E. Cotes. 1972. Spirometric standard for healthy nonsmoking adults. Am. Rev. Respir. Dis. 105: 30-41 [Medline].
16. Haralick, R. M.. 1979. Statistical and structural approaches to texture. Proceedings IEEE 67: 786-804 .
17. Everhart, J. J. 1994. Automated segmentation of lung tissue from pulmonary CT slices [Master's thesis]. The University of Iowa, Iowa City, IA.
18. Cornelis, J., J. De Becker, M. Bister, C. Vanhove, G. Demonceau, and A. Cornelis. 1992. Techniques for cardiac image segmentation. In Proceedings of the 14th IEEE EMBS Conference, Vol. 14. IEEE, Piscataway, NJ. 1906-1908.
19. Bister, M., J. Cornelis, and Y. Taeymans. 1991. Towards automated analysis in 3-D cardiac MR imaging. In Information Processing in Medical Imaging: Lecture Notes in Computer Science. Springer Verlag, London.
20. Ferdeghini, E. M., B. Pinamonti, E. Picano, F. Lattanzi, R. Bussani, G. Slavich, A. Benassi, F. Camerini, L. Landini, and A. L'Abbate. 1991. Quantitative texture analysis in echocardiography: application to diagnosis of myocarditis. J. Clin. Ultrasound 19: 263-270 [Medline].
21. Uppaluri, R., T. Mitsa, and J. R. Galvin. 1995. Fractal analysis of high-resolution CT images as a tool for quantification of lung disease. In E. A. Hoffman, editor. Medical Imaging 1995: Physiology and Function from Multidimensional Images, Vol. 2433. SPIE, Bellingham, WA. 133-142.
22. Fleagle, S. R., W. Stanford, T. L. Burns, and D. J. Skorton. 1994. Feasibility of quantitative texture analysis of cardiac magnetic resonance imagery: preliminary results. In E. A. Hoffman and R. S. Acharya, editors. Medical Imaging 1994: Physiology and Function from Multidimensional Images, Vol. 2168. SPIE, Bellingham, WA. 23-32.
23. Andrews, H. C. 1972. Introduction to Mathematical Techniques in Pattern Recognition. Wiley, New York.
24. Sonka, M., V. Hlavac, and R. Boyle. 1993. Image Processing, Analysis and Machine Vision. Chapman & Hall, London.
25.
Hoffman, E. A..
1985.
Effect of body orientation on regional lung expansion: a computed tomographic approach.
J. Appl. Physiol.
59:
468-480
26. Dixon, W. J., and F. J. Massey, Jr. 1983. Introduction to Statistical Analysis. McGraw Hill, New York.
27. Pentland, A. P.. 1984. Fractal-based description of natural scenes. IEEE Trans Pattern Anal. and Machine Intelligence 6: 661-675 .
28. Chen, C., J. S. Daponte, and M. D. Fox. 1989. Fractal feature analysis and classification in medical imaging. IEEE Trans. Med. Imaging 8: 133-142 .
29. Wu, C. M., Y. C. Chen, and K. S. Hsieh. 1992. Texture features for classification of ultrasonic liver images. IEEE Trans. Med. Imaging 11: 141-152 .
30. Lundahl, T., W. J. Ohley, W. S. Kuklinski, D. O. Williams, H. Gewirtz, and A. S. Most. 1985. Analysis and interpolation of angiographic images by use of fractals. In IEEE Proceedings on Computers in Cardiology. IEEE, Los Alamitos, CA. 355-358.
31. Goddard, P. A., E. M. Nicholson, G. Laszlo, and I. Watt. 1982. Computed tomography in pulmonary emphysema. Clin. Radiol. 33: 379-387 [Medline].
32. Hayhurst, M. D., W. MacNee, D. C. Flenley, D. Wright, A. McLean, D. Lamb, A. J. A. Wightman, and J. Best. 1984. Diagnosis of pulmonary emphysema by computed tomography. Lancet 2: 320-322 [Medline].
This article has been cited by other articles:
 |
|
 |
|
  A. C. Best, J. Meng, A. M. Lynch, C. M. Bozic, D. Miller, G. K. Grunwald, and D. A. Lynch Idiopathic Pulmonary Fibrosis: Physiologic Tests, Quantitative CT Indexes, and CT Visual Scores as Predictors of Mortality Radiology, March 1, 2008; 246(3): 935 - 940. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. R. Froese, K. Ask, R. Labiris, T. Farncombe, D. Warburton, M. D. Inman, J. Gauldie, and M. Kolb Three-dimensional computed tomography imaging in an animal model of emphysema Eur. Respir. J., December 1, 2007; 30(6): 1082 - 1089. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. A. Hoffman, B. A. Simon, and G. McLennan State of the Art. A Structural and Functional Assessment of the Lung via Multidetector-Row Computed Tomography: Phenotyping Chronic Obstructive Pulmonary Disease Proceedings of the ATS, August 1, 2006; 3(6): 519 - 532. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. B. Fain, S. R. Panth, M. D. Evans, A. L. Wentland, J. H. Holmes, F. R. Korosec, M. J. O'Brien, H. Fountaine, and T. M. Grist Early Emphysematous Changes in Asymptomatic Smokers: Detection with 3He MR Imaging. Radiology, June 1, 2006; 239(3): 875 - 883. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. R. Boser, H. Park, S. F. Perry, M. G. Menache, and F. H. Y. Green Fractal Geometry of Airway Remodeling in Human Asthma Am. J. Respir. Crit. Care Med., October 1, 2005; 172(7): 817 - 823. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J.-M. Kuhnigk, V. Dicken, S. Zidowitz, L. Bornemann, B. Kuemmerlen, S. Krass, H.-O. Peitgen, S. Yuval, H.-H. Jend, W. S. Rau, et al. Informatics in Radiology (infoRAD): New Tools for Computer Assistance in Thoracic CT. Part 1. Functional Analysis of Lungs, Lung Lobes, and Bronchopulmonary Segments RadioGraphics, March 1, 2005; 25(2): 525 - 536. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 N J Screaton and T Koh Emphysema and smoking-related lung diseases Imaging, October 1, 2004; 16(1): 50 - 60. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
F. Chabat, G.-Z. Yang, and D. M. Hansell Obstructive Lung Diseases: Texture Classification for Differentiation at CT Radiology, September 1, 2003; 228(3): 871 - 877. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. C. Best, A. M. Lynch, C. M. Bozic, D. Miller, G. K. Grunwald, and D. A. Lynch Quantitative CT Indexes in Idiopathic Pulmonary Fibrosis: Relationship with Physiologic Impairment Radiology, August 1, 2003; 228(2): 407 - 414. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 N L Muller and H Coxson Chronic obstructive pulmonary disease * 4: Imaging the lungs in patients with chronic obstructive pulmonary disease Thorax, November 1, 2002; 57(11): 982 - 985. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. Suki Fluctuations and Power Laws in Pulmonary Physiology Am. J. Respir. Crit. Care Med., July 15, 2002; 166(2): 133 - 137. [Full Text] [PDF]
|
|
|
|
|
|
A. Madani, C. Keyzer, and P.A. Gevenois Quantitative computed tomography assessment of lung structure and function in pulmonary emphysema Eur. Respir. J., October 1, 2001; 18(4): 720 - 730. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. S. Gierada, R. D. Yusen, I. A. Villanueva, T. K. Pilgram, R. M. Slone, S. S. Lefrak, and J. D. Cooper Patient Selection for Lung Volume Reduction Surgery: An Objective Model Based on Prior Clinical Decisions and Quantitative CT Analysis Chest, April 1, 2000; 117(4): 991 - 998. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. Mishima, T. Hirai, H. Itoh, Y. Nakano, H. Sakai, S. Muro, K. Nishimura, Y. Oku, K. Chin, M. Ohi, et al. Complexity of terminal airspace geometry assessed by lung computed tomography in normal subjects and patients with chronic obstructive pulmonary disease PNAS, August 3, 1999; 96(16): 8829 - 8834. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. UPPALURI, E. A. HOFFMAN, M. SONKA, P. G. HARTLEY, G. W. HUNNINGHAKE, and G. MCLENNAN Computer Recognition of Regional Lung Disease Patterns Am. J. Respir. Crit. Care Med., August 1, 1999; 160(2): 648 - 654. [Abstract] [Full Text]
|
|
|
|
|
|
R. UPPALURI, E. A. HOFFMAN, M. SONKA, G. W. HUNNINGHAKE, and G. MCLENNAN Interstitial Lung Disease . A Quantitative Study Using the Adaptive Multiple Feature Method Am. J. Respir. Crit. Care Med., February 1, 1999; 159(2): 519 - 525. [Abstract] [Full Text]
|
|
|
|
|
|
R. THURNHEER, H. ENGEL, W. WEDER, U. STAMMBERGER, I. LAUBE, E. W. RUSSI, and K. E. BLOCH Role of Lung Perfusion Scintigraphy in Relation to Chest Computed Tomography and Pulmonary Function in the Evaluation of Candidates for Lung Volume Reduction Surgery Am. J. Respir. Crit. Care Med., January 1, 1999; 159(1): 301 - 310. [Abstract] [Full Text]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Proc. Am. Thorac. Soc. | Am. J. Respir. Cell Mol. Biol. |