Published online before print. 10.1148/radiol. ... chest radiography with regard to the accuracy of detection of lung nodules. MATERIALS AND ... temporal subtraction images were produced by using a program developed at the University of ...
Radiology
Shingo Kakeda, MD Katsumi Nakamura, MD Koji Kamada, MD Hideyuki Watanabe, MD Hajime Nakata, MD Shigehiko Katsuragawa, PhD Kunio Doi, PhD
Improved Detection of Lung Nodules by Using a Temporal Subtraction Technique1 PURPOSE: To evaluate the effect of a temporal subtraction technique for digital chest radiography with regard to the accuracy of detection of lung nodules.
Index terms: Diagnostic radiography, observer performance Diagnostic radiology, 60.1215 Lung, abnormalities, 60.31, 60.3211, 60.3212, 60.3213 Lung neoplasms, diagnosis, 60.1215, 60.31, 60.3211, 60.3212, 60.3213 Subtraction, digital, 60.1215 Published online before print 10.1148/radiol.2241010719 Radiology 2002; 224:145–151 Abbreviations: Az ⫽ area under ROC curve CR ⫽ computed radiography ROC ⫽ receiver operating characteristic 1 From the Department of Radiology, University of Occupational and Environmental Health School of Medicine, Iseigaoka 1-1, Yahatanisi-ku, Kitakyushushi 807-8555, Japan (S. Kakeda, K.N., K.K., H.W., H.N.); Research Center, Nippon Bunri University, Ooita-shi, Japan (S. Katsuragawa); and Kurt Rossmann Laboratories for Radiologic Image Research, Department of Radiology, University of Chicago, Ill (K.D.). Received August 17, 2001; revision requested May 10; final revision received January 8, 2002; accepted January 29. Address correspondence to S. Kakeda (e-mail: kakeda @med.uoeh-u.ac.jp). K. D. is a shareholder of R2 Technology, Los Altos, Calif. It is the policy of the University of Chicago that investigators disclose publicly actual or potential significant financial interests that may appear to be affected by the research activities. ©
RSNA, 2002
Author contributions: Guarantors of integrity of entire study, S. Kakeda, K.N.; study concepts and design, S. Kakeda, K.N.; literature research, S. Kakeda, K.K.; clinical studies, S. Kakeda, K.N., H.W.; data acquisition and analysis/interpretation, S. Kakeda, K.N.; statistical analysis, S. Kakeda, K.N., S. Katsuragawa; manuscript preparation, S. Kakeda, K.K.; manuscript definition of intellectual content and editing, S. Kakeda, H.N., K.D.; manuscript revision/review and final version approval, all authors.
MATERIALS AND METHODS: Twenty solitary lung nodules smaller than 30 mm in diameter, including 10 lung cancers and 10 benign nodules, were used. The nodules were grouped subjectively according to their subtlety. For nonnodular cases, 20 nodules without perceptible interval changes were selected. All chest radiographs were obtained by using a computed radiographic system, and temporal subtraction images were produced by using a program developed at the University of Chicago. The effect of the temporal subtraction image was evaluated by using an observer performance study, with use of receiver operating characteristic analysis. RESULTS: Observer performance with temporal subtraction images was substantially improved (Az ⫽ 0.980 and 0.958), as compared with that without temporal subtraction images (Az ⫽ 0.920 and 0.825) for the certified radiologists and radiology residents, respectively. The temporal subtraction technique clearly improved diagnostic accuracy for detecting lung nodules, especially subtle cases. CONCLUSION: The temporal subtraction technique is useful for improving detection accuracy for peripheral lung nodules on digital chest radiographs. ©
RSNA, 2002
Chest radiography is currently the most frequently used screening procedure for chest diseases owing to its economical aspects and ease of performance. Small lung nodules are, however, often missed, and radiologists may fail to detect lung nodules in up to 30% of cases with actual positive diagnoses (1–3). Although double reading involving a second radiologist and comparison of a current chest radiograph with a previous one can improve detection accuracy for lung nodules, it is unavoidable that radiologists miss a certain percentage of lung nodules. Nodules that are difficult to detect are reported to be small (1,4), of low contrast, and overlapping with normal anatomic structures (5–9). The temporal subtraction technique is a method with which a previous chest radiograph is subtracted from a current radiograph so that interval changes are enhanced. The purpose of the temporal subtraction technique in the interpretation of a chest radiograph is to alert the radiologist to the location of newly developed abnormalities and to assist radiologists in their final decision. Although temporal subtraction techniques and their possible advantages have been reported (10 –12), additional studies seem to be needed to further validate the usefulness of temporal subtraction. The purpose of our study was to evaluate the effect of the temporal subtraction technique for digital chest radiography with regard to the accuracy of detection of lung nodules.
MATERIALS AND METHODS Digital Radiography and Temporal Subtraction All chest radiographs were exposed at 100 kV with a 10:1 grid and were obtained by using a computed radiographic (CR) system (FCR; Fuji Photo Film, Tokyo, Japan). The imaging plate (model ST-V; Fuji) was 35 ⫻ 43 cm (matrix size, 1,760 ⫻ 2,140; 10-bit gray level; pixel size, 200 m). The output image was printed on 23.5 ⫻ 23.5-cm film (CR Film; 145
Radiology
Fuji) (67% reduction) by using a laser printer (model FL-IMD; Fuji). Image reading of the CR data was performed by using an automatic exposure data recognition method. CR images were processed by using a sigmoid long-contrast Hurter and Driffield curve and slight edge enhancement at higher frequencies. The enhancement factor was 0.2, with a frequency range of more than about 0.35 cycle per millimeter. The gradation processing parameters were as follows: rotation amount, 0.9; gradation type, E; rotation center, 1.6; and gradation shifting amount, ⫺0.2. The unprocessed CR image data of current and previous images were used to provide the subtraction image. Image data from chest radiographs were transferred to a workstation (O2; Silicon Graphics, Mountain View, Calif). Temporal subtraction images were created by using a temporal subtraction program developed at the University of Chicago. Previous and current images were subsampled to 586 ⫻ 586 pixels, and the previous image was shifted and rotated to correct for variations in patient positioning between the previous and current images. After global matching was performed with low-resolution images, a number of template regions of interest (32 ⫻ 32 pixels) and the corresponding search area regions of interest (64 ⫻ 64 pixels) were selected from the previous and current images, respectively. Local matching of template regions of interest with corresponding regions of interest and shift values ⌬x and ⌬y were determined for all pairs of selected regions of interest by using a cross-correlation technique. The previous image was nonlinearly warped according to local shift vectors for “best matching.” For further matching, image warping was repeated. Finally, a temporal subtraction image was obtained by subtracting a second warped previous image from a current image (10 –12). It took approximately 15 seconds to produce a temporal subtraction image from a pair of current and previous CR images. When multiple previous images were available, the best-matched temporal subtraction images were selected.
Case Selection and Classification Our institutional review board did not require its approval or informed consent for this study. Consecutive cases with solitary lung nodules up to 30 mm in diameter seen at our institution between January 1997 and February 1998 were selected from a thoracic computed tomography (CT) file. From them we further selected 146
䡠
Radiology
䡠
July 2002
20 cases in which the diagnosis was proved and current and previous chest radiographs were available (nodule size range, 7–30 mm in diameter; mean, 19.4 mm; median, 20.0 mm). The final diagnosis was primary lung cancer in 10 cases and benign nodules in 10 cases. Pathologic proof was obtained by means of surgical resection or fine-needle biopsy in all primary lung cancers (four adenocarcinomas, four squamous cell carcinomas, and two small cell carcinomas). Benign nodules were diagnosed by means of surgical resection in one case (intrapulmonary lymph node), by means of transbronchial bronchoscopic biopsy in three cases (organizing pneumonia, one case; cryptococcosis, one case; and aspergilloma, one case), and by means of clinical follow-up in six cases with organizing pneumoniae. There were eight women and 12 men, with an age range of 25– 81 years (mean age, 62.4 years). Two certified radiologists (S. Kakeda, K.N.) independently classified the radiographs showing nodules, in accordance with the subtlety of the lung nodule, by using a five-point scale: Subtlety 1 ⫽ extremely subtle, subtlety 2 ⫽ very subtle, subtlety 3 ⫽ subtle, subtlety 4 ⫽ relatively obvious, or subtlety 5 ⫽ obvious. They reached a consensus regarding the classification of the subtlety of the nodules when their ratings differed. There were six (30%) cases with subtlety 1, six (30%) with subtlety 2, five (25%) with subtlety 3, two (10%) with subtlety 4, and one (5%) with subtlety 5. Twenty additional nonnodular cases without perceptible interval changes were selected on the basis of CT findings. Nonnodular cases included calcifications or strandlike opacities without perceptible interval changes. There were 10 women and 10 men, with an age range of 46 –77 years (mean age, 60.0 years). The nodules were located in all six parts of the lung fields as follows: four on the right upper lung (one overlapping the clavicle), seven on the right middle lung (four overlapping the hilum), six on the right lower lung (three overlapping the diaphragm), one on the left upper lung, one on the left middle lung, and one on the left lower lung, overlapping the diaphragm. Four cases with strandlike opacities and one with calcifications were included in the nonnodular cases. Calcifications were located on the right upper lung, and strandlike opacities were located as follows: one on the right upper lung, one on the left lower lung, and two on both lungs. The interval between acquisition of
current and previous radiographs was 60 – 1,245 days (mean, 515 days) in nodular cases and 120 – 665 days (mean, 300 days) in nonnodular cases.
Image Display For subjective evaluation and the observer performance study, current and previous radiographs were mounted on a conventional view box as radiographs. Next to the view box, the temporal subtraction image was displayed on the 1,200 ⫻ 1,600-line, 20-inch-diagonal screen, cathode-ray-tube, or CRT, monitor (model GDM4011P; Silicon Graphics). The room had uniformly low ambient light conditions. The observers were permitted to manipulate monitor brightness and contrast by using a track ball or push buttons.
Subjective Evaluation of Temporal Subtraction First, the quality of the temporal subtraction images in the 40 cases was subjectively rated by the two certified radiologists (S. Kakeda, K.N.) by using a fivepoint scale: Quality 1 ⫽ very good, quality 2 ⫽ good, quality 3 ⫽ reasonably good, quality 4 ⫽ poor, and quality 5 ⫽ very poor, according to the degree of rib shadow disappearance. For each nodule case, the two certified radiologists subjectively evaluated the relative change in confidence in nodule detection between using the radiograph alone and using the radiograph plus temporal subtraction. A five-point rating scale was used: Markedly improved ⫽ ⫹2, moderately improved ⫽ ⫹1, unchanged ⫽ 0, moderately degraded ⫽ ⫺1, and markedly degraded ⫽ ⫺2. For all evaluations the two certified radiologists worked independently, and a final rating score was determined by means of consensus when scores differed. First the current and previous radiographs were presented, then the temporal subtraction image was displayed on a CRT monitor for interpretation and rating.
Observer Performance Study Next, an observer performance study was performed. Eight radiologists (four certified radiologists and four radiology residents with 2–3 years of training) participated as observers. Before the test, six training cases were shown, and the observers were informed of (a) the fraction (50%) of lung nodule cases included in the study, (b) the role of the temporal subtraction image as a “second opinion,” (c) whether an abnormality was present, Kakeda et al
Radiology
conventional interpretation, and the observer marked his or her confidence level by using a black pencil on a 7-cm line. Current and previous radiographs were evaluated on posteroanterior views. The temporal subtraction image was then displayed on a CRT monitor; the radiologist viewed it, as well as the current and previous images, and marked his or her confidence level by using a red pencil on the same line, if this level was different from the first confidence level.
Data Analysis
Figure 1. Graph shows subjective evaluation of the effect of temporal subtraction images on ease of nodule detection according to nodule subtlety. Black bar ⫽ extremely subtle, gray bar ⫽ very subtle, bar with narrow diagonal stripes ⫽ subtle, bar with wide diagonal stripes ⫽ relatively obvious, and white bar ⫽ obvious.
Figure 2. Images show lung cancer in a 68year-old woman. (a) Previous posteroanterior radiograph, (b) current posteroanterior radiograph, and (c) temporal subtraction image; this last image enhanced the visibility of a nodule (arrows) overlapping the diaphragm.
and (d) the definition of a “solitary” nodule as one 7–30 mm in diameter. For the 20 nodular and 20 nonnodular cases, a sequential test method was used (13), only one case was shown at a time, and reading time was not limited. A continuous rating scale with a line-marking method was used to record each observer’s confidence level regarding the presence of a lung nodule. Current and previous radiographs were shown first for Volume 224
䡠
Number 1
For the subjective evaluation part of our study, we evaluated the relationship between the subjective rating score of relative change in confidence and the subtlety of the nodules. Receiver operating characteristic (ROC) analysis was used for comparison of observer performances in detecting the presence or absence of lung nodules without and with the temporal subtraction images for all 40 cases. Estimates of the area under the ROC curve (Az ) and SDs were computed by using the computer program (LABROC5) provided by Metz et al (14). The observers were not required to indicate the location of a nodule, which is the standard procedure for ROC analysis. In an observer performance study, the difference among individual observers should be substantial in marking the continuous rating scale. In this sense, it was difficult to categorize the cases marked in the midpoint of the “absence” and “presence” into falsenegative or true-positive findings. Therefore, in this observer performance study, we calculated the difference between the confidence levels without (first rating) and with (second rating) the temporal subtraction images and assumed that a clinically relevant change in confidence levels occurred only when the difference calculated in this way was greater than 20 units on a 0 –100 confidence rating scale. The full length of the 7-cm line in the confidence rating scale corresponded to 100 units. A shift of more than 20 units in the direction of a correct diagnosis implied that use of the temporal subtraction image was beneficial. Similarly, a shift of more than 20 units in the direction of an incorrect diagnosis implied that use of the temporal subtraction image was detrimental. The statistical significance of the difference in Az values without and with the temporal subtraction images was estimated by using the Dorfman-Berbaum-Metz method, in which both interobserver variation and case sample variation are considered. In addition, the difference between the average number of cases affected beneficially Improved Detection of Lung Nodules
䡠
147
Comparison of Az Values with and without Temporal Subtraction Images A z Value
Radiology
Reviewers
Figure 3. Graph shows ROC curves for detecting lung nodules with and without use of temporal subtraction images.
Residents A B C D Average Certified radiologists E F G H Average Overall average
Without Subtraction Image
With Subtraction Image
0.764 ⫾ 0.078 0.914 ⫾ 0.044 0.787 ⫾ 0.071 0.835 ⫾ 0.066 0.825 ⫾ 0.066
0.947 ⫾ 0.038 0.972 ⫾ 0.024 0.937 ⫾ 0.049 0.977 ⫾ 0.027 0.958 ⫾ 0.019
0.920 ⫾ 0.046 0.884 ⫾ 0.059 0.945 ⫾ 0.034 0.932 ⫾ 0.046 0.920 ⫾ 0.026 0.873 ⫾ 0.069
0.993 ⫾ 0.009 1.000 0.993 ⫾ 0.009 0.934 ⫾ 0.046 0.980 ⫾ 0.031 0.969 ⫾ 0.026*
Note.—Data are means ⫾ SDs. * P ⫽ .027, Dorfman-Berbaum-Metz method.
and those affected detrimentally by the temporal subtraction images was analyzed by using a two-tailed Student t test.
RESULTS Subjective Evaluation All of the quality ratings of the temporal subtraction images were quality 1 (very good) to 3 (reasonably good). Specifically, 15 (38%) of the 40 temporal subtraction images were rated as quality 1; 19 (48%), as quality 2 (good); and six (15%), as quality 3 (reasonably good), whereas none were rated as either quality 4 (poor) or 5 (very poor). With regard to subjective evaluation of the effect of the temporal subtraction images, 10 (50%) of the 20 nodules were markedly improved (⫹2), seven (35%) were moderately improved (⫹1), and three (15%) were unchanged (⫹0). Confidence in the existence of the nodules was increased, since all nodules were clearly enhanced on the temporal subtraction images. Artifacts indistinguishable from the nodules, which might have increased false-positive findings, were not observed. Figure 1 shows the relationship between the subtlety of the nodules and the subjective rating. Six (60%) of the 10 nodules rated as markedly improved (⫹2) were of subtlety 1 (extremely subtle), and two (20%) were of subtlety 2 (very subtle), and two (20%) were of subtlety 3 (subtle). Four (57%) of the seven nodules rated as moderately improved (⫹1) were of subtlety 2 (very subtle), and three (43%) were of subtlety 3 (subtle). Six (60%) of the 10 nodules rated as markedly improved (⫹2) overlapped the diaphragm, hilum, or clavicle 148
䡠
Radiology
䡠
July 2002
Figure 4. Graph shows the number of cases (⬎20%) affected by temporal subtraction images in confidence rating with regard to nodular cases (A–H represent the eight readers).
(Fig 2). The nodules rated as unchanged (⫹0) tended to be slightly larger (mean diameter, 23.0 mm) than those rated as moderately (⫹1) (20.1 mm) or markedly (⫹2) improved (19.3 mm).
Observer Performance Study Average ROC curves demonstrated a significant improvement in the accuracy of nodule detection with the temporal subtraction images (Fig 3). Az values for all eight observers are shown in the Table. The average Az value increased significantly from 0.873 without the temporal subtraction image to 0.969 with the image (P ⫽ .027). Individually, the use of temporal subtraction images was beneficial to seven (88%) of the eight observers.
The average Az improvement with use of the temporal subtraction images was greater for the subgroup of radiology residents than for the certified radiologists (from 0.825 to 0.958 vs from 0.920 to 0.980). The average Az value for the radiology residents with use of the temporal subtraction images was greater than that for the certified radiologists without use of the temporal subtraction images (0.958 vs 0.920). The results of a clinically relevant change in confidence level are shown in Figures 4 and 5. Among the 20 nodule cases, the average number of cases affected beneficially was significantly larger than that affected detrimentally (mean number of cases, 6.75 [34%] vs 0.25 [1%]; Kakeda et al
Radiology Figure 5. Graph shows the number of cases (⬎20%) affected by temporal subtraction images in confidence rating with regard to nonnodular cases (A–H represent the eight readers).
Figure 6. Graph shows the average number of cases (⬎20%) affected by temporal subtraction images according to subtlety in nodular cases.
tween the subtlety of the nodules and the average number of clinically relevant changes in confidence level. The temporal subtraction images clearly improved the diagnostic accuracy of lung nodule detection, especially of subtle cases (Fig 7). The average number of cases affected beneficially by the temporal subtraction images was 3.4 (56%) of six cases with subtlety 1 (extremely subtle), 2.4 (40%) of six with subtlety 2 (very subtle), and 0.9 (18%) of five with subtlety 3 (subtle); none had subtlety 4 (relatively obvious) or 5 (obvious). The average number of cases affected detrimentally was 0.25 (4%) of six with subtlety 1.
DISCUSSION Figure 7. Images show organizing pneumonia in a 77-year-old man. (a) Previous posteroanterior radiograph, (b) current posteroanterior radiograph, and (c) temporal subtraction image; this last image enhanced the visibility of a subtle nodule (arrow) that was difficult to recognize on conventional radiographs alone.
P ⬍ .001). Two nodules affected detrimentally were missed by the radiology Volume 224
䡠
Number 1
residents on both chest radiographs and temporal subtraction images, although these nodules were clearly visible on the temporal subtraction images. This occurred because of inexperience and carelessness. Among the 20 nonnodular cases, the average number of cases affected beneficially was significantly larger than that affected detrimentally (mean number of cases, 4.88 [24%] vs 0.88 [4%]; P ⫽ .015). All detrimental effects were caused by misregistration artifacts. Figure 6 shows the relationship be-
Difazio et al (11) reported that the temporal subtraction technique can significantly improve diagnostic accuracy for newly developed lung abnormalities larger than 10 mm in diameter on chest radiographs. The results of the current observer study also indicated that reading chest radiographs with temporal subtraction images significantly increased Az values in diagnosing subtle lung nodules. In the current study, nodule cases were selected on the basis that the nodules were smaller than 30 mm in diameter, which means that most of these were visually subtle. The criterion of selection of nodules smaller than 30 mm in diameter was chosen because nodules likely to be missed in clinical settings were the subjects of this study. If the obvious nodules that could have easily been detected with chest radiographs had been included in this study, the improvement in Improved Detection of Lung Nodules
䡠
149
Radiology
diagnostic accuracy with the temporal subtraction images would have been less than the results described in this study. In this observer performance study, we analyzed the effect of the temporal subtraction images separately in nodular and nonnodular cases. In nodular cases, use of temporal subtraction images was substantially beneficial, which indicates that the images can help prevent observers from missing lung nodules. It is important to note that the cases with true-positive detections on chest radiographs were not affected detrimentally by the temporal subtraction images. Our observer test indicated that the use of temporal subtraction images was also clearly beneficial, even in nonnodular cases. One reason for this result seems to be that observers were not detrimentally influenced by some of the misregistration artifacts on the temporal subtraction images. It is not difficult for observers to distinguish between misregistration artifacts and lung nodules by inspecting original chest radiographs. This also seems to indicate that observers can use temporal subtraction images effectively as a second opinion. For only one observer, temporal subtraction images did not improve diagnostic accuracy because of an increase in false-positive detection. This observer was not able to distinguish between lung nodules and minor artifacts related to misregistration artifacts on the temporal subtraction images, probably because of insufficient training on temporal subtraction images. It seems necessary for observers to be familiar with typical patterns of artifacts in order to reduce such interpretation errors. Malpractice litigation involving radiologists occurs more frequently as a result of missed diagnoses of lesions than as a result of any other type of radiologic work. In particular, lung cancer lesions have been reported as the most frequently missed lesions leading to such litigation (15–18). Therefore, in clinical settings, radiologists may hesitate in diagnosing normal cases with confidence. In our study, 24% of nonnodular cases were affected beneficially by use of the temporal subtraction images. This finding was due to the increase in the observers’ confidence in diagnosing absence of nodules in nonnodular cases with use of the temporal subtraction images; it means a decrease in false-positive findings with use of chest radiographs alone. We separately analyzed Az values obtained from the certified radiologists and radiology residents to determine whether the effect of the temporal subtraction im150
䡠
Radiology
䡠
July 2002
ages on an observer’s performance was dependent on clinical experience. The results indicated that use of the temporal subtraction images was more beneficial for the radiology residents than for the certified radiologists; the Az value for the radiology residents using temporal subtraction images was almost equal to that for the certified radiologists. In addition, it is important to note that the effect of using the temporal subtraction images for the certified radiologists also was beneficial. Misregistration artifacts are caused by mismatching normal anatomic structures in current and previous images and are due mostly to the ribs, diaphragm, clavicle, and blood vessels. Ishida et al (19) reported that misregistration artifacts on temporal subtraction images obtained at screening examinations were markedly reduced with use of an improved temporal subtraction scheme based on a shift-vector analysis and an iterative image warping technique. In our study, by using the same scheme, we evaluated chest radiographs obtained from daily clinical practice. Results of the subjective evaluation regarding the quality of the temporal subtraction images showed that they were all of very good to reasonably good quality, although the average interval between current and previous radiograph acquisition was 308 days. Although temporal subtraction images with moderate misregistration artifacts need to be improved, 85% of the temporal subtraction images are very good– or good-quality images that would be adequate for clinical use. Most of the misregistration artifacts observed in the current study were caused by normal anatomic structures. We used the best-matched temporal subtraction images. Although this may have reduced the proportion of cases with misregistration artifacts, we believe that it did not affect the results of the study. In clinical settings, it is possible to choose the best match from multiple previous images in cases with more than one previous radiograph. However, at the time this article was written, to our knowledge, a best-match selection process could not be automated. The temporal subtraction system has been installed in our department, and its daily routine use is being tested in practical diagnostic procedures. Currently, temporal subtraction images can be produced in approximately 15 seconds. This time will be shortened with the anticipated progress of computer technology. In clinical settings, it is important for radiologists to use temporal subtraction images effec-
tively as a second opinion. We believe radiologists will be able to easily achieve this skill by gaining experience in reading temporal subtraction images. Our study had several limitations. It is likely that lateral chest radiographs would have been helpful for detecting nodules. To our knowledge, however, temporal subtraction images of lateral views cannot be produced at this time. We used only posteroanterior views to make comparisons simple; also, some cases lacked lateral chest radiographs. This may not reflect the clinical reality for the usefulness of temporal subtraction. “Satisfaction of search” has been suggested as a possible source of error in reading radiographs with multiple abnormalities: It has been used to refer to a phenomenon in which detection of one abnormality interferes with detection of other abnormalities on the same radiograph (20 –23). Before our observer test, the radiologists were informed that objects of detection were solitary nodules 7–30 mm in diameter and that reading time was not limited. Therefore, the radiologists’ attention was focused specifically on the task of detecting solitary lung nodules. This may have caused the underestimation of the false-positive rate in the current study. Acknowledgments: The authors are grateful to Takatoshi Aoki, MD, Toshimi Uozumi, MD, Satoshi Kawanami, MD, Suzushi Kusano, MD, Takayuki Ohguri, MD, Syuichiro Yamamoto, MD, Kazuhisa Iida, MD, and Katsuya Yahara, MD, for participating as observers. References 1. Heelan RT, Flehinger BJ, Melamed MR, et al. Non-small-cell lung cancer: results of the New York screening program. Radiology 1984; 151:289 –293. 2. Forrest JV, Friedman PJ. Radiologic errors in patients with lung cancer. West J Med 1981; 134:485– 490. 3. Naidich DP, Zerhouni EA, Siegelman SS. Computed tomography of the thorax. New York, NY: Raven, 1984; 171–199. 4. Brogdon BG, Kelsey CA, Moseley RD Jr. Factors affecting perception of pulmonary lesions. Radiol Clin North Am 1983; 21:633– 654. 5. Revesz G, Kundel HL, Graber MA. The influence of structured noise on the detection of radiologic abnormalities. Invest Radiol 1974; 9:479 – 486. 6. Kundel HL, Revesz G. Lesion conspicuity, structure noise, and film reader error. AJR Am J Roentgenol 1976; 126:1233–1238. 7. Sorenson JA, Mitchell CR, Armstrong JD Jr, et al. Effects of improved contrast on lung-nodule detection: a clinical ROC study. Invest Radiol 1987; 22:772–780. 8. Austin JH, Romney BM, Goldsmith LS. Missed bronchogenic carcinoma: radiographic findings in 27 patients with a potentially resectable lesion evident in retrospect. Radiology 1992; 182:115–122. 9. Noguchi M, Morikawa A, Kawasaki M, et al. Small adenocarcinoma of the lung: Kakeda et al
Radiology
10.
11.
12.
13.
histologic characteristics and prognosis. Cancer 1995; 75:2844 –2852. 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. Difazio MC, MacMahon H, Xu XW, et al. Digital chest radiography: effect of temporal subtraction images on detection accuracy. Radiology 1997; 202:447– 452. Ishida T, Ashizawa K, Engelmann R, Katsuragawa S, MacMahon H, Doi K. Application of temporal subtraction for detection of interval changes on chest radiographs: improvement of subtraction images using automated initial image matching. J Digit Imaging 1999; 12: 77– 86. Kobayashi T, Xu XW, MacMahon H, Metz CE, Doi K. Effect of a computer-aided diagnosis scheme on radiologists’ perfor-
Volume 224
䡠
Number 1
14.
15. 16. 17.
18. 19.
mance in detection of lung nodules on radiographs. Radiology 1996; 199:843– 848. Metz CE, Herman BA, Shen JH. Maximum-likelihood estimation of receiver operating (ROC) characteristic curves from continuously-distributed data. Stat Med 1998; 17:1033–1053. Berlin L. Malpractice and radiologists, update 1986: an 11.5-year perspective. AJR Am J Roentgenol 1986; 147:1291–1298. Spring DB, Tennenhouse DJ. Radiology malpractice lawsuits: California jury verdicts. Radiology 1986; 159:811– 814. Hamer MM, Morlock F, Foley HT, Ros PR. Medical malpractice in diagnostic radiology: claims, compensation, and patient injury. Radiology 1987; 164:263–266. Dehlen RT, Foley HT. Medical malpractice claims in diagnostic radiology: update (letter). Radiology 1989; 170:277. Ishida T, Katsuragawa S, Nakamura K,
20.
21.
22.
23.
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. Samuel S, Kundel HL, Nodine CF, Toto LC. Mechanism of satisfaction of search: eye position recording in the reading of chest radiographs. Radiology 1995; 194: 895–902. Tuddenham WJ. Visual search, image organization, and reader error in roentgen diagnosis: studies of the psychophysiology of roentgen image perception. Radiology 1962; 78:694 –704. Tuddenham WJ. Problem of perception in chest roentgenology: facts and fallacies. Radiol Clin North Am 1963; 1:277– 289. Smith MJ. Error and variation in diagnostic radiology. Springfield, Ill: Thomas, 1967; 80:103.
Improved Detection of Lung Nodules
䡠
151
