(Radiology. 2000;214:497-502.)
© RSNA, 2000
Diagnosis of Gastric Cancers: Comparison of Conventional Radiography and Digital Radiography with a 4 Million-Pixel Charge-coupled Device1
Gen Iinuma, PhD, MD,
Kyosuke Ushio, PhD, MD,
Tsutomu Ishikawa, PhD, MD,
Shigeru Nawano, MD,
Ryuzou Sekiguchi, MD and
Mituo Satake, MD
1 From the Department of Diagnostic Radiology, National Cancer Center Hospital, 1-1 Tsukiji 5-chome, Chuo-ku, Tokyo 104-0045, Japan (G.I., K.U., T.I.) and the Department of Diagnostic Radiology, National Cancer Center Hospital East, Chiba, Japan (S.N., R.S., M.S.). From the 1997 RSNA scientific assembly. Received November 30, 1997; revision requested February 23, 1998; final revision received May 13, 1999; accepted July 20. Address reprint requests to G.I. (e-mail: giinuma@gan2 .ncc.go.jp).
 |
Abstract
|
|---|
PURPOSE: To evaluate the differences in accuracy and observer performance at conventional radiography and at digital radiography with a 4 million–pixel charge-coupled device (CCD) for the diagnosis of gastric cancers.
MATERIALS AND METHODS: A prospective study was performed of 225 patients with suspected gastric cancer who were referred to our hospital from January 1997 through February 1997. One hundred twelve patients were examined at conventional radiography and 113 were examined at digital radiography, and 24 and 27 patients had gastric cancer, respectively. Six radiologists interpreted the images, with attention to tumor findings. They were blinded to the clinical details, and their interpretations were rated against those of three other radiologists who examined the patients and who were aware of the clinical information such as endoscopic features and/or histopathologic findings in biopsy specimens. Receiver operating characteristic (ROC) analysis was used to compare the differences in observer performance for the diagnosis of gastric cancers at conventional radiography and at digital radiography.
RESULTS: The overall sensitivity was 64.6% at conventional radiography versus 75.3% at digital radiography (P = .287); specificities were 84.5% and 90.5%, respectively (P = .011); and the positive predictive values were 53.1% and 71.3%, respectively (P = .036). ROC analysis clearly showed higher diagnostic performance at digital radiography than at conventional radiography.
CONCLUSION: The data demonstrate the high diagnostic value of digital radiography with a 4 million–pixel CCD for gastric cancers. The technique has considerable potential as an alternative to conventional gastrointestinal radiography.
Index terms: Diagnostic radiology, observer performance • Gastrointestinal tract, neoplasms, 72.11, 72.121, 72.321 • Gastrointestinal tract, radiography, 72.11, 72.121, 72.321 • Radiography, comparative studies, 72.11, 72.121 • Radiography, digital, 72.121 • Receiver operating characteristic (ROC) curve • Stomach, neoplasms, 72.321
|
Introduction
|
|---|
The demand for the digitization of medical images has been increasing, and digital radiography based on image intensifiers and television cameras has now become widely used in digital subtraction angiography (1–3). This system can provide immediate image display and more rapid data acquisition, compared with that of conventional radiography. However, the system has crucial limitations, especially in image resolution. Recently, a new 4 million–pixel charge-coupled device (CCD)–based digital radiographic (CCD-DR) system has been developed. This system has a large matrix size (2,000 x 2,000), and its image resolution is almost equal to that of conventional radiography.
In this article, we report our experience with this system at gastrointestinal examination. The purpose of this study was to compare the accuracy of CCD-DR with that of conventional radiography in the diagnosis of gastric cancers and to conduct a receiver operating characteristic (ROC) analysis to evaluate the differences in the performance of the observers at CCD-DR and at conventional radiography.
|
MATERIALS AND METHODS
|
|---|
Digital Radiographic System
Our digital radiographic system (DR-2000H; Hitachi Medical, Tokyo, Japan), demonstrated in Figure 1, was composed of an x-ray source, an image intensifier, a television camera for fluoroscopy, an image processing unit, a display, a laser printer, and a CCD unit. One image was acquired per second with a 2,029 x 2,044 matrix. Each was automatically processed and optimized by using several image-processing algorithms, including 
conversion. The images were instantly displayed on a high-resolution display, and hard-copy images were made with a laser printer for side-by-side examination. The system was connected directly to a picture archiving and communication system, or PACS.
View larger version (19K):
[in this window]
[in a new window]
|
Figure 1. Diagram of our digital radiographic system, which is equipped with a new type of CCD camera unit with a resolution of 4 million pixels. PACS = picture archiving and communication system, TV = television.
|
|
Conventional Radiographic System
Conventional radiographs were obtained with a standard screen-film system that was equipped for fluoroscopy. The system had a spot-filming device to convey films without a cassette, which provided an automatic supply of up to 50 10 x 12-inch films. In Japan, gastrointestinal examinations are usually performed with this system, and the remote unit is generally used for this.
Patients
A prospective study was performed from January 1, 1997, to February 28, 1997. All patients with suspected gastric cancer who were referred to our hospital in this period were included. During the 2 months, 225 patients were referred for upper gastrointestinal examination. Among these patients, 58 underwent detailed studies for gastric cancer, and the other 167 were examined for screening or follow up. After assigning the patients to stratified groups for the purpose of examination, we alternately assigned the patients to the conventional radiographic or CCD-DR group.
This method did not provide true randomization, but each group was confirmed to be nearly balanced with regard to patient characteristics. Our institutional review board approved the examination of the patients in the CCD-DR group in this clinical trial, and informed consent was obtained from each patient.
Image Selection
Usually, 10–15 images were obtained in each patient. In this study, to reduce the burden on the observers, the best four images in four fixed positions were selected in each patient by experienced technicians who had no access to the clinical information. The CCD-DR images were printed with a laser printer and were interpreted as were the conventional radiographs (Fig 2). Using the four images, we covered almost the entire stomach with the double-contrast method.
View larger version (127K):
[in this window]
[in a new window]
|
Figure 2. Digital double-contrast radiographs in four fixed positions, which are basically standardized at routine upper gastrointestinal examinations in Japan. Using these four images, we covered almost all of the patient's stomach with the double-contrast method. A, Frontal and B, oblique views of the angular part of the antrum and the lower to middle gastric body. Oblique views of the C, middle to upper gastric body and D, gastric fundus.
|
|
Observers
Six radiologists (including S.N., R.S., and M.S.), all experienced with gastrointestinal examination, independently interpreted all of the radiographs in the 225 patients, with attention to the findings compatible with gastric cancer. They were blinded to the clinical details of each patient, and their interpretations were made as the examinations were performed. The observers were required to report each finding with a corresponding confidence rating to indicate the clinical importance. Six confidence rating categories were predefined with the following corresponding qualitative labels: 1, definitely absent; 2, probably absent; 3, possibly absent; 4, possibly present; 5, probably present; and 6, definitely present (4). Only the abnormalities relevant to gastric cancer were used for the purpose of determining the ROC curves. Furthermore, the observers also noted their opinion of the technical image quality on the available images for each patient with the following five grades: 1, very poor; 2, poor; 3, fair; 4, good; and 5, excellent.
Consensus Panel
A consensus panel of three independent radiologists (G.I., K.U., and T.I.) who had examined the 225 patients in their practices was convened to establish the diagnoses. They were experienced in this area and were aware of the clinical information, such as patient characteristics (age, sex, size and weight, chief symptoms, purpose of the examination), endoscopic features, and histopathologic findings in biopsy specimens. Furthermore, by using all of the other images in addition to the four selected images, they established a radiographic diagnosis for each patient on the basis of all of the evidence.
Statistical Analysis
To ensure the comparability of the background factors of the patients who underwent CCD-DR and of those who underwent conventional radiography, characteristics such as age, sex, size, and weight were compared by using the unpaired t test.
To compare the accuracy of the two modalities, the interpretations of the six radiologists were rated against those of the consensus panel as the standard. Accuracy was compared in terms of sensitivity, specificity, and positive predictive values for each reader. For this analysis, confidence ratings 1–3 were defined as negative findings and confidence ratings 4–6 were defined as positive findings. To compare the mean difference in the modalities across the readers, a generalized es timating equation (5,6) with empirical variance was used to account for the correlation due to multiple readings in the same patients. In addition, to compare the overall accuracy of both modalities, ROC curves and the areas under the curves were obtained for each reader.
The five grade ratings for the impressions of each observer were used to compare the image quality of the modalities. For each reader, the number of patients in whom images were rated were plotted by grade, and these points were joined. By assuming normality, the mean ratings across the readers were compared by using a linear mixed-effect model with modality, consensus diagnosis, and reader as the fixed effects and with the observer and the observer's interaction with the modality as the random effects (7).
All analyses except the generation of the ROC curves were conducted with SAS version 6.12 software (SAS/STATR Software: Changes and Enhancements for Release 6.12. Cary, NC: SAS Institute, 1996). ROCFIT (University of Chicago, Ill) software was used to calculate the areas under the ROC curves (8).
|
RESULTS
|
|---|
Results of the Consensus Panel
One hundred thirteen patients underwent digital radiography, and 112 underwent conventional radiography. No statistically significant difference between the two groups was evident for patient characteristics (age, sex, size, and weight). Each group included 29 patients who were scheduled for detailed examination for gastric cancers, but it was impossible to detect cancerous lesions in three patients in the CCD-DR group and in five patients in the conventional radiography group. An additional gastric cancer was found in one of the patients screened in the CCD-DR group. Consensus findings for all 225 patients are listed in Table 1.
The clinical diagnoses in the 51 patients with gastric cancers are summarized in Table 2. No substantial difference in the size of the gastric cancer was found between the two groups. Allocation and stratification were sufficiently conducted since the rates of gastric cancer (CCD-DR group, 27 of 113 patients; conventional radiography group, 24 of 112 patients) and the clinical characteristics were essentially balanced between the groups. Figures 3 and 4 are representative digital radiographs of two patients from the CCD-DR group.
View larger version (164K):
[in this window]
[in a new window]
|
Figure 3. Frontal digital radiograph obtained in a 66-year-old woman. A small 2.0 x 1.5-cm depressed type of early gastric cancer (arrows) is visible at the posterior wall of the lower gastric body, near the gastric angle. Findings at histopathologic examination of the resected specimens revealed submucosal invasion.
|
|
View larger version (140K):
[in this window]
[in a new window]
|
Figure 4. Frontal digital radiograph obtained in a 62-year-old woman. An advanced 5.0 x 6.0-cm gastric cancer (arrows) is demonstrated in the posterior wall of the gastric angle. The cancer invaded through the proper muscle layer and reached the serosa.
|
|
Sensitivity, Specificity, and Positive Predictive Values
Sensitivities, specificities, and positive predictive values for the six observers are shown in Tables 3–5. Most of the readers had more accurate results with CCD-DR than with conventional radiography. The mean sensitivity of CCD-DR was 75.3% (122 of 162) versus 64.6% (93 of 144) for conventional radiography (P = .287), while the specificities were 90.5% (467 of 516) and 84.5% (446 of 528), respectively (P = .011). The positive predictive values were 71.3% (122 of 171) and 53.1% (93 of 175) (P = .036).
ROC Analysis
The ROC analysis showed that each of six observers achieved more accurate results with CCD-DR than with conventional radiography (Table 6). The ROC curves for the six observers are shown in Figure 5. With all levels of the false-positive fraction, the true-positive fraction of CCD-DR was higher than that of conventional radiography. This was especially true for the curves for all observers, except for that of observer B, on which the true-positive fraction with CCD-DR was considerably higher at a low false-positive fraction.
View larger version (33K):
[in this window]
[in a new window]
|
Figure 5. ROC curves obtained for each observer. All observers achieved more accurate results with CCD-DR (black line) than with conventional radiography (gray line). The diagnostic accuracy of CCD-DR is clearly superior to that of conventional radiography.
|
|
In this article, in which the diagnosis of cancer was based on a fixed cutoff point, we regarded the ROC curve and the areas under the curve as complimentary evidence of the overall performance with the two modalities for each reader. Thus, we did not perform a formal ROC analysis.
Image Quality
The observers' impressions of image quality at CCD-DR and at conventional radiography for the five grade categories are shown in Figure 6. The horizontal axis indicates the five ratings of image quality, and the vertical axis shows the number of patients. For each observer, the CCD-DR distribution showed a tendency to shift to a quality rating higher than that of conventional radiography. The variation was also considerably smaller with CCD-DR than with conventional radiography. Linear mixed-effect models also showed a higher image quality with CCD-DR (P = .018) and a greater variation with conventional radiography (P = .001). Residual plots and goodness-of-fit statistics did not show any violation of the assumption of normality.
|
DISCUSSION
|
|---|
At present, we have two options for gastrointestinal radiography, both with certain limits. For conventional radiography, the storage of films requires space, and if the images are lost, they cannot be reproduced. For digital radiography, computed radiography (9,10) has sufficient image resolution (11–14), but imaging plates are expensive, and time is necessary to acquire the images. Conventional digital radiography based on an image intensifier–television camera system (15,16) does not use imaging plates, but the image resolution is generally limited. However, in 1992, Takahashi et al (17–19) described an image intensifier and television–based digital radiographic system with a 2,048 x 2,048 matrix and an image quality equivalent to that of conventional radiography for the diagnosis of gastrointestinal disease.
Recently, CCD technology has made great advances, and the application of equipment that uses more than 4 million pixels has become common in various fields, including medicine. The CCD directly catches x-ray phosphor emissions and converts them into digital data. For radiographic diagnosis, the CCD has already been successfully applied to digital mammography, and the system has shown excellent spatial resolution (20,21). In 1996, we developed a digital radiographic approach that uses a 4 million–pixel CCD for use in gastrointestinal examinations. Image resolution is excellent and matches that of conventional radiography.
In the present study of CCD-DR, we conducted a prospective, stratified examination of 225 patients who were assigned to two groups that were essentially balanced with regard to the patients' clinical characteristics. Although the observation of the images in the four fixed positions was not 100% complete and although it was a source of some bias, this bias was not worse than that of the conventional approach. Although this diagnostic approach was somewhat different from that usually applied, the criteria for image selection at CCD-DR and at conventional radiography were essentially equivalent.
Under the present conditions, our data showed significantly greater specificity and positive predictive values at CCD-DR than at conventional radiography. The results demonstrated the superiority of CCD-DR in terms of accuracy. Furthermore, the diagnostic value of CCD-DR was considerably greater than that of conventional radiography on the ROC curves for the six observer. All observers achieved more accurate results with CCD-DR than with conventional radiography. The curves for six observers correlated, as expected, with their observation of the same 225 patients, and the results were consistent among observers.
It is noteworthy that the observers' impressions of image quality was also better with CCD-DR; a statistically significant difference was found between the two data sets. Thus, the imaging precision with CCD-DR was superior to that of conventional radiography. Presumably, this is linked to the optimization after image acquisition to provide standardized processing. The double-contrast imaging method for gastrointestinal examination uses barium as the contrast medium.
Fine mucosal patterns can be visualized due to contrast between the barium and the gas. This requires not only high spatial resolution but also a wide latitude. Standardized processing of CCD-DR images is, thus, particularly useful for this purpose.
CCD-DR allows the rapid acquisition of data and the instant image display during examinations, which is a major advantage over the necessity for fluoroscopy and the need to use film with other modalities. In addition, CCD-DR has the potential to optimize image quality and to reduce the radiation dose. Because of these favorable features, this system lends itself to application in various radiologic examinations, especially those that use fluoroscopy.
In conclusion, our data indicate a high diagnostic value for CCD-DR in the detection of gastric cancers. Thus, we regard CCD-DR as a good alternative to conventional radiography for gastrointestinal examinations.
|
Acknowledgments
|
|---|
The authors thank the readers—Hirohito Matue, MD, Yasunori Mizuguchi, MD, and Masaaki Yamashiro, MD—for recording the interpretations. We also are grateful to Tomoyuki Fujita, RT, Satoshi Ogasawara, RT, and Kouji Sekimoto, RT, for in-house technical support. In addition, we express our appreciation to Malcolm Moore, PhD, for his assistance with the preparation of the manuscript and to Seiichiro Yamamoto, PhD, for his assistance with the statistical analysis.
|
Footnotes
|
|---|
Abbreviations: CCD = charge-coupled device
CCD-DR = CCD-based digital radiography
ROC = receiver operating characteristic
Author contributions: Guarantor of integrity of entire study, G.I.; study concepts, G.I., K.U.; study design, G.I., T.I.; definition of intellectual content, G.I.; literature research, G.I.; clinical studies, all authors; data acquisition and analysis, G.I.; statistical analysis, G.I.; manuscript preparation, editing, and review, G.I.
|
References
|
|---|
-
Hillman BJ, Ovitt TW, Nudelman S, Fisher HD, III, Frost MM, Capp MP. Digital video subtraction angiography of renal vascular abnormalities. Radiology 1981; 139:277-280.[Abstract/Free Full Text]
-
Harrington DP, Boxt LM, Murray PD. Digital subtraction angiography: overview of technical principles. AJR Am J Roentgenol 1982; 139:781-786.[Abstract/Free Full Text]
-
Ovitt TW, Newell JD, II. Digital subtraction angiography: technology, equipment, and techniques. Radiol Clin North Am 1985; 23:177-184.[Medline]
-
Ackerman SJ, Gitlin JN, Gayler RW, Flagle CD, Bryan RN. Receiver operating characteristic analysis of fracture and pneumonia detection: comparison of laserdigitized workstation images and conventional analog radiographs. Radiology 1993; 186:263-268.[Abstract/Free Full Text]
-
Liang K, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika 1986; 73:13-22.[Abstract/Free Full Text]
-
Toledano A, Gatsonis CA. Regression analysis of correlated receiver operating characteristics data. Acad Radiol 1995; 2:S30-S36.
-
Searle SR, Casella G, McCulloch CE. Variance components New York, NY: Wiley, 1992.
-
Metz CE. Quantification of failure to demonstrate statistical significance: the usefulness of confidence intervals. Invest Radiol 1989; 24:234-245.[Medline]
-
Sonoda M, Takano M, Miyahara J, Kato H. Computed radiography utilizing scanning laser stimulated luminescence. Radiology 1983; 148:833-838.[Abstract/Free Full Text]
-
Merritt CR, Matthews CC, Scheinhorn D, Balter S. Digital imaging of chest. J Thorac Imaging 1985; 1:1-13.
-
Oestmann JW, Kopans D, Hall DA, McCarthy KA, Rubens JR, Greene R. A comparison of digitized storage phosphors and conventional mammography in the detection of malignant microcalcifications. Invest Radiol 1988; 23:725-728.[Medline]
-
Schaefer CM, Greene R, Oestmann JW, et al. Digital storage phosphor imaging versus conventional film radiography in CT-documented chest disease. Radiology 1990; 174:207-210.[Abstract/Free Full Text]
-
Ishigaki T, Endo T, Ikeda M, et al. Subtle pulmonary disease: detection with computed radiography versus conventional chest radiography. Radiology 1996; 201:51-60.[Abstract/Free Full Text]
-
Higashida Y, Moribe N, Morita K, et al. Detection of subtle microcalcifications: comparison of computed radiography and screen-film mammography. Radiology 1992; 183:483-486.[Abstract/Free Full Text]
-
Feczko PJ, Ackerman LV, Kastan DJ, Halpert RD. Digital radiography of the gastrointestinal tract. Gastrointest Radiol 1988; 13:191-196.[Medline]
-
Steiner E, Mueller PR, Hahn PF, Taaffe J, Ferrucci JT. Digital videofluorography for direct digital spot filming of gastrointestinal studies. Gastrointest Radiol 1989; 14:193-201.[Medline]
-
Takahashi M, Ueno S, Yoshimatsu S, et al. Gastrointestinal examinations with digital radiography. RadioGraphics 1992; 12:969-978.[Abstract]
-
Takahashi M, Ueno S, Tsuchigame T, et al. Development of a 2,048 x 2,048-pixel image intensifier-TV digital radiography system: basic imaging properties and clinical application. Invest Radiol 1992; 27:898-907.[Medline]
-
Ogura T, Fujita H. Resolution properties of a 2048 x 2048 matrix image intensifier-TV based digital radiography system. Med Phys 1994; 21:463-469.[Medline]
-
Yaffe MJ. Direct digital mammography using a scanned-slot CCD imaging system. Med Prog Technol 1993; 19:13-21.[Medline]
-
Hejazi S, Trauernicht DP. System considerations in CCD-based x-ray imaging for digital chest radiography and digital mammography. Med Phys 1997; 24:287-297.[Medline]
This article has been cited by other articles:
|
|
|
|
 
N. A. Obuchowski
Receiver Operating Characteristic Curves and Their Use in Radiology
Radiology,
October 1, 2003;
229(1):
3 - 8.
[Abstract]
[Full Text]
[PDF]
|
|
|
TOP
Abstract
Introduction
= observer A, 
= observer B, • = observer C, 
= observer D, 
= observer E, 
= observer F.