DOI: 10.1148/rg.231025129
(Radiographics. 2003;23:255-265.)
© RSNA, 2003
Imaging & Therapeutic Technology |
Computer-aided Diagnosis in Chest Radiography: Results of Large-Scale Observer Tests at the 1996–2001 RSNA Scientific Assemblies1
Hiroyuki Abe, MD,
Heber MacMahon, MD,
Roger Engelmann, MS,
Qiang Li, PhD,
Junji Shiraishi, PhD,
Shigehiko Katsuragawa, PhD,
Masahito Aoyama, PhD,
Takayuki Ishida, PhD,
Kazuto Ashizawa, MD,
Charles E. Metz, PhD and
Kunio Doi, PhD
1 From the Kurt Rossmann Laboratories for Radiologic Image Research, Department of Radiology, MC-2026, University of Chicago, 5841 S Maryland Ave, Chicago, IL 60637. Recipient of an Excellence in Design award for an education exhibit at the 2001 RSNA scientific assembly. Received July 19, 2002; revision requested August 22 and received September 12; accepted September 23. Supported by grant CA62625 from the U.S. Public Health Service. Address correspondence to H.A. (e-mail: habe@uchicago.edu).
 |
Abstract
|
|---|
Since 1996, computer-aided diagnosis (CAD) schemes have been presented as interactive demonstrations on computer workstations at each scientific assembly of the Radiological Society of North America. The schemes involved (a) detection of pulmonary nodules, (b) temporal subtraction, (c) detection of interstitial lung disease, (d) differential diagnosis of interstitial lung disease, and (e) distinction between benign and malignant pulmonary nodules on chest radiographs. Large-scale observer tests were carried out to examine how radiologists can benefit from CAD systems. Observer performance was evaluated by analysis of receiver operating characteristic (ROC) curves. The statistical significance of the difference between the areas under the ROC curves without and with CAD was analyzed with the Student t test. In all of the tests, the diagnostic accuracy of the radiologists in total improved significantly when CAD was used. This result provides additional evidence that CAD has the potential to improve the performance of radiologists in their decision-making process in interpreting chest radiographs.
© RSNA, 2003
Index Terms: Computers, diagnostic aid • Diagnostic radiology, observer performance • Images, interpretation • Lung, interstitial disease, 60.917 • Lung, nodule, 60.30 • Receiver operating characteristic (ROC) curve
|
Introduction
|
|---|
Computer-aided diagnosis (CAD) schemes have been developed by many investigators over the past 2 decades. Recently, interest in CAD has increased as digital technology has been applied to many areas of medical imaging (1). However, most radiologists do not have any experience with use of CAD. To facilitate an understanding of the concept and benefits of CAD, one effective approach is to demonstrate a prototype system to radiologists, thereby providing them with firsthand experience. Therefore, since 1996 we have demonstrated various CAD schemes for chest radiography at the scientific assemblies of the Radiological Society of North America (RSNA) so that radiologists could experience the benefits and limitations of CAD. Five CAD schemes for digital chest radiographs have been presented in the form of real-time observer tests. These CAD schemes, each of which was developed at our institution, were for (a) detection of pulmonary nodules, (b) temporal subtraction, (c) detection of interstitial lung disease, (d) differential diagnosis of interstitial lung disease, and (e) distinction between benign and malignant pulmonary nodules. In this article, we summarize the basic principles of each of these CAD schemes, along with its user interface and the results of the corresponding observer test.
We previously reported the results of our 1996 real-time observer study of CAD for nodule detection (2). The results given here represent an extension of that work to additional observers and to additional CAD schemes. The observer performance data presented in this article have certain limitations due to practical constraints that will be described later; however, each CAD scheme presented here has been shown in more scientifically rigorous studies to improve radiologists’ performance (3–8).
|
Interactive Demonstration
|
|---|
From 1996 to 2001, we exhibited CAD schemes as interactive demonstrations with four computer workstations. To assist radiologists in their understanding of the concept of CAD and its potential usefulness, we implemented an observer test that simulated a clinical setting for each CAD scheme (9). In addition, we provided observers with printed copies of receiver operating characteristic (ROC) curves indicating their personal results so that they could see the effect of CAD on their performance at a glance. In 1996, we presented only a nodule detection program as an observer test, but in each subsequent year we added one new CAD program and a corresponding observer test, as listed in the Introduction. Although an energy subtraction program was presented in 1998, we excluded it from this article because of technical problems with data storage.
Attendees who wished to participate in an observer test could choose from any of the tests available at each meeting. Each user interface was designed not only to demonstrate a CAD program but also to collect data for an observer test. The basic structure of each interface consisted of an image display, controls for operations such as zooming and windowing, and input controls for the observer test. The CAD schemes presented for observer tests in all of the years were identical and were not updated.
Because the main purpose of our observer tests was demonstration of the CAD programs, there was no limitation on the number of times that each attendee could take the tests. To help identify whether an observer took the same test more than once, we asked each observer to enter his or her RSNA badge number before starting each test. In addition, the observers were asked to indicate prior to the test whether they were chest radiologists, other radiologists, radiology residents, or nonradiologists. Subsequently, an informed consent form for each session was presented on the display. The informed consent form included a brief description about the nature of each observer study; an explanation that there was no direct risk, benefit, or reimbursement; and a statement that individual data would be kept confidential and only aggregate results would be published. At this time, an observer was asked to express electronically whether the policy was accepted by clicking the "Accept" or "Do not accept" button. Only the observers who accepted the policy proceeded to the observer study.
The observer tests, with the exception of the test demonstrating the differential diagnosis of interstitial disease, were performed as follows: At the beginning of each test, there was a training session in which the observers acquainted themselves with the operation of the system. In these training sessions, three to eight cases, which were not used in the subsequent test, were presented, depending on the type of the test. Observers were asked to indicate their confidence level for the presence or absence of an abnormality in each case by clicking with a mouse on one or more horizontal bars on the display. Points toward the right or left end of a bar indicated the observer’s greater confidence in a positive or negative result, respectively. In the test for differential diagnosis of interstitial disease, the observers were asked to indicate their most likely diagnosis ("top 1") as well as their second and third most likely diagnoses ("top 2" and "top 3") for each case.
In each test, observers were asked to indicate their diagnostic decisions for each case first without CAD and then with the CAD results displayed. This test procedure is known as a sequential test in ROC observer studies (3). After the observer’s unaided and aided responses had been obtained for a case, the truth for the case (such as the true location of the nodule, whether a nodule was benign or malignant, or the correct disease) was displayed. All the malignant nodules in the observer tests of nodule detection, temporal subtraction, or distinction between benign and malignant pulmonary nodules on chest radiographs were diagnosed by pathologic examination. The diagnoses of benign nodules in those observer tests were determined by pathologic examination or by observation of no change or decrease in nodule size over an interval of 2 years. The location of a nodule was determined by consensus of three chest radiologists. For the cases used in the test of interstitial lung disease detection, the existence and location of interstitial disease were determined by consensus of two chest radiologists. Diagnoses of interstitial lung disease in the test of differential diagnosis of interstitial disease were based on a detailed clinical correlation or on pathologic or bacteriologic proof of the pulmonary lesion. The instructions to observers included the approximate fraction of abnormal images (or malignant cases) included in the test case set, as well as the approximate performance level of the CAD programs. There was no time limit for completing the tests. All digital images (either digitized film images or directly acquired computed radiography images) had a matrix size on the order of 2,000 x 2,000 pixels and 10–12 bits of resolution.
For all tests except that which demonstrated the differential diagnosis of interstitial disease, we employed a proper binormal ROC curve–fitting program to evaluate the performance of the observers by fitting their confidence ratings (10–13). A proper binormal ROC curve can deal with degenerate data sets (10), which may occur when ROC curves are fit to small data sets. We used small case sets in our study to reduce the time required for observers’ participation. The statistical significance of the difference between the areas under the proper ROC curves without and with CAD was determined by application of the Student t test, which accounts for reader-sample variation but not case-sample variation. In the data analysis, the total of chest radiologists, other radiologists, and radiology residents was referred to as "all radiologists" for simplicity.
The scientific value of the results of these observer tests is limited for the following reasons. The tests were not performed under ideal conditions: There were a number of distractions such as crowds, noise, and ambient light because the tests were performed in the exhibit areas of the RSNA meetings, whereas rigorous observer tests should be performed in a dark, quiet room. Moreover, the image display quality was not of a clinical grade and was not always the same from year to year. Although we intended to use only data from each observer’s first use of a test data set, some initial data may have been overwritten by later tests due to a technical error. In addition, because of various other technical issues, not all of the data from the observer tests were processed for analysis. We also eliminated the data sets from some observers due to possible bias and other factors; for example, we eliminated data sets from some observers who are involved with CAD research at our facility and who may have seen some cases previously. Finally, in terms of statistics, case-sample variation was not taken into account in the calculation of the statistical significance of the results.
|
Detection of Pulmonary Nodules
|
|---|
One CAD scheme detected pulmonary nodules by use of a difference-image technique followed by feature extraction and rule-based analysis (14). On the interface of the interactive demonstration, computer output indicating suspicious lesions was indicated by arrows on chest images displayed on the monitor (Fig 1). The sensitivity of the CAD scheme was about 80% and its false-positive rate was approximately one per image for the set of cases presented in the test. From 1996 to 1999, 20 normal chest images and 20 chest images with subtle pulmonary nodules were selected for the test. In 2000 and 2001, however, 10 normal and 10 abnormal cases were used in the test for simplicity. Owing to technical errors, the 1998 data were not processed for analysis. In addition, training cases (seven cases for 1996–1999, four cases for 2000–2001) consisting of normal and abnormal cases were included. The observers were asked to indicate their confidence levels without and with CAD in detecting pulmonary nodules. One hundred twenty-seven radiologists (35 chest radiologists, 63 other radiologists, and 29 residents) have participated in the observer tests since 1996. The results are shown in Table 1 and Figure 2. In each group, there was a statistically significant improvement (P < .001) in nodule detection accuracy with use of CAD.
View larger version (65K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 1. Demonstration of the user interface for the nodule detection scheme. A chest radiograph is shown with the CAD results presented as arrows; the arrow in the upper right lung indicates a false-positive finding, whereas the arrow in the lower left lung indicates a true nodule. On the upper right side of the screen, there are control buttons for adjusting the window setting or zooming the image. A "confidence bar" for entering the observer’s confidence level is on the lower right side.
|
|
|
Temporal Subtraction for Detection of Interval Changes
|
|---|
In temporal subtraction, a previous radiograph is used as a subtraction mask to selectively enhance areas of interval change. With this program, which is based on an iterative image-warping technique (15–17), an observer can easily recognize new opacities as isolated dark foci that stand out from an essentially uniform background (Fig 3). Ten cases with subtle pulmonary nodules and 10 normal cases were presented for the test. In addition, eight training cases representing normal and abnormal patients were selected, with each case including current and previous images. Asequential test similar to that used for nodule detection was employed. The observers were asked to indicate their confidence levels without and with the temporal subtraction image in detecting pulmonary nodules. In this test, the performance of the observers was evaluated on a per-lung basis, and a confidence bar was provided for each of the lungs. Thirty-one radiologists (14 chest radiologists, 14 other radiologists, and three residents) participated in the 1997 observer test. (Because of technical errors, we present only the results of the observer test in 1997.) The results are shown in Table 2 and Figure 4. Within each group except the residents, use of the subtraction image produced a statistically significant improvement in detection accuracy.
View larger version (138K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3. Demonstration of the user interface for the temporal subtraction scheme for detection of interval changes. The interface, which consists of chest radiographs, control buttons, and confidence bars, is similar to that of the nodule detection program. Three chest radiographs are displayed: a current radiograph (upper right), a previous radiograph (upper left), and a temporal subtraction image (lower right). A confidence bar is placed at the bottom of each lung in the current chest image. A newly developed nodule stands out as a black focus (arrow) on the grayish background.
|
|
View larger version (37K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4. Average ROC curves for the radiologists in nodule detection without and with temporal subtraction (TS) images. The radiologists’ performance improved significantly (P < .001) with temporal subtraction.
|
|
|
Detection of Interstitial Lung Disease
|
|---|
Another computerized scheme enables automated detection and characterization of interstitial disease on chest radiographs based on an analysis of texture patterns by use of Fourier transformation (18–23). In our interactive demonstration, the locations of interstitial abnormalities were indicated by markers on chest images displayed on the monitor (Fig 5). The sensitivity and specificity of the CAD scheme were approximately 90% for the cases included in our demonstration.
View larger version (122K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 5. Demonstration of the user interface for detection of interstitial disease. Using the bar on the right side of the screen, observers indicated their level of confidence as to whether interstitial disease was present in the lungs. A chest radiograph is shown with CAD indicators.
|
|
Ten normal chest images and 10 chest images that showed interstitial disease, which included nodular opacities, reticular opacities, and a honeycomb pattern, were used for the test. An additional six chest images, which included normal and abnormal images, were selected for the training session. The observers were asked to indicate their confidence level without and with CAD in detecting interstitial lung disease. Twenty-eight radiologists (eight chest radiologists, 13 other radiologists, and seven residents) have participated in this observer test since 1999. The results are shown in Table 3 and Figure 6. A statistically significant improvement in detection accuracy was obtained by the residents (P < .05) and also by all radiologists (P < .01) with CAD.
View larger version (33K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6. Average ROC curves for the radiologists in detection of interstitial lung disease without and with CAD. The radiologists’ performance improved significantly (P < .01) with CAD.
|
|
|
Differential Diagnosis of Interstitial Lung Disease
|
|---|
An artificial neural network (ANN) was designed to determine the likelihood of disease for 11 types of interstitial lung disease based on input data for 10 clinical parameters and 16 radiologic findings (24–26). The likelihood of each disease was shown by a numerical value from 0 to 1 in a bar graph displayed on the monitor (Fig 7). Four sets of seven chest images that showed interstitial lung diseases were selected for the test from a database that included 11 kinds of interstitial lung disease, and an additional three chest images that showed interstitial lung disease were employed for the training session. The area under the ROC curve of this computer program, which describes its ability to provide a differential diagnosis, was approximately 0.96 for the set of cases in this test. Fifty-three radiologists (16 chest radiologists, 25 other radiologists, and 12 residents) have participated in the observer test since 2000. The results are shown in Table 4. In this test, we determined the sensitivity only for the "top 1," "top 2," and "top 3" categories to avoid a time-consuming procedure for obtaining ROC curves, which would require observers to indicate their confidence levels for all 11 diseases in each case. In each group, there was a statistically significant improvement in diagnostic accuracy (P < .01) by use of the ANN output.
View larger version (133K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 7. Demonstration of the user interface for differential diagnosis of interstitial disease. Eleven diseases are listed on the upper right side of the screen, next to which the ANN outputs are shown as numerical values and bar graphs. On the lower right side, the chief symptom and clinical parameters are shown. Control buttons are located at the top. EG = eosinophilic granuloma, IPF = idiopathic pulmonary fibrosis, Lym Ca = lymphangitic carcinomatosis, PCP = Pneumocystis carinii pneumonia, RCC = renal cell carcinoma, TBC = tuberculosis.
|
|
|
Distinction between Benign and Malignant Pulmonary Nodules
|
|---|
The purpose of another automated computerized scheme was to differentiate between benign and malignant nodules. The likelihood of malignancy was determined from a clinical parameter (age) and several image features by use of linear discriminant analysis (27,28). In this demonstration, the likelihood of malignancy was shown by numerical values from 0 to 100 displayed on the monitor (Fig 8). Two sets of 20 chest images that showed solitary nodules, which consisted of 10 benign and 10 malignant nodules, were selected for the test, and an additional seven chest images that showed either benign or malignant nodules were used for the training session. The accuracy of the CAD scheme in distinguishing between benign and malignant nodules was about 80%. The observers were asked to indicate their confidence levels without and with CAD in diagnosis of the nodules as malignant. Twenty-eight radiologists (seven chest radiologists, 14 other radiologists, and seven residents) participated in the observer test at the 2001 RSNA meeting. The results are shown in Table 5 and Figure 9. Within each group except the chest radiologists, there was a statistically significant improvement (P < .05) in accuracy when the CAD output was available.
View larger version (117K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 8. Demonstration of the user interface for distinction between benign and malignant pulmonary nodules. In addition to an entire chest radiograph, a nodule is shown within a magnified image on the left side. The CAD result is presented above the magnified nodule as a numerical value. A confidence bar and control buttons are placed below the magnified nodule and above the entire chest radiograph, respectively.
|
|
View this table:
[in this window]
[in a new window]
|
TABLE 5. ROC Areas for Distinction between Benign and Malignant Nodules on Chest Radiographs without and with CAD
|
|
View larger version (36K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 9. Average ROC curves for the radiologists in differentiation between benign and malignant pulmonary nodules without and with CAD. The radiologists’ performance improved significantly (P < .001) with CAD.
|
|
|
Conclusions
|
|---|
In all observer tests conducted with five different types of CAD schemes, there was a trend that radiologists were able to improve their performance with CAD. We believe that many radiologists were able to understand the basic concepts of CAD and gain personal experience regarding the benefits and limitations of CAD for chest radiographs.
|
Acknowledgments
|
|---|
We are grateful to all observers who participated in this interactive demonstration at the RSNA meetings and also to E. Lanzl for improving the manuscript.
|
Footnotes
|
|---|
Abbreviations: ANN = artificial neural network,
CAD = computer-aided diagnosis,
ROC = receiver operating characteristic
H.M., S.K., C.E.M., and K.D. are shareholders in R2 Technology, Los Altos, Calif. K.D. is a shareholder in Deus Technologies, Rockville, Md.
|
References
|
|---|
- Doi K, MacMahon H, Katsuragawa S, Nishikawa RM, Jiang Y. Computer-aided diagnosis in radiology: potential and pitfalls. Eur J Radiol 1999; 31:97-109.[CrossRef][Medline]
- MacMahon H, Engelmann R, Behlen F, et al. Computer-aided diagnosis of pulmonary nodules: results of a large-scale observer test. Radiology 1999; 213:723-726.[Abstract/Free Full Text]
- Kobayashi T, Xu XW, MacMahon H, Metz CE, Doi K. Effect of a computer-aided diagnosis scheme on radiologists’ performance in detection of lung nodules on radiographs. Radiology 1996; 199:843-848.[Abstract/Free Full Text]
- Difazio MC, MacMahon H, Xu XW, et al. Effect of time interval difference images on detection accuracy in digital chest radiography. Radiology 1997; 202:447-452.[Abstract/Free Full Text]
- Uozumi T, Nakamura K, Watanabe H, Nakata H, Katsuragawa S, Doi K. ROC analysis of detection of metastatic pulmonary nodules on digital chest radiographs with temporal subtraction. Acad Radiol 2001; 8:871-878.[CrossRef][Medline]
- Monnier-Cholley L, MacMahon H, Katsuragawa S, Morishita J, Ishida T, Doi K. Computer-aided diagnosis for detection of interstitial infiltrates in chest radiographs: evaluation by means of ROC analysis. AJR Am J Roentgenol 1998; 171:1651-1656.[Abstract/Free Full Text]
- Ashizawa K, MacMahon H, Ishida T, et al. Effect of an artificial neural network on radiologists’ performance in the differential diagnosis of interstitial lung disease using chest radiographs. AJR Am J Roentgenol 1999; 172:1311-1315.[Abstract/Free Full Text]
- Shiraishi J, Abe H, Engelmann R, Aoyama M, MacMahon H, Doi K. Computer-aided diagnosis in the distinction between benign and malignant solitary pulmonary nodules in chest radiographs: ROC analysis on radiologists’ performance. Radiology; (in press).
- Engelmann R, MacMahon H, Metz CE, Hoffmann KR, Doi K. Implementation of a technique for performing real-time ROC observer studies to examine the effects of CAD schemes on the performance of radiologists. In: Doi K, MacMahon H, Giger ML, Hoffmann KR, eds. Computer-aided diagnosis in medical imaging. Amsterdam, the Netherlands: Elsevier, 1999; 71-76.
- Pan X, Metz CE. The proper binormal model: parametric receiver operating characteristic curve estimation with degenerate data. Acad Radiol 1997; 4:380-389.[CrossRef][Medline]
- Metz CE, Pan X. "Proper" binormal ROC curves: theory and maximum-likelihood estimation. J Math Psychol 1999; 43:1-13.[CrossRef][Medline]
- Metz CE, Herman BA, Shen JH. Maximum-likelihood estimation of receiver operating characteristic (ROC) curves from continuously distributed data. Stat Med 1998; 17:1033-1053.[CrossRef][Medline]
- Metz CE. ROC methodology in radiologic imaging. Invest Radiol 1986; 21:720-733.[Medline]
- Xu XW, Doi K, Kobayashi T, MacMahon H, Giger ML. Development of an improved CAD scheme for automated detection of lung nodules in digital chest images. Med Phys 1997; 24:1395-1403.[CrossRef][Medline]
- Kano A, Doi K, MacMahon H, Hassell DD, Giger ML. Digital image subtraction of temporally sequential chest images for detection of interval change. Med Phys 1994; 21:453-461.[CrossRef][Medline]
- Ishida T, Ashizawa K, Engelmann R, Katsuragawa S, MacMahon H, Doi K. Application of temporal subtraction for detection of interval change in chest radiographs: improvement of subtraction image using automated initial image matching. J Digit Imaging 1999; 12:77-86.[Medline]
- Ishida T, Katsuragawa S, Nakamura K, MacMahon H, Doi K. Iterative image warping technique for temporal subtraction of sequential chest radiographs to detect interval change. Med Phys 1999; 26:1320-1329.[CrossRef][Medline]
- Katsuragawa S, Doi K, MacMahon H. Image feature analysis and computer-aided diagnosis in digital radiography: detection and characterization of interstitial lung disease in digital chest radiographs. Med Phys 1988; 15:311-319.[CrossRef][Medline]
- Katsuragawa S, Doi K, MacMahon H. Image feature analysis and computer-aided diagnosis in digital radiography: classification of normal and abnormal lungs with interstitial disease in chest images. Med Phys 1989; 16:38-44.[CrossRef][Medline]
- Katsuragawa S, Doi K, Nakamori N, MacMahon H. Image feature analysis and computer-aided diagnosis in digital radiography: effect of digital parameters on the accuracy of computerized analysis of interstitial disease in digital chest radiographs. Med Phys 1990; 17:72-78.[CrossRef][Medline]
- Katsuragawa S, Doi K, MacMahon H, Monnier-Cholley L, Morishita J, Ishida T. Quantitative analysis of geometric-pattern features of interstitial infiltrates in digital chest radiographs. J Digit Imaging 1996; 9:137-144.[Medline]
- Ishida T, Katsuragawa S, Kobayashi T, MacMahon H, Doi K. Computerized analysis of interstitial disease in chest radiographs: improvement of geometric-pattern feature analysis. Med Phys 1997; 24:915-924.[CrossRef][Medline]
- Katsuragawa S, Doi K, MacMahon H, Monnier-Cholley L, Ishida T, Kobayashi T. Classification of normal and abnormal lungs with interstitial diseases by rule-based method and artificial neural networks. J Digit Imaging 1997; 19:108-114.[CrossRef]
- Asada N, Doi K, MacMahon H, et al. Potential usefulness of artificial neural network for differential diagnosis of interstitial lung diseases: a pilot study. Radiology 1990; 177:857-860.[Abstract/Free Full Text]
- Ashizawa K, Ishida T, MacMahon H, Vyborny CJ, Katsuragawa S, Doi K. Artificial neural networks in chest radiology: application to the differential diagnosis of interstitial lung disease. Acad Radiol 1999; 6:2-9.[CrossRef][Medline]
- Abe H, Ashizawa K, Katsuragawa S, MacMahon H, Doi K. Use of an artificial neural network to determine the diagnostic value of specific clinical and radiologic parameters in the diagnosis of interstitial lung disease on chest radiographs. Acad Radiol 2002; 9:13-17.[CrossRef][Medline]
- Nakamura K, Yoshida H, Engelmann R, et al. Computerized analysis of the likelihood of malignancy in solitary pulmonary nodules by use of artificial neural networks. Radiology 2000; 214:823-830.[Abstract/Free Full Text]
- Aoyama M, Li Q, Katsuragawa S, MacMahon H, Doi K. Automated computerized scheme for distinction between benign and malignant solitary pulmonary nodules on chest images. Med Phys 2002; 29:701-708.[CrossRef][Medline]
This article has been cited by other articles:
|
|
|
|
 
F. Girvin and J. P. Ko
Pulmonary Nodules: Detection, Assessment, and CAD
Am. J. Roentgenol.,
October 1, 2008;
191(4):
1057 - 1069.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
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]
|
|
|
|
|
|
|
K. Peldschus, P. Herzog, S. A. Wood, J. I. Cheema, P. Costello, and U. J. Schoepf
Computer-Aided Diagnosis as a Second Reader: Spectrum of Findings in CT Studies of the Chest Interpreted as Normal
Chest,
September 1, 2005;
128(3):
1517 - 1523.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
K Doi
Current status and future potential of computer-aided diagnosis in medical imaging
Br. J. Radiol.,
January 1, 2005;
78(suppl_1):
S3 - s19.
[Abstract]
[Full Text]
[PDF]
|
|
|
Top
Abstract
Introduction
