| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1
Department of Radiology, Osaka City University Hospital, 1-5-7 Asahi-machi, Abeno-ku, Osaka, Osaka 545-8586 Japan.
2
Department of Radiology, Iwate Medical University, 19-1 Uchimaru, Morioka, Iwate 020-8505 Japan.
3
Department of Radiology, Ehime University, Shizukawa, Shigenobu-cho, Onsen-gun, Ehime 791-0295 Japan.
4
Department of Radiology, Yamaguchi University, 1144 Ogushi, Ube, Yamaguchi 755-8506 Japan.
5
Department of Radiology, Kanazawa University, 5-11-80, Kotateno, Kanazawa, Ishikawa 920-0942 Japan.
6
Medical Engineering Division, Toshiba, 1-1-1 Shibaura, Minato-ku, Tokyo 105-8001 Japan.
7
Medical Imaging Division, Konica, 1 Sakura-cho, Hino, Tokyo 191-8511 Japan.
8
Department of Information Science, Faculty of Engineering, Gifu University, 1-1, Yanagido, Gifu, Gifu 501-1193 Japan.
9
Department of Radiological Technology, Nagoya University School of Health Sciences, 1-1-20 Taikouminami, Higashi-ku, Nagoya, Aichi 461-8673 Japan.
10
Kurt Rossmann Laboratories for Radiologic Image Research, Department of Radiology, The University of Chicago, 5841 S. Maryland Ave., MC2026, Chicago, IL 60637.
Received January 22, 1999;
accepted after revision June 14, 1999.
Presented at the annual meetings of the Radiological Society of North America, Chicago, November 1997 and November 1998.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. One hundred and fifty-four conventional chest radiographs with a lung nodule and 93 radiographs without a nodule were selected from 14 medical centers and were digitized by a laser digitizer with a 2048 x 2048 matrix size (0.175-mm pixels) and a 12-bit gray scale. Lung nodule images were classified into five groups according to the degrees of subtlety shown. The observations of 20 participating radiologists were subjected to ROC analysis for detecting solitary pulmonary nodules. Experimental results (areas under the curve, Az) obtained from observer studies were used for characterization of five groups of lung nodules with different degrees of subtlety.
RESULTS. ROC analysis showed that the database included a wide range of various nodules yielding Az values from 0.574 to 0.991 for the five categories of cases for different degrees of subtlety.
CONCLUSION. This database can be useful for many purposes, including research, education, quality assurance, and other demonstrations.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
When the image database is established, one of the most important properties associated with the database is the level of radiologists' performance for detection of abnormalities included in the database. If users of this database know the average level of radiologists' performance, it would be valuable information to radiologists who can compare their performance with the average performance of other radiologists. However, to our knowledge, an image database that includes the radiologists' performance is not available. Therefore, in this study, a digital image database (www.macnet.or.jp/jsrt2/cdrom_nodules.html) for chest radiographs with and without solitary pulmonary nodules was developed for common use by radiologists and researchers. In addition, the radiologists' performance for detection of pulmonary nodules included in the database was evaluated using receiver operating characteristic (ROC) analysis.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Three experienced chest radiologists excluded images with multiple pulmonary nodules or of poor image quality. The database included 247 posteroanterior chest images, which consisted of 154 images with and 93 images without a nodule. One hundred nodules were malignant and 54 were benign. Of the patients with nodules, 68 were men and 86 were women, whereas patients without nodules included 51 men and 42 women. The average age of patients with nodules was 60 years old. Two patients were 21-30 years old, seven were 31-40 years old, 24 were 41-50 years old, 37 were 51-60 years old, 53 were 61-70 years old, and 29 were 71 years old or older. Two patients' ages were unknown.
The database included 31 nodules that were 0-10 mm, 52 that were 11-15 mm, 36 that were 16-20 mm, 14 that were 21-25 mm, 17 that were 26-30 mm, and four that were 31-60 mm. The average size of all nodules included in the database was 17.3 mm.
To investigate the distribution of anatomic locations of nodules, right and left lungs on the radiographs were divided into upper, middle, and lower regions. Forty-two nodules were found in the upper right lung, 30 in the middle right lung, 23 in the lower right lung, 20 in the upper left lung, 21 in the middle left lung, and 18 in the lower left lung.
The database contains a text file including the following information about each patient: nodule size (mm), age, sex, final diagnosis, degree of subtlety, anatomic location of the nodule, x- and y-coordinates of the center of the nodule in a digital image (upper left corner was defined as the origin of x- and y-coordinates), and nodule classification as malignant or benign. Furthermore, a simple schematic diagram that indicates each location of a lung nodule is contained in a tagged image file format file for visual assistance. The database does not include information about calcification of nodules.
The images in the database were grouped according to the degree of subtlety of the lung nodule based on the consensus of three chest radiologists. The levels of subtlety in detecting a lung nodule were defined as follows: level 1, extremely subtle (detection is extremely difficult because of low contrast, small size, or overlap with a normal structure); level 2, very subtle (detection is very difficult); level 3, subtle (detection is difficult); level 4, relatively obvious (detection is relatively easy); and level 5, obvious (detection is easy). Twenty-five cases were categorized as level 1, 29 as level 2, 50 as level 3, 38 as level 4, and 12 as level 5.
An ROC study of detection of lung nodules was performed to evaluate the radiologists' performance and to characterize this database. All images in this database were printed on film (34.6 cm x 34.6 cm [14 inch x 14 inch]) with a Li-7DD laser printer (Konica). Twenty radiologists (nine chest and 11 general radiologists) from four institutions participated in this study. Their experience ranged from 2 to 22 years (average, 12.1 years). All images were presented in a randomized sequence for detection of lung nodules. For ROC experiments, we used a continuous rating scale and the curve fitting program LABROC5 (Metz CE, The University of Chicago) [5, 6].
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
|
ROC curves obtained from radiologists at four institutions were similar (the numbers of observers at institutions A, B, C, and D were three, eight, four, and five) (Fig. 2). We found no statistically significant differences among the average Az values of the four institutions.
|
ROC curves were calculated for detection of patients with nodules in each group of different degrees of subtlety as positive cases and all patients without nodules as negative cases. These curves were clearly distinguished as five curves corresponding to degree of subtlety (Fig. 3). The relationship between the degree of subtlety and the average confidence rating score (maximum = 50), which were obtained in ROC, experiments indicated a high correlation coefficient, r = 0.999 (Fig. 4). These results indicate that the images showing nodules in each group of the database are distinctly different and cover a wide range of nodules that differ in their visual appearance.
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The proper choice of digital image parameters is important for constructing a digital image database. Small pixels provide high spatial resolution but require large amounts of storage space and increase the computation time for image processing. The parameters used for images in this database may not be optimal for chest images of patients with interstitial disease, but they are appropriate for detection of lung nodules [7, 8]. The pixel size and the gray-scale levels for the computed radiography system that has been used widely for chest radiography are similar to those used for the images in this database. Therefore, we believe that the image quality of this database is acceptable for evaluation of lung nodules.
Although this database was constructed for research on digital imaging, it might be useful for many different purposes, including training and education. One finding in this study indicated that the degree of subtlety correlated with the size of nodules and the age of patients as shown in Table 2 (two patients were not included because of unknown ages). Table 2 shows that the smaller the nodule size and the younger the patient, the more subtle the pulmonary nodule appears, and thus the more difficult nodules are to detect on chest images. Although the average detectability increased with increased average nodule size up to 20 mm, the average detectability decreased in nodules larger than 20 mm probably because the large nodules included in this database tend to overlap normal anatomic structures and thus are difficult to detect (Table 3).
|
|
If this database is used for ROC analysis or other observer performance studies, using all 247 cases may be difficult in some experiments. However, subsets of this database can be used for various studies intended for particular purposes. For example, a subset may be selected for a small number of cases or combinations of cases with several different degrees of subtlety. To illustrate this flexibility, we created three subsets of images using different hypothetic combinations of degrees of subtlety. The three subsets consisted of levels 1-3 as subtle, levels 2-4 as intermediate subtlety, and levels 3-5 as obvious. ROC curves show the expected detection performance using these subset data (Fig. 5). Therefore, subsets such as these can be used for testing various techniques and methods such as different display devices and data compression techniques.
|
In conclusion, we have developed a common image database for chest radiographs with and without a lung nodule. Twenty radiologists' performance in detecting solitary pulmonary nodules included in the database was evaluated using an ROC analysis. The results of our ROC analysis for detection of lung nodules indicated that this database includes a wide range of nodules with different degree of subtlety. Therefore, we believe that this database will be useful for many purposes, including research, education, quality assurance, and demonstration.
Acknowledgments
We thank Takahiro Kozuka, Ryusaku Yamada, and Akiyoshi Ohtsuka, for their encouragement and support of this work; Kazuhisa Kosakai and Akane Utsunomiya, Osaka City University Hospital, for their technical support; 20 radiologists at four institutions for their participation in the ROC observer studies; and 14 institutions for their kind cooperation in providing clinical images for the creation of this database.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  L. Monnier-Cholley, F. Carrat, B. P. Cholley, J.-M. Tubiana, and L. Arrive Detection of Lung Cancer on Radiographs: Receiver Operating Characteristic Analyses of Radiologists', Pulmonologists', and Anesthesiologists' Performance Radiology, December 1, 2004; 233(3): 799 - 805. [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]
|
|
|
|
|
|
J. Krejza, Z. Mariak, and J. Lewko Standardization of Flow Velocities with Respect to Age and Sex Improves the Accuracy of Transcranial Color Doppler Sonography of Middle Cerebral Artery Spasm Am. J. Roentgenol., July 1, 2003; 181(1): 245 - 252. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. Shiraishi, H. Abe, R. Engelmann, M. Aoyama, H. MacMahon, and K. Doi Computer-aided Diagnosis to Distinguish Benign from Malignant Solitary Pulmonary Nodules on Radiographs: ROC Analysis of Radiologists' Performance--Initial Experience Radiology, May 1, 2003; 227(2): 469 - 474. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
U. Rapp-Bernhardt, F. W. Roehl, R. C. Gibbs, H. Schmidl, U. W. Krause, and T. M. Bernhardt Flat-Panel X-ray Detector Based on Amorphous Silicon versus Asymmetric Screen-Film System: Phantom Study of Dose Reduction and Depiction of Simulated Findings Radiology, May 1, 2003; 227(2): 484 - 492. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Tsukuda, A. Heshiki, S. Katsuragawa, Q. Li, H. MacMahon, and K. Doi Detection of Lung Nodules on Digital Chest Radiographs: Potential Usefulness of a New Contralateral Subtraction Technique Radiology, April 1, 2002; 223(1): 199 - 203. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
