- Email: [email protected]
CRT diagnosis of pulmonary disease: influence of monitor brightness and room illuminance on observer performance
Computerized Medical Imaging and Graphics PERGAMON
Computerized Medical Imaging and Graphics 26 (2002) 181±185
www.elsevier.com/locate/compmedimag
CRT diagnosis of pulmonary disease: in¯uence of monitor brightness and room illuminance on observer performance Shunichi Ishihara a,*, Kazuhiro Shimamoto b,1, Mitsuru Ikeda c,2, Katsuhiko Kato a, Yoshine Mori a, Tsuneo Ishiguchi a, Takeo Ishigaki a b
a Department of Radiology, Nagoya University School of Medicine, 65 Tsuruma-cho, Showa-ku, Nagoya 466-8550, Japan Department of Radiological Technology, Nagoya University School of Health Sciences, 1-1-20 Daikominami, Higashi-ku, Nagoya 461-8673, Japan c Department of Medical Information and Records, Nagoya University Hospital, 65 Tsuruma-cho, Showa-ku, Nagoya 466-8550, Japan
Received 10 September 2001; accepted 17 December 2001
Abstract Using a 21-in. cathode ray tube (CRT) monitor (2048 £ 2560 £ 8 bits), six radiologists interpreted 12 images with interstitial lung disease under six conditions of CRT luminance (50 and 400 cd/m 2) and room illuminance (20, 120 and 480 lx), and 10 radiologists interpreted 25 images with pulmonary nodules under nine conditions of CRT luminance (50, 200 and 500 cd/m 2) and room illuminance (20, 120 and 480 lx). Observer's performance for interstitial disease was relatively better at 120 lx. Four hundred and eighty lux illuminance with 50 cd/ m 2 CRT luminance, which degraded the detectability of pulmonary nodule signi®cantly (p , 0.05), should be avoided for clinical use. Published by Elsevier Science Ltd. Keywords: Observer performance; Cathode ray tube display; Image interpretation; Computed radiography; Chest radiography
1. Introduction
2. Materials and methods
With current technology, cathode ray tube (CRT) reading can be as ef®cient and accurate as ®lm interpretation, and it is now widely accepted in medical practice. However, there are several factors which affect observer performance, including CRT luminance and room illuminance [1±4]. It should be an essential requirement for CRT reading to establish the optimal reading environment for clinical use. Also, the degradation of a CRT monitor will cause a decline in the brightness level, and correlation between CRT luminance and the observer's performance should be discussed from the viewpoint of the safety of CRT diagnosis. Accordingly, this study is designed to evaluate the in¯uence of CRT luminance and room illuminance on observer diagnostic performance in the detection of interstitial pulmonary diseases and pulmonary nodules.
2.1. Experiment 1 (interstitial pulmonary diseases)
* Corresponding author. Tel.: 181-52-744-2327; fax: 181-52-744-2335. E-mail address: [email protected] (S. Ishihara). 1 Tel.: 181-52-719-1562; fax: 181-52-719-1509. 2 Tel.: 181-52-744-2666; fax: 181-52-744-2973. 0895-6111/02/$ - see front matter. Published by Elsevier Science Ltd. PII: S 0895-611 1(02)00004-6
Twelve hemithoraces were used for test images. All images were selected from the test images used in the previous study [5]. Six were normal cases, and six had interstitial disease (subtle pulmonary disease in two cases, moderate abnormality in two cases and obvious shadow in two cases). Twenty-one-inch CRT monitor, RS 252 (Konica Ltd, Tokyo) with a resolution of 2048 £ 2560 £ 8 bits and a maximum luminance of 512 cd/m 2 was used for the study. CRT luminance was set at 50 cd/m 2 (14.6 fL) and 400 cd/m 2 (116.8 fL) using a luminance meter (Konica, Dome Imaging System), and room illuminance was set at 20, 120 and 480 lx at the console desk using an illuminance meter (TOPCON, digital illuminance meter IM-3). Six experienced chest radiologists (clinical experience: 15±35 years) interpreted 12 test images under six kinds of combination of CRT luminance and room illuminance. Image reading sessions were conducted in the viewing workstation room of our department. The interval between reading sessions was about 30 min. Prior to the experiment, the reader was informed that the images might contain interstitial
182
S. Ishihara et al. / Computerized Medical Imaging and Graphics 26 (2002) 181±185
Fig. 1. Brier score comparison among six viewing conditions (interstitial disease). (A) CRT luminance was ®xed; (B) room illuminance was ®xed. Note: Observer's performance was relatively better at 120 lx than at 20 and 480 lx illuminance.
pulmonary diseases of various grades. Observers were asked to determine the con®dence level for the presence or absence of interstitial pulmonary disease with a continuously distributed scale and subjective evaluation of the image quality with one of the ®ve levels (1 poor, 3 the lowest level acceptable for clinical use, 5 excellent). The order of 12 test images was randomized for each reading session. No time limit was imposed for reading. The tone scale was ®xed. For statistical analysis, Brier score (see Appendix A) was used for the comparison of observer performance in the detection of signals. In this study, the lower the Brier score, the better the observer's performance. For our experimental design, the differences among six kinds of reading conditions should be evaluated, and therefore, the analysis of variance (ANOVA) of pseudo-values computed by the jackknife analysis method was used. For subjective feeling, the ANOVA was employed. 2.2. Experiment 2 (pulmonary nodules) One normal computed radiography (CR) chest image was
selected from our clinical database, and used for generating 25 test images using a SUN workstation. The lung ®eld was divided into four parts: right upper, right lower, left upper and left lower (the border between the upper and the lower zone is the upper margin of the seventh rib). In 50 of 100 quadrants, one small nodule between 10 and 16 mm in diameter was inserted electronically with the previously reported method [6]. Then, the number of simulated nodule in one test image varied from 0 to 4. The same CRT monitor as Experiment 1 was used. CRT luminance was set at 50 cd/m 2 (14.6 fL), 200 cd/m 2 (58.4 fL), and 500 cd/m 2 (146 fL), and room illuminance was set at 20, 120, and 480 lx at the console desk. In a total of nine combinations of CRT luminance and room illuminance, reading sessions were held. The tone scale was ®xed. To eliminate the learning effect of an image showed just before, unrelated pictures were displayed between test images. The order of 25 test images was randomized for each reading session. Also, the order of nine reading conditions was randomized for each observer. Reading time for each test image was limited within 30 s. The interval between reading sessions was about 30±60 min and
Fig. 2. Subjective evaluation of image quality. (A) CRT luminance was ®xed; (B) room illuminance was ®xed. Note: The higher CRT luminance, or the darker the room illuminance, the better the score of subjective estimation.
S. Ishihara et al. / Computerized Medical Imaging and Graphics 26 (2002) 181±185
183
Fig. 3. Brier score comparison among nine viewing conditions (pulmonary nodule). (A) CRT luminance was ®xed; (B) room illuminance was ®xed. Note: Observer's performance with 50 cd/m 2 CRT at 480 lx illuminance was signi®cantly worse when compared with that in the other eight conditions (p , 0.05).
these experiments were ®nished in 2 days. Readers were 10 chest radiologists (clinical experience: 10±33 years). Observers were asked to determine the con®dence level for the presence or absence of pulmonary nodules in 100 parts using a continuously distributed scale and to note the location of the nodule on a schematic illustration of the lung. For statistical analysis, the same way as Experiment 1 was employed. 3. Results 3.1. Experiment 1 Among six conditions, Brier score varied from 0.196 to 0.242. At both 50 and 400 cd/m 2 CRT luminance, Brier score was the lowest at 120 lx room illuminance without statistically signi®cant difference (Fig. 1). As for the subjective evaluation, 400 cd/m 2 at 20 lx was the best, and 50 cd/ m 2 at 480 lx was the worst. Whether 50 or 400 cd/m 2 luminance, the average score at 20 lx was signi®cantly better than at 480 lx (p , 0.05; Fig. 2). 3.2. Experiment 2 Among nine sessions, Brier score varied from 0.1692 to 0.2174 (Fig. 3). With a 50 cd/m 2 CRT monitor, Brier score at 20 and 120 lx was signi®cantly lower than that at 480 lx (p , 0.05). In other words, observer performance with 50 cd/m 2 CRT at 480 lx illuminance was signi®cantly worse when compared with that in the other eight conditions. However, there was no signi®cant difference in Brier score among the three different level of room illuminance with both 200 and 500 cd/m 2 CRT. 4. Discussion As for the in¯uence of room illuminance on the observer's performance, the detectability of small pulmonary nodules at 170 lx illuminance was better than at 70 and
480 lx illuminance [1]. In Experiment 1, the observer's performance in detecting interstitial lung disease was relatively better at 120 lx than at 20 and 480 lx illuminance. In Experiment 2, the observer's performance in detecting lung nodule was worse at 480 lx than at 20 and 120 lx when the CRT luminance was 50 cd/m 2. However, there was no signi®cant difference in the observer's performance at 200 and 500 cd/m 2. These results would show that the optimum room illuminance level exists at the intermediate level. In Experiment 1, the darker the room illuminance, the better the subjective estimation. The discrepancy between the observer's performance and the subjective estimation would be explained by the following reasons: (1) image contrast may be improved under lower ambient light; and (2) low room illuminance may reduce the performance during a long-time image interpretation due to mental or eye fatigue [1]. In general, the luminance of CRT monitor as well as the intensity of the ambient light affects the physiological response of the eye in image perception [2]. In Experiment 1, the observer's performance in detecting interstitial lung disease was relatively better at 400 cd/m 2 than at 50 cd/m 2 when the room illuminance was 20 lx. There was no change in the observer's performance at 120 and 480 lx. In Experiment 2, the observer's performance in detecting lung nodule was worse at 50 cd/m 2 than at 200 and 500 cd/m 2 when the room illuminance was 480 lx. However, there was no significant difference in the observer's performance at 20 and 120 lx. Krupinski et al. [3] reported that observer performance in mammogram interpretation was better at 140 fL (480 cd/m 2) than at 80 fL (274 cd/m 2) without a statistically signi®cant difference, and that viewing time was signi®cantly longer at 80 fL than at 140 fL. Song et al. [4] also reported that observer performance in detecting solitary pulmonary nodules was slightly better at 100 fL (343 cd/ m 2) than at 65 fL (223 cd/m 2) without a statistically signi®cant difference. These reports, suggesting that higher the CRT luminance, the better the observer performance, are compatible with our results. In Experiment 1, the subjective estimation was better at 400 cd/m 2 than at 50 cd/m 2. When
184
S. Ishihara et al. / Computerized Medical Imaging and Graphics 26 (2002) 181±185
CRT luminance was changed, the observer's performance paralleled subjective evaluation. These results suggested that higher the CRT luminance, the better the observer performance. However, the lower limit of CRT luminance was not established. In Experiment 2, our results suggested that a combination of high room illuminance (480 lx) and low CRT luminance (50 cd/m 2) can affect the detectability of the pulmonary nodule. However, except for this condition, there was no statistically signi®cant difference in Brier score, and the change of neither CRT luminance nor room illuminance in¯uenced the observer's performance. The discrepancy between the previous study [1] and current study existed, but, all observers in this study were highly experienced diagnostic radiologists. The human visual performance of such experts would have a wide range of adaptability to the ambient light level and CRT luminance, and unfavorable conditions may not degrade their ability of reading so much. At 20 and 120 lx, the change of CRT luminance did not have an in¯uence on observer performance. Both Krupinski et al. [3] and Song et al. [4] reported that higher the CRT luminance, relatively better is the observer performance. This discrepancy would be explained by the same reason as a wide range of adaptability. To reduce learning effect, we used only one normal CR chest image, and unrelated pictures were displayed between test images. However, in a short-time reading session, learning effect could not be eliminated completely. Also, much longer reading time will be necessary to produce severe eye fatigue or mental fatigue. In Experiment 2, it took about 30 min in one reading session for each observer. In clinical practice, reading time should be longer than reading sessions, and therefore, physical or mental fatigue may play a little role in this study. Experiments with much longer time will be necessary to evaluate effect of fatigue on the observer's performance. In conclusion, our results suggested that close interaction between CRT luminance and room illuminance may exist in CRT reading. CRT luminance and room illuminance affected the observer's diagnostic performance in detecting subtle pulmonary diseases as well as pulmonary nodules. Both low CRT luminance and high room illuminance should be avoided for clinical use. Besides for CRT luminance and room illuminance, task and the different type of images in the experiment designs will affect the observer's performance. Further studies will be required to clarify the optimal reading environment.
diseases and pulmonary nodules and to establish the optimal reading environment for clinical use. Twenty-one-inch CRT monitor with a resolution of 2048 £ 2560 £ 8 bits and a maximum luminance of 512 cd/m 2 was used for the study. Six chest radiologists interpreted 12 hemithoraces under six kinds of combination of CRT luminance (50 and 400 cd/m 2 and room illuminance 20, 120, and 480 lx at the console desk), and described the con®dence level for the presence or absence of interstitial pulmonary disease with a continuously distributed scale. Brier score analysis showed that the observer's performance in detecting interstitial lung disease was relatively better at 120 lx than at 20 and 480 lx illuminance. Similarly, 10 chest radiologists interpreted 25 test images with electronically inserted pseudo-nodules under nine conditions of CRT luminance (50, 200, and 500 cd/m 2) and room illuminance (20, 120 and 480 lx). Combination of high room illuminance (480 lx) and low CRT luminance (50 cd/m 2) degraded the detectability of the pulmonary nodule signi®cantly (p , 0.05). This study would show that the optimum room illuminance level exists in intermediate level, and also suggested that higher the CRT luminance, the better the observer performance. In conclusion, CRT luminance and room illuminance affected the observer's diagnostic performance in detecting subtle pulmonary diseases as well as pulmonary nodules. Both low CRT luminance and high room illuminance should be avoided for clinical use.
5. Summary
BS
This study was designed to evaluate the in¯uence of CRT luminance and room illuminance on the observer's diagnostic performance in the detection of interstitial pulmonary
In this study, the lower the BS, the better the observer's performance.
Acknowledgements This study was supported by a grant funded by the Ministry of Health, Labor, and Welfare in Japan.
Appendix A The Brier score (BS) is one of the best-known scoring rules [7,8]. Brier index provides a score that quali®es the accuracy of a set of judgments, by comparing the expressed probabilities to the actual outcomes, and it has been applied to medical studies [9]. Let yi indicate the true state of the event, such that yi 0 if the event is nonsignal and yi 1 if the event is signal, where the subscript i indexes the individual event (1) [9]. Further, let pi denote the observers (or physicians) probability estimate that the ith event will be the signal one [9]. The de®nition of BS is as follows [9,10]: M 1 X y 2 pi 2 M i1 i
1
S. Ishihara et al. / Computerized Medical Imaging and Graphics 26 (2002) 181±185
References [1] Itoh Y, Ishigaki T, Sakuma S, et al. In¯uence of CRT workstation on observer's performance. Comput Meth Progm Biomed 1992; 37(4):253±8. [2] Wang J, Langer S. A brief review of human perception factors in digital displays for picture archiving and communications systems. J Digit Imag 1997;10(4):158±68. [3] Krupinski EA, Roehrig H, Furukawa T, Tang C. In¯uence of monitor luminance and tone scale on observer detection performance. Proc SPIE 1998;3340:99±104. [4] Song KS, Lee JS, Kim HY, Lim TH. Effect of monitor luminance on the detection of solitary pulmonary nodule: ROC analysis. Proc SPIE 1999;3663:212±6. [5] Ishigaki T, Endo T, Ikeda M, et al. Subtle pulmonary disease: detection with computed radiography versus conventional chest radiography. Radiology 1996;201(1):51±60. [6] Itoh S, Ikeda M, Arahata S, et al. Lung cancer screening: Minimum tube current required for helical CT. Radiology 2000;215(1):175±83. [7] Ikeda M, Itoh S, Ishigaki T, Yamauchi K. Application of resampling techniques to the statistical analysis of the Brier score. Meth Inform Med 2001;40(3):259±64. [8] Poses RM, Cebul RD, Centor RM. Evaluating physicians' probabilistic judgments. Med Decis Making 1988;8(4):233±40. [9] Redelmeier DA, Bloch DA, Hickam DH. Assessing predictive accuracy: how to compare Brier scores. J Clin Epidemiol 1991;44(11):1141±6. [10] Spiegelhalter DJ. Probabilistic prediction in patient management and clinical trials. Stat Med 1986;5(5):421±33.
Shunichi Ishihara was born in Tsushima, Aichi Prefecture, Japan on August 19, 1969. He graduated from Nagoya University, School of Medicine, Japan in 1994. Further, he graduated from the Graduate School Division of Medical Research, Nagoya University, Japan in 2000. He is currently on the staff in the Department of Radiology, Toyohashi Municipal Hospital.
Kazuhiro Shimamoto was born in Gamagori, Aichi Prefecture, Japan on December 8, 1958. He graduated from Nagoya University, School of Medicine, Japan in 1983. He received a PhD degree in 1992 from Nagoya University. He is currently an Associate Professor in the Department of Radiological Technology, Nagoya University School of Health Sciences. He is a member of the Radiological Society of North America, and the American Institute of Ultrasound in Medicine.
185
Mitsuru Ikeda graduated from Tokyo Kougyo University in March 1977 with a Bachelor of Engineering degree. Further, he graduated from Nagoya University School of Medicine in March 1982 with a Bachelor of Medicine degree. He received a PhD degree in March 1986 from Nagoya University. From 1986 to 1992, he was an Assistant Professor in the Department of Radiology, Nagoya University Hospital. At present, he is an Associate Professor in the Department of Medical Information and Medical Records, Nagoya University Hospital. Dr Ikeda is a member of the Radiological Society of North America and the Japan Radiological Society.
Katsuhiko Kato was born in Nagoya, Aichi Prefecture, Japan on February 28, 1960. He graduated from Nihon University School of Technology, Department of Civil Engineering, Japan in 1984. He graduated from Kawasaki Medical School, Japan in 1995. He received a PhD degree in 2000 from Nagoya University. He is currently on the medical staff in the Department of Radiology, Nagoya University School of Medicine. He is a member of the Radiological Society of North America and the Society of Nuclear Medicine.
Yoshine Mori was born in Inazawa, Aichi Prefecture, Japan on November 6, 1971. He graduated from Tsukuba University, School of Medicine, Japan in 1997. He is a student of post-graduated school of Nagoya University School of Medicine. He is a member of the Japan Radiological Society.
Tsuneo Ishiguchi was born in Shirahama, Wakayama Prefecture, Japan on June 18, 1952. He graduated from Nagoya University School of Medicine, Japan in 1977. He received a PhD degree in 1990 from Nagoya University. He is currently an Associate Professor in the Department of Radiology, Aichi Medical University. He is a member of the Radiological Society of North America. His current research interests include 3D vascular imaging and endovascular stent-grafting for aortic aneurysms.
Takeo Ishigaki was born in Tokyo on April 17, 1942. He graduated from University of Tokyo, Faculty of Medicine in 1968. He received a PhD degree in 1976 with research on TV monitor diagnosis for the upper GI tract from Nagoya City University, Japan. At present, he is a Professor and a Chairman of the Department of Radiology, Nagoya University School of Medicine. He is a member of the Radiological Society of North America. His present research includes digital radiography, PACS, and telemedicine.
Computerized Medical Imaging and Graphics 26 (2002) 181±185
www.elsevier.com/locate/compmedimag
CRT diagnosis of pulmonary disease: in¯uence of monitor brightness and room illuminance on observer performance Shunichi Ishihara a,*, Kazuhiro Shimamoto b,1, Mitsuru Ikeda c,2, Katsuhiko Kato a, Yoshine Mori a, Tsuneo Ishiguchi a, Takeo Ishigaki a b
a Department of Radiology, Nagoya University School of Medicine, 65 Tsuruma-cho, Showa-ku, Nagoya 466-8550, Japan Department of Radiological Technology, Nagoya University School of Health Sciences, 1-1-20 Daikominami, Higashi-ku, Nagoya 461-8673, Japan c Department of Medical Information and Records, Nagoya University Hospital, 65 Tsuruma-cho, Showa-ku, Nagoya 466-8550, Japan
Received 10 September 2001; accepted 17 December 2001
Abstract Using a 21-in. cathode ray tube (CRT) monitor (2048 £ 2560 £ 8 bits), six radiologists interpreted 12 images with interstitial lung disease under six conditions of CRT luminance (50 and 400 cd/m 2) and room illuminance (20, 120 and 480 lx), and 10 radiologists interpreted 25 images with pulmonary nodules under nine conditions of CRT luminance (50, 200 and 500 cd/m 2) and room illuminance (20, 120 and 480 lx). Observer's performance for interstitial disease was relatively better at 120 lx. Four hundred and eighty lux illuminance with 50 cd/ m 2 CRT luminance, which degraded the detectability of pulmonary nodule signi®cantly (p , 0.05), should be avoided for clinical use. Published by Elsevier Science Ltd. Keywords: Observer performance; Cathode ray tube display; Image interpretation; Computed radiography; Chest radiography
1. Introduction
2. Materials and methods
With current technology, cathode ray tube (CRT) reading can be as ef®cient and accurate as ®lm interpretation, and it is now widely accepted in medical practice. However, there are several factors which affect observer performance, including CRT luminance and room illuminance [1±4]. It should be an essential requirement for CRT reading to establish the optimal reading environment for clinical use. Also, the degradation of a CRT monitor will cause a decline in the brightness level, and correlation between CRT luminance and the observer's performance should be discussed from the viewpoint of the safety of CRT diagnosis. Accordingly, this study is designed to evaluate the in¯uence of CRT luminance and room illuminance on observer diagnostic performance in the detection of interstitial pulmonary diseases and pulmonary nodules.
2.1. Experiment 1 (interstitial pulmonary diseases)
* Corresponding author. Tel.: 181-52-744-2327; fax: 181-52-744-2335. E-mail address: [email protected] (S. Ishihara). 1 Tel.: 181-52-719-1562; fax: 181-52-719-1509. 2 Tel.: 181-52-744-2666; fax: 181-52-744-2973. 0895-6111/02/$ - see front matter. Published by Elsevier Science Ltd. PII: S 0895-611 1(02)00004-6
Twelve hemithoraces were used for test images. All images were selected from the test images used in the previous study [5]. Six were normal cases, and six had interstitial disease (subtle pulmonary disease in two cases, moderate abnormality in two cases and obvious shadow in two cases). Twenty-one-inch CRT monitor, RS 252 (Konica Ltd, Tokyo) with a resolution of 2048 £ 2560 £ 8 bits and a maximum luminance of 512 cd/m 2 was used for the study. CRT luminance was set at 50 cd/m 2 (14.6 fL) and 400 cd/m 2 (116.8 fL) using a luminance meter (Konica, Dome Imaging System), and room illuminance was set at 20, 120 and 480 lx at the console desk using an illuminance meter (TOPCON, digital illuminance meter IM-3). Six experienced chest radiologists (clinical experience: 15±35 years) interpreted 12 test images under six kinds of combination of CRT luminance and room illuminance. Image reading sessions were conducted in the viewing workstation room of our department. The interval between reading sessions was about 30 min. Prior to the experiment, the reader was informed that the images might contain interstitial
182
S. Ishihara et al. / Computerized Medical Imaging and Graphics 26 (2002) 181±185
Fig. 1. Brier score comparison among six viewing conditions (interstitial disease). (A) CRT luminance was ®xed; (B) room illuminance was ®xed. Note: Observer's performance was relatively better at 120 lx than at 20 and 480 lx illuminance.
pulmonary diseases of various grades. Observers were asked to determine the con®dence level for the presence or absence of interstitial pulmonary disease with a continuously distributed scale and subjective evaluation of the image quality with one of the ®ve levels (1 poor, 3 the lowest level acceptable for clinical use, 5 excellent). The order of 12 test images was randomized for each reading session. No time limit was imposed for reading. The tone scale was ®xed. For statistical analysis, Brier score (see Appendix A) was used for the comparison of observer performance in the detection of signals. In this study, the lower the Brier score, the better the observer's performance. For our experimental design, the differences among six kinds of reading conditions should be evaluated, and therefore, the analysis of variance (ANOVA) of pseudo-values computed by the jackknife analysis method was used. For subjective feeling, the ANOVA was employed. 2.2. Experiment 2 (pulmonary nodules) One normal computed radiography (CR) chest image was
selected from our clinical database, and used for generating 25 test images using a SUN workstation. The lung ®eld was divided into four parts: right upper, right lower, left upper and left lower (the border between the upper and the lower zone is the upper margin of the seventh rib). In 50 of 100 quadrants, one small nodule between 10 and 16 mm in diameter was inserted electronically with the previously reported method [6]. Then, the number of simulated nodule in one test image varied from 0 to 4. The same CRT monitor as Experiment 1 was used. CRT luminance was set at 50 cd/m 2 (14.6 fL), 200 cd/m 2 (58.4 fL), and 500 cd/m 2 (146 fL), and room illuminance was set at 20, 120, and 480 lx at the console desk. In a total of nine combinations of CRT luminance and room illuminance, reading sessions were held. The tone scale was ®xed. To eliminate the learning effect of an image showed just before, unrelated pictures were displayed between test images. The order of 25 test images was randomized for each reading session. Also, the order of nine reading conditions was randomized for each observer. Reading time for each test image was limited within 30 s. The interval between reading sessions was about 30±60 min and
Fig. 2. Subjective evaluation of image quality. (A) CRT luminance was ®xed; (B) room illuminance was ®xed. Note: The higher CRT luminance, or the darker the room illuminance, the better the score of subjective estimation.
S. Ishihara et al. / Computerized Medical Imaging and Graphics 26 (2002) 181±185
183
Fig. 3. Brier score comparison among nine viewing conditions (pulmonary nodule). (A) CRT luminance was ®xed; (B) room illuminance was ®xed. Note: Observer's performance with 50 cd/m 2 CRT at 480 lx illuminance was signi®cantly worse when compared with that in the other eight conditions (p , 0.05).
these experiments were ®nished in 2 days. Readers were 10 chest radiologists (clinical experience: 10±33 years). Observers were asked to determine the con®dence level for the presence or absence of pulmonary nodules in 100 parts using a continuously distributed scale and to note the location of the nodule on a schematic illustration of the lung. For statistical analysis, the same way as Experiment 1 was employed. 3. Results 3.1. Experiment 1 Among six conditions, Brier score varied from 0.196 to 0.242. At both 50 and 400 cd/m 2 CRT luminance, Brier score was the lowest at 120 lx room illuminance without statistically signi®cant difference (Fig. 1). As for the subjective evaluation, 400 cd/m 2 at 20 lx was the best, and 50 cd/ m 2 at 480 lx was the worst. Whether 50 or 400 cd/m 2 luminance, the average score at 20 lx was signi®cantly better than at 480 lx (p , 0.05; Fig. 2). 3.2. Experiment 2 Among nine sessions, Brier score varied from 0.1692 to 0.2174 (Fig. 3). With a 50 cd/m 2 CRT monitor, Brier score at 20 and 120 lx was signi®cantly lower than that at 480 lx (p , 0.05). In other words, observer performance with 50 cd/m 2 CRT at 480 lx illuminance was signi®cantly worse when compared with that in the other eight conditions. However, there was no signi®cant difference in Brier score among the three different level of room illuminance with both 200 and 500 cd/m 2 CRT. 4. Discussion As for the in¯uence of room illuminance on the observer's performance, the detectability of small pulmonary nodules at 170 lx illuminance was better than at 70 and
480 lx illuminance [1]. In Experiment 1, the observer's performance in detecting interstitial lung disease was relatively better at 120 lx than at 20 and 480 lx illuminance. In Experiment 2, the observer's performance in detecting lung nodule was worse at 480 lx than at 20 and 120 lx when the CRT luminance was 50 cd/m 2. However, there was no signi®cant difference in the observer's performance at 200 and 500 cd/m 2. These results would show that the optimum room illuminance level exists at the intermediate level. In Experiment 1, the darker the room illuminance, the better the subjective estimation. The discrepancy between the observer's performance and the subjective estimation would be explained by the following reasons: (1) image contrast may be improved under lower ambient light; and (2) low room illuminance may reduce the performance during a long-time image interpretation due to mental or eye fatigue [1]. In general, the luminance of CRT monitor as well as the intensity of the ambient light affects the physiological response of the eye in image perception [2]. In Experiment 1, the observer's performance in detecting interstitial lung disease was relatively better at 400 cd/m 2 than at 50 cd/m 2 when the room illuminance was 20 lx. There was no change in the observer's performance at 120 and 480 lx. In Experiment 2, the observer's performance in detecting lung nodule was worse at 50 cd/m 2 than at 200 and 500 cd/m 2 when the room illuminance was 480 lx. However, there was no significant difference in the observer's performance at 20 and 120 lx. Krupinski et al. [3] reported that observer performance in mammogram interpretation was better at 140 fL (480 cd/m 2) than at 80 fL (274 cd/m 2) without a statistically signi®cant difference, and that viewing time was signi®cantly longer at 80 fL than at 140 fL. Song et al. [4] also reported that observer performance in detecting solitary pulmonary nodules was slightly better at 100 fL (343 cd/ m 2) than at 65 fL (223 cd/m 2) without a statistically signi®cant difference. These reports, suggesting that higher the CRT luminance, the better the observer performance, are compatible with our results. In Experiment 1, the subjective estimation was better at 400 cd/m 2 than at 50 cd/m 2. When
184
S. Ishihara et al. / Computerized Medical Imaging and Graphics 26 (2002) 181±185
CRT luminance was changed, the observer's performance paralleled subjective evaluation. These results suggested that higher the CRT luminance, the better the observer performance. However, the lower limit of CRT luminance was not established. In Experiment 2, our results suggested that a combination of high room illuminance (480 lx) and low CRT luminance (50 cd/m 2) can affect the detectability of the pulmonary nodule. However, except for this condition, there was no statistically signi®cant difference in Brier score, and the change of neither CRT luminance nor room illuminance in¯uenced the observer's performance. The discrepancy between the previous study [1] and current study existed, but, all observers in this study were highly experienced diagnostic radiologists. The human visual performance of such experts would have a wide range of adaptability to the ambient light level and CRT luminance, and unfavorable conditions may not degrade their ability of reading so much. At 20 and 120 lx, the change of CRT luminance did not have an in¯uence on observer performance. Both Krupinski et al. [3] and Song et al. [4] reported that higher the CRT luminance, relatively better is the observer performance. This discrepancy would be explained by the same reason as a wide range of adaptability. To reduce learning effect, we used only one normal CR chest image, and unrelated pictures were displayed between test images. However, in a short-time reading session, learning effect could not be eliminated completely. Also, much longer reading time will be necessary to produce severe eye fatigue or mental fatigue. In Experiment 2, it took about 30 min in one reading session for each observer. In clinical practice, reading time should be longer than reading sessions, and therefore, physical or mental fatigue may play a little role in this study. Experiments with much longer time will be necessary to evaluate effect of fatigue on the observer's performance. In conclusion, our results suggested that close interaction between CRT luminance and room illuminance may exist in CRT reading. CRT luminance and room illuminance affected the observer's diagnostic performance in detecting subtle pulmonary diseases as well as pulmonary nodules. Both low CRT luminance and high room illuminance should be avoided for clinical use. Besides for CRT luminance and room illuminance, task and the different type of images in the experiment designs will affect the observer's performance. Further studies will be required to clarify the optimal reading environment.
diseases and pulmonary nodules and to establish the optimal reading environment for clinical use. Twenty-one-inch CRT monitor with a resolution of 2048 £ 2560 £ 8 bits and a maximum luminance of 512 cd/m 2 was used for the study. Six chest radiologists interpreted 12 hemithoraces under six kinds of combination of CRT luminance (50 and 400 cd/m 2 and room illuminance 20, 120, and 480 lx at the console desk), and described the con®dence level for the presence or absence of interstitial pulmonary disease with a continuously distributed scale. Brier score analysis showed that the observer's performance in detecting interstitial lung disease was relatively better at 120 lx than at 20 and 480 lx illuminance. Similarly, 10 chest radiologists interpreted 25 test images with electronically inserted pseudo-nodules under nine conditions of CRT luminance (50, 200, and 500 cd/m 2) and room illuminance (20, 120 and 480 lx). Combination of high room illuminance (480 lx) and low CRT luminance (50 cd/m 2) degraded the detectability of the pulmonary nodule signi®cantly (p , 0.05). This study would show that the optimum room illuminance level exists in intermediate level, and also suggested that higher the CRT luminance, the better the observer performance. In conclusion, CRT luminance and room illuminance affected the observer's diagnostic performance in detecting subtle pulmonary diseases as well as pulmonary nodules. Both low CRT luminance and high room illuminance should be avoided for clinical use.
5. Summary
BS
This study was designed to evaluate the in¯uence of CRT luminance and room illuminance on the observer's diagnostic performance in the detection of interstitial pulmonary
In this study, the lower the BS, the better the observer's performance.
Acknowledgements This study was supported by a grant funded by the Ministry of Health, Labor, and Welfare in Japan.
Appendix A The Brier score (BS) is one of the best-known scoring rules [7,8]. Brier index provides a score that quali®es the accuracy of a set of judgments, by comparing the expressed probabilities to the actual outcomes, and it has been applied to medical studies [9]. Let yi indicate the true state of the event, such that yi 0 if the event is nonsignal and yi 1 if the event is signal, where the subscript i indexes the individual event (1) [9]. Further, let pi denote the observers (or physicians) probability estimate that the ith event will be the signal one [9]. The de®nition of BS is as follows [9,10]: M 1 X y 2 pi 2 M i1 i
1
S. Ishihara et al. / Computerized Medical Imaging and Graphics 26 (2002) 181±185
References [1] Itoh Y, Ishigaki T, Sakuma S, et al. In¯uence of CRT workstation on observer's performance. Comput Meth Progm Biomed 1992; 37(4):253±8. [2] Wang J, Langer S. A brief review of human perception factors in digital displays for picture archiving and communications systems. J Digit Imag 1997;10(4):158±68. [3] Krupinski EA, Roehrig H, Furukawa T, Tang C. In¯uence of monitor luminance and tone scale on observer detection performance. Proc SPIE 1998;3340:99±104. [4] Song KS, Lee JS, Kim HY, Lim TH. Effect of monitor luminance on the detection of solitary pulmonary nodule: ROC analysis. Proc SPIE 1999;3663:212±6. [5] Ishigaki T, Endo T, Ikeda M, et al. Subtle pulmonary disease: detection with computed radiography versus conventional chest radiography. Radiology 1996;201(1):51±60. [6] Itoh S, Ikeda M, Arahata S, et al. Lung cancer screening: Minimum tube current required for helical CT. Radiology 2000;215(1):175±83. [7] Ikeda M, Itoh S, Ishigaki T, Yamauchi K. Application of resampling techniques to the statistical analysis of the Brier score. Meth Inform Med 2001;40(3):259±64. [8] Poses RM, Cebul RD, Centor RM. Evaluating physicians' probabilistic judgments. Med Decis Making 1988;8(4):233±40. [9] Redelmeier DA, Bloch DA, Hickam DH. Assessing predictive accuracy: how to compare Brier scores. J Clin Epidemiol 1991;44(11):1141±6. [10] Spiegelhalter DJ. Probabilistic prediction in patient management and clinical trials. Stat Med 1986;5(5):421±33.
Shunichi Ishihara was born in Tsushima, Aichi Prefecture, Japan on August 19, 1969. He graduated from Nagoya University, School of Medicine, Japan in 1994. Further, he graduated from the Graduate School Division of Medical Research, Nagoya University, Japan in 2000. He is currently on the staff in the Department of Radiology, Toyohashi Municipal Hospital.
Kazuhiro Shimamoto was born in Gamagori, Aichi Prefecture, Japan on December 8, 1958. He graduated from Nagoya University, School of Medicine, Japan in 1983. He received a PhD degree in 1992 from Nagoya University. He is currently an Associate Professor in the Department of Radiological Technology, Nagoya University School of Health Sciences. He is a member of the Radiological Society of North America, and the American Institute of Ultrasound in Medicine.
185
Mitsuru Ikeda graduated from Tokyo Kougyo University in March 1977 with a Bachelor of Engineering degree. Further, he graduated from Nagoya University School of Medicine in March 1982 with a Bachelor of Medicine degree. He received a PhD degree in March 1986 from Nagoya University. From 1986 to 1992, he was an Assistant Professor in the Department of Radiology, Nagoya University Hospital. At present, he is an Associate Professor in the Department of Medical Information and Medical Records, Nagoya University Hospital. Dr Ikeda is a member of the Radiological Society of North America and the Japan Radiological Society.
Katsuhiko Kato was born in Nagoya, Aichi Prefecture, Japan on February 28, 1960. He graduated from Nihon University School of Technology, Department of Civil Engineering, Japan in 1984. He graduated from Kawasaki Medical School, Japan in 1995. He received a PhD degree in 2000 from Nagoya University. He is currently on the medical staff in the Department of Radiology, Nagoya University School of Medicine. He is a member of the Radiological Society of North America and the Society of Nuclear Medicine.
Yoshine Mori was born in Inazawa, Aichi Prefecture, Japan on November 6, 1971. He graduated from Tsukuba University, School of Medicine, Japan in 1997. He is a student of post-graduated school of Nagoya University School of Medicine. He is a member of the Japan Radiological Society.
Tsuneo Ishiguchi was born in Shirahama, Wakayama Prefecture, Japan on June 18, 1952. He graduated from Nagoya University School of Medicine, Japan in 1977. He received a PhD degree in 1990 from Nagoya University. He is currently an Associate Professor in the Department of Radiology, Aichi Medical University. He is a member of the Radiological Society of North America. His current research interests include 3D vascular imaging and endovascular stent-grafting for aortic aneurysms.
Takeo Ishigaki was born in Tokyo on April 17, 1942. He graduated from University of Tokyo, Faculty of Medicine in 1968. He received a PhD degree in 1976 with research on TV monitor diagnosis for the upper GI tract from Nagoya City University, Japan. At present, he is a Professor and a Chairman of the Department of Radiology, Nagoya University School of Medicine. He is a member of the Radiological Society of North America. His present research includes digital radiography, PACS, and telemedicine.