| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 Department of Radiology, Osaka University Graduate School of Medicine, 2-2
Yamadaoka, Suita, Osaka, 565-0871, Japan.
2 Mitsubishi Space Software Co., Ltd., Fuji Techno-square, 5-4-36
Tsukaguchi-honmachi, Amagasaki, Hyogo 661-0001, Japan.
Received October 25, 2001;
accepted after revision February 11, 2002.
Address correspondence to M. Tsubamoto.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. Thirty healthy patients and 30 patients with new multifocal hazy pulmonary opacities that were confirmed by serial chest CT examinations were evaluated with and without temporal subtraction images. Chest radiographs were taken from either film-screen or digital radiography images and were digitized with a spatial resolution of 0.171 mm per pixel. Temporal subtraction images were produced by an iterative image-warping technique. We designed an observer performance study in which observers (six chest radiologists and four residents) indicated their confidence level for the presence or absence of hazy pulmonary opacities on two sets of images, current and previous radiographs only (set A), and current and previous radiographs with temporal subtraction images (set B). Receiver operating characteristic curves were generated.
RESULTS. For chest radiologists, observer performance with set B (with temporal subtraction images; Az = 0.947) was superior to that with set A (without temporal subtraction images; Az = 0.916) (p < 0.05). For residents, no statistically significant difference was found between sets A and B.
CONCLUSION. The temporal subtraction technique clearly improves diagnostic accuracy for the detection of multifocal hazy pulmonary opacities on chest radiographs, especially when the observers are experienced chest radiologists who have sufficient skill to evaluate the patient's condition as revealed on the images.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Recently, some researchers reported that a computer-assisted diagnosis (CAD) system that includes temporal subtraction images can improve the accuracy of diagnosis [3, 4]. However, to our knowledge, no study has evaluated the detection of multifocal hazy pulmonary opacities using the temporal subtraction method in relatively large numbers of cases. The purpose of this study was to evaluate the accuracy of temporal subtraction for the detection of multifocal hazy pulmonary opacities, which are sometimes difficult to detect directly on chest radiographs.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Previous and Current Chest Radiographs
Thirty consecutive chest radiographs with newly detected hazy pulmonary
opacities and 30 chest radiographs with normal findings were selected from the
daily case load referred for CT at our institution for use in the observer
test. Chest CT was performed using either a HiSpeed Advantage scanner (General
Electric Medical Systems, Milwaukee, WI) or an X-vigor scanner (Toshiba,
Tokyo, Japan) at 10-mm intervals and 10-mm collimation during breath-holding
after inspiration from the apex to the lung base. If an abnormal opacity was
detected, thin-section CT with 1-mm collimation was also performed. When the
opacity was ground-glass opacity and previous CT images with normal findings
for that patient existed, the case was included in this study as a case with
abnormal findings. Ground-glass opacity was defined as hazy increased
attenuation with preservation of bronchial and vascular markings
[5]. Any limitation of the
extent or the severity of the opacity was not noted. The chest radiographs
obtained on the date nearest each current and previous CT study were selected
as the current and previous radiographs in cases with abnormal findings.
CT was performed for healthy patients for clinical reasons such as screenings for metastasis and confirming some opacities on chest radiography, then all chest radiographs with normal findings were confirmed by CT. The following abnormal conditions were found: Pneumocystis carinii pneumonia (n = 13), hypersensitivity pneumonitis (n = 7), eosinophilic pneumonia (n = 3), drug-induced pneumonitis (n = 2), cytomegalovirus infection (n = 2), lung edema (n = 1), chronic eosinophilic pneumonia (n = 1), and cellular interstitial pneumonia (n = 1). The diagnosis of abnormal findings was made by transbronchial lung biopsy or by reaction to the treatment. The cases with normal findings were selected in reference to the CT findings.
The chest radiographs obtained on the most recent date were selected. The previous chest radiographs with normal findings were available for all cases of normal and abnormal findings. The abnormality and normality of both previous and current radiographs were confirmed by CT. High-resolution CT was performed if an abnormal opacity was detected on conventional CT. High-resolution CT was performed for all abnormal findings included in this study. However, we did not perform high-resolution CT for patients with normal findings; conventional CT was considered adequate to confirm the normality. The intervals between chest radiography and CT ranged from 0 to 2 days (average, 0.8 days).
Digital Subtraction System
Thirty cases with abnormal findings and 30 cases with normal findings, with
their respective comparison images that were taken with various film-screen or
digital radiography imaging, provided a total of 120 chest radiographs. All
were digitized with a spatial resolution of 0.171 mm per pixel and 4096 gray
levels by means of a Film Scanner 300 digitizer (Canon, Tokyo, Japan). Sixty
temporal subtraction images were generated from the image pair data by means
of an iterative image-warping technique
[6]. These digitized data were
transferred to a new commercially available CAD system (Truedia/XR; Mitsubishi
Space Software, Amagasaki, Japan). This CAD system was composed of an image
server PC (Pentium III with 550 MHz; RAM, 256 MB; hard drive, 27 GB) (Intel,
Santa Clara, CA) and a diagnostic workstation PC (Pentium III with 550 MHz;
RAM, 256 MB; and hard drive, 5 GB) (Intel) with three monitors that display
previous, current, and temporal subtraction images. The process of temporal
subtraction consisted of normalization of density and contrast images,
correction for lateral inclination by image rotation, rib cage edge detection
based on image profile analysis, smoothing of low-resolution images using a
gaussian filter (matrix size, 128 x 128), lung segmentation using the
rib cage edge, initial image matching (determination of global shift using
cross-correlation technique), nonlinear image warping, and image subtraction
[6,
7] (Fig.
1A,1B,1C,1D,1E).
Time needed was about 30 sec to digitize an image, 8 sec to send it to the
server, and 5 sec to make a subtraction image.
|
|
|
|
|
Interpretation of Chest Radiographs and Temporal Subtraction
Images
Six senior board-certified chest radiologists and four residents evaluated
both image sets—set A (current and previous radiographs only) and set B
(current and previous radiographs with temporal subtraction images)—on a
monitor equipped with this system. To reduce learning effects, we required
that at least 4 weeks elapse between each interpretation session. Before the
image interpretation, four training cases not included in the observer test
were distributed to the observers to familiarize them with the scoring system.
Observers recorded the presence or absence of hazy pulmonary opacities on a
monitor. They used a continuous rating scale (0-50), on which 0 indicated
complete confidence that the lung did not have new hazy pulmonary opacities
and 50 indicated complete confidence that the lung did have new hazy pulmonary
opacities [8,
9]. No time constraints were
placed on observers. No mention was made of the frequency of opacities that
were seen.
Statistical Analysis
The data were evaluated using LABROC 4 algorithm (Charles E. Metz, Chicago,
IL), which calculates two parameters, a and b. Parameter
a represents the maximum-likelihood estimates of the intercept and
slope of the ROC curve when it is plotted on normal deviate axes, and
parameter b represents the area under the ROC curve
(Az) [9].
The Az value of the ROC curve ranges from 0.5 to 1 and
increases when diagnostic performance approaches that of the gold standard (in
this case, the existence of new ground-glass shadow on CT). The statistical
significance of the difference of Az values of each
observer between set A and set B was evaluated using the Student's two-tailed
t test for paired data.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
|
|
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
This result has a paradoxical aspect. The absolute improvement of Az values with the use of the subtraction images was greater for residents (from 0.765 to 0.855; difference, 0.090) than for senior board-certified chest radiologists (from 0.916 to 0.947; difference, 0.031); however, only the difference for senior board-certified chest radiologists reached statistical significance. This paradox may be explained in two ways. First, there were fewer residents than chest radiologists, and second, the skills of the residents were various and their Az values varied widely. The high Az value of the senior board-certified chest radiologists without temporal subtraction (0.916) suggested either that the observers were highly skilled or that the opacities were not as subtle as one might suppose. In fact, the opacities were multifocal, so the potential bias toward a positive result may be possible because the more lobes that are involved, the more likely the observers are to detect the abnormality. Although the improvement of senior board-certified chest radiologists (from 0.916 to 0.947) was statistically significant, the difference was slight. In other words, the improvement of Az values in resident observers was considered to reflect a level of inexperience, although their improvement did not reach statistical significance. It would have been more accurate to use general radiologists with varied skills as observers for better analysis of this technique.
Misregistration and the difference of condition of exposure between previous and current radiographs caused false-positive findings that looked like hazy pulmonary opacities on temporal subtraction images [4, 6]. In our study, we used various film-screen or digital radiography images. Another contributing factor may be that it was difficult for the patients whose chest radiographs showed hazy pulmonary opacities to adequately breath-hold. Therefore, several pseudoabnormalities were due to misregistration and the different conditions of exposure on temporal subtraction images in our study.
Difazio et al. [3] reported that the digital subtraction technique for chest radiography can improve both sensitivity and specificity for the detection of subtle new abnormalities greater than 1 cm in diameter when paired digitized previous and current chest radiographs are viewed in conjunction with the temporal subtraction images. Those authors also reported that moderate amounts of residual misregistration do not lead to false-positive results in most cases when compared with previous and current chest radiographs in conjunction with the temporal subtraction images. Their mean Az value increased from 0.89 without to 0.98 with the temporal subtraction images. However, in our study, no statistically significant improvement of Az values was noted with or without temporal subtraction images for the residents. This might be because in some cases the residents in our study could not differentiate the misregistration from the new hazy pulmonary opacities with the temporal subtraction images. The difference in the results between both studies might depend on the experience of the observers and the characteristics of cases included in each study: focal abnormal opacities were used in the study by Difazio et al., and multifocal were used in our study. In addition, as previously described, a statistical paradox may be caused by the small number of resident observers and the great difference in their skills.
We used various film-screen and digital radiography images in our study, which is one of the advantages of the subtraction technique. Digital subtraction is considered useful for comparing a film with a different kind of film. In clinical settings we usually have various kinds of films to compare—for example, a film obtained in another hospital or a film obtained using a portable system.
Our study has several limitations. First, chest radiologists and physicians ordinarily use films for the interpretation of chest radiographs in daily clinical practice. Therefore, an accurate assessment of this CAD system with a monitor requires comparison of interpretations obtained from only films and from films with temporal subtraction images on the CAD system. Second, in our study, the degree of difficulty for the detection of each abnormality was not considered. However, the high Az values for chest radiologists without temporal subtraction images suggests that it was not difficult to detect the hazy pulmonary opacities in this study. Third, we did not require observers to check the location of the abnormality. It might be that misinterpretation of artifacts made the analysis biased toward a positive result. This possibility, however, must be low because all cases included in this study had multifocal opacities. On this point, further examination is needed. Fourth, our system does not have a way to process the original data of digital radiography. We had to redigitize the image even when the image was digital. Finally, we did not evaluate the causes of misinterpretation. It is important for the establishment of better CAD systems to assess the causes of misinterpretation with this system. However, we did not require the observers to check the location of the opacities, so we could not analyze the cause of the misinterpretation. On this point also, further investigation is needed. However, we achieved the purpose of our study—namely, the evaluation of the usefulness of the CAD system for observers with different levels of experience.
In conclusion, the temporal subtraction technique clearly improves diagnostic accuracy for the detection of hazy pulmonary opacities on chest radiographs, especially when the observers are board-certified chest radiologists who have sufficient skill to evaluate the films and the status of patients. In addition, diagnostic accuracy for the detection of hazy pulmonary opacities on chest radiographs was improved for residents even though the improvement did not reach statistical significance.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  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]
|
|
|
|
 |
|
 S. Kakeda, K. Kamada, Y. Hatakeyama, T. Aoki, Y. Korogi, S. Katsuragawa, and K. Doi Effect of temporal subtraction technique on interpretation time and diagnostic accuracy of chest radiography. Am. J. Roentgenol., November 1, 2006; 187(5): 1253 - 1259. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Kakeda, J. Moriya, H. Sato, T. Aoki, H. Watanabe, H. Nakata, N. Oda, S. Katsuragawa, K. Yamamoto, and K. Doi Improved Detection of Lung Nodules on Chest Radiographs Using a Commercial Computer-Aided Diagnosis System Am. J. Roentgenol., February 1, 2004; 182(2): 505 - 510. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
