Eur Radiol (2008) 18: 2148 2154 DOI 10.1007/s00330-008-1008-8 COMPUTER APPLICATIONS Yusuke Kawasumi Takayuki Yamada Hideki Ota Masahiro Tsuboi Kei Takase Akihiro Sato Shuichi Higano Tadashi Ishibashi Shoki Takahashi High-resolution monochrome liquid crystal display versus efficient household colour liquid crystal display: comparison of their diagnostic performance with unenhanced CT images in focal liver lesions Received: 6 July 2007 Revised: 4 March 2008 Accepted: 7 March 2008 Published online: 8 May 2008 # European Society of Radiology 2008 Y. Kawasumi (*). T. Yamada. H. Ota. M. Tsuboi. K. Takase. A. Sato. S. Higano. T. Ishibashi. S. Takahashi Department of Diagnostic Radiology, Tohoku University Graduate School of Medicine, 1 1 Seiryo, Aoba, Sendai, Miyagi, 980 8574, Japan e-mail: ssu@rad.med.tohoku.ac.jp Tel.: +81-22-7177312 Fax: +81-22-7177316 Abstract This study compared the ability of an efficient household liquid crystal display (LCD) and a medical high-resolution LCD to detect small hepatic lesions on unenhanced computed tomography (CT) images. We obtained the images from 100 subjects who had undergone abdominal CT. They consisted of 41 patients with a single space-occupying lesion (SOL) in the liver and 59 control subjects with no SOL. Independently, five radiologists rated their confidence concerning the presence of hepatic SOLs on a continuous scale from 0 to 1. Receiver-operating characteristic (ROC) analysis was performed using the jackknife method using the program LABMRMC. We evaluated the differences in A z based on the 95% confidence intervals. The mean A z of the five observers was 0.9594 with the efficient household LCD vs. 0.9335 with the medical high-resolution LCD. The difference was 0.0422, and the 95% confidence interval was 0.1101 to 0.0257 (p=0.2203). There was no significant difference in the A z value between the two types of LCDs. The diagnostic performance with the household LCD was comparable to that with the high-resolution LCD, implying that the former type of LCD can be used to diagnose CT images. Keywords Liver. Space-occupying lesion. Computed tomography. Receiver-operating characteristics curve Abbreviations MDCT: Multidetector-row helical computed tomography. CT: Computed tomography. MR: Magnetic resonance. SOL: Space-occupying lesion. CRT: Cathode ray tube. LCD: Liquid crystal display. SMPET: Society of Motion Picture and Television Engineers. PACS: Picture archiving communication system. ROC: Receiver-operating characteristic. Az: Area under ROC curve Introduction Currently, multidetector-row helical computed tomography (MDCT) generates an immense number of digital images, which radiologists are obliged to assess. This, together with developments in medical information systems, has driven the conversion of imaging diagnosis from film-based diagnosis to monitor-based diagnosis supported by a picture archiving communication system (PACS) associated with an efficient viewer system [1]. Furthermore, the cathode ray tube (CRT) display used for monitors has largely been superseded by liquid crystal displays (LCD) because the latter require less power and space. The use of high-resolution LCD is generally recommended for radiologic diagnosis; however, its cost is one of the obstacles to the widespread use of PACS [2, 3]. In addition, the degree of LCD resolution required for different imaging modalities is not clear. Not every type of radiological examination may need a high-resolution display [4 13]. If household LCDs could be used at least partially, PACS would become less expensive.
2149 This study compared the ability to detect small lesions on non-contrast CT images using a household LCD and a medical high-resolution LCD. Materials and methods This retrospective study was conducted according to the principles of the Declaration of Helsinki of the World Medical Association [14]. Informed consent was not required for our retrospective study, which involved review of previously obtained image data. Case selection As the abdomen is the site subjected to CT most frequently in our institution, we compared the diagnostic performance of a high-resolution monochrome LCD versus a household colour LCD for detecting small space-occupying lesions (SOLs) in the liver, because hepatic SOLs such as metastasis and haemangioma are often subtle abnormalities on unenhanced CT images. Our reporting system was searched for abdominal CT examinations performed between October 2003 and March 2005 to obtain study cases. A general radiologist with more than 10 years of experience and a radiology resident reviewed the cases and selected 100 subjects who had undergone abdominal CT examinations with and without intravenous contrast agents. They comprised 40 patients with a single SOL in the liver (positive findings) and 60 control subjects without a SOL. The presumptive diagnoses for the positive cases on unenhanced and enhanced CT examinations were hemangioma for 10 patients and liver metastasis for 30 patients (Figs. 1 and 2). The average age was 59.8 years (range, 13 83 years). They comprised 61 men and 39 women. The reconstruction slice thickness of the CT images assessed was 7 or 10 mm. viewer was MeIFIS Marosis Me-view (Pioneer, Japan) running on Windows 2000 (Microsoft, CA). Household LCD-based system The household LCD used in this study was a 19-inch colour monitor (S190M; Nanao, Japan) with a display of 1,280 1,024 lines, 32-bit colour, and a gamma-correction function. The maximum luminance and contrast ratio of the monitor were 250 cd/m 2 and 1,000:1, respectively. The graphics card used was a WinFast PX6800 TDH (32-bit colour; Leadtek Research, Taiwan). Although we could not measure the luminance, the luminance of the monitor was at least 250 cd/m 2. The greyscale tones were not necessarily adjusted to DICOM Part14 GSDF [15]. The software used for the DICOM viewer was efilm Version 1.5.3 (Merge Healthcare, Canada) running on Windows 2000 (Microsoft). For both monitors, the brightness and contrast controls were initially set at the default parameters and then reset by one of the authors, using a Society of Motion Picture and Television Engineers (SMPTE) test pattern [16], before the reading examination. Ambient lighting conditions were 20~30 lux for both devices. Using the two different LCDs, differences in the resolution, colour mode, colour depth, and luminance also had to be considered when assessing the images used. Medical high-resolution LCD-based system The medical high-resolution LCD used in the study was a 20.8-inch greyscale portrait monitor (ME315L plus; Totoku, Japan) with a display of 1,536 2,043 lines, an 11-bit greyscale, and a gamma-correction function (Table 1). The maximum luminance and contrast ratio of the monitor were 700 cd/m 2 and 900:1, respectively. The graphics card used was a Matrox MED 3MP DVI (10-bit greyscale; Matrox, Canada). We could not measure the luminance; however, as the manufacturer s specified luminance was 410 cd/m 2, the luminance of the monitor was approximately 400 cd/m 2. The greyscale tones were adjusted to Digital Imaging and Communications in Medicine (DICOM) Part 14 Greyscale Standard Display Function (GSDF) [15]. The software used for the DICOM Fig. 1 CT image of a liver haemangioma. There is a low-density area in segment 7 (white arrow)
2150 Assessing the images Five radiologists were recruited for the study. They had from 5 years to more than 25 years of experience in general radiology as well as an additional one or two subspecialties out of either gastrointestinal, thoracic, cardiovascular, or breast radiology. All of the observers used the medical high-resolution LCD in their first rating session and the household LCD in their second session, which was conducted at least 4 weeks later. There was no time limit for each reading session. The display layout was stack mode, and all the observers manually paged the CT images. The images had a fixed resolution with an equal number of pixels in both directions (512 512 pixels). All the observers were free to adjust the width and level of the window. Scoring method The observers rated their confidence in evaluating the presence of hepatic SOLs on a continuous scale from 0 to 1, where 0 implied that the observer was convinced that there was no SOL, and 1 implied that an SOL was absolutely present. In addition, they were also asked to mark the location of the lesion in a schema of the liver and to record its Couinaud segment. Statistical analysis Fig. 2 a An extremely subtle low-density area (haemangioma) in segment 6 on an unenhanced CT (white arrow). b On a contrastenhanced CT, the haemangioma is enhanced at the periphery and is more conspicuous (black arrow) Receiver-operating characteristic (ROC) analysis [4 12, 17, 18] was performed by the jackknife method using LABMRMC (Charles E. Metz, Department of Radiology, University of Chicago) [19]. The software computed the area under the ROC curve (A z ) for all five observers for each LCD. It also calculated the A z for each observer for each LCD. ROC curves were generated using plotroc (Charles E. Metz, Department of Radiology, University of Table 1 Features of the diagnostic imaging systems based on the two displays Monitor type High-resolution Household Colour tone 11-bit greyscale 32-bit colour Resolution 1,536 2,043 1,280 1,024 Brightness 700 cd/m 2 250 cd/m 2 Contrast ratio 900:1 1,000:1 Graphics card Matrox MED 3MP DVI 10-bit greyscale Leadtek Research Inc. WinFast PX6800 TDH 32-bit colour Image quality adjustment Use SMPTE test pattern Use SMPTE test pattern Use time <1 month <1 month Dicom viewer MeIFis Marosis Me-view efilm 1.5.3 Ambient lighting 20 30 lux 20 30 lux
2151 LCD. The A z values and 95% confidence intervals of the individual observers did not differ significantly (Table 3). The results of the ROC analysis revealed no significant difference in the diagnostic performance between the two types of LCD. Discussion Fig. 3 The receiver-operating characteristic (ROC) curves pooled for the five observers Chicago) [19]. We evaluated the differences in A z based on the 95% confidence intervals. Results All five observers scored 1 of the 60 control cases at an average of more than 0.8 points. Both radiologists who initially selected the subjects reviewed this case and confirmed a SOL in the liver (presumptive metastasis). As a result, the subjects in our study consisted of 41 positive cases and 59 control cases. For the positive study cases, there was no disagreement in the locations of the SOLs identified by the observers. Therefore, we skipped the analysis of the location ROC (LROC) in this study. The ROC curves of all five observers for each LCD are shown in Fig. 3, and the areas under the ROC curves (A z ) are presented in Table 2. The A z of the efficient household LCD was 0.9594 (standard error, 0.0251) and that of the medical high-resolution LCD was 0.9335 (standard error, 0.0450). The difference was 0.0422 (standard error, 0.0342), and the 95% confidence interval was 0.1101 to 0.0257 (p=0.23303). There was no significant difference in the area under the ROC curve (A z ) between the two types of This study found no significant difference between the diagnostic performance in detecting hepatic SOLs on unenhanced CT examinations using a standard household LCD or a medical high-resolution LCD. To rate LCD performance, specifications such as the resolution, colour mode (colour or greyscale), colour depth, gamma value, luminance, and contrast ratio are generally considered. The household LCD had a lower resolution (1,280 1,024 lines) than the medical LCD (1,536 2,043 lines) in our study. Despite this difference in resolution, there was no significant difference in the diagnostic performance between the two types of monitors. Given that CT images consist of 512 512 pixels, a 1-megapixel LCD may be sufficient to display images with the real number of pixels. Doyle et al. [4] compared the observer performance of detecting wrist fractures on computer radiographs (CR) using monitors with an even larger difference in resolution: the colour monitor of a personal computer (PC) (1,024 768 lines) versus a greyscale portrait monitor on a PACS workstation (1,728 2,304 lines). They found no significant difference in the ability to detect wrist fractures on the two displays. Usami et al. [6] compared the detectability of nodular lesion depicted on chest radiographs using four kinds of LCDs with different resolution (1,024 1,280, 1,200 1,600, 1,536 2,048, 2,048 2,560 lines) and a CRT (2,048 2,560). There were no significant differences in performance among the four LCDs and the CRT, although a radiograph generally requires higher resolution than a CT image to demonstrate fine structures. Based on the results of our study and those of Doyle et al. [4] and Usami et al. [6], high-resolution monitors do not appear necessary for displaying CT images. The difference in the colour mode of the monitors in the present study, i.e., the colour display of the household LCD versus the monochrome display of the medical highresolution LCD, did not result in a significant difference in Table 2 Areas under the receiver-operating characteristic (ROC) curve pooled for the five observers High-resolution LCD Household LCD Difference A z values 0.8959 0.9381-0.0422 Standard error 0.0450 0.0251 0.0342 95% confidence interval (-0.1101, 0.0257), p=0.2203
2152 Table 3 Areas under the receiver-operating characteristic (ROC) curve for each individual observer Observer A z values 95% confidence interval for difference High-resolution LCD Household LCD 1 0.9314 0.9647-0.0822 0.0156 2 0.7687 0.9148-0.3710 0.0788 3 0.9897 0.9839-0.0170 0.0286 4 0.9104 0.9433-0.2395 0.1737 5 0.8791 0.8836-0.0816 0.0725 diagnostic performance. Previous studies comparing colour versus monochrome monitors, including the study of Doyle et al. [4] and a study that assessed the detection of early brain infarction by CT using a colour LCD and a greyscale CRT (Pärtan et al. [5]), also found no significant difference in the diagnostic performance. The study of Pärtan et al. [5] was similar to ours in terms of detecting subtle hypodense lesions on non-contrast CT. Therefore, whether the display is colour or monochrome may not be important in interpreting CT examinations, although the black and white gradation of colour LCDs is generally considered less than that of monochrome LCDs because a colour filter is inserted in colour LCD and there is a decrease in brightness. Colour depth, which represents the feasibility of displaying minute differences in gradation, has never been neatly evaluated in radiological diagnosis, to our knowledge. In this study, the colour depth was thought to depend on the performance of the display itself, since the transmitted degradation in the colour depth of the image was negligible owing to the connection of a graphics card to the LCD through a digital visual interface (DVI). The difference in colour depth between the 8-bit depth of the household colour LCD and the 11-bit depth of the medical high-resolution LCD did not cause a significant difference in the ability to detect a subtle hypodense lesion on unenhanced CT images. This may be because both viewer systems we used were based on the Windows operating system (Microsoft), which can display colour depth only up to an 8-bit scale. More specifically, both the 8-bit LCD and 11-bit LCD simultaneously displayed only 256-step (8- bit) data, although the 11-bit LCD should theoretically have shown finer gradation than the 8-bit LCD because the former could select 256 steps of the 2,048-step (11-bit) data, whereas the latter just displayed all the data from the 256 steps (8-bit). Monitor luminance is one of the important factors affecting display performance. The American College of Radiology (ACR) [20] standard for digital image data management requires a luminance of at least 50 footlamberts (171.3 cd/m 2 ). In our study, the luminance values of the household LCD and medical high-resolution LCD were about 250 and 410 cd/m 2, respectively, both exceeding the recommended luminance. From our results, a difference in luminance may be regarded as causing only a negligible difference in diagnostic performance, if the recommended value is met. As the contrast ratios of both displays were comparable, the effect of the contrast ratio on diagnostic performance was assumed to be very little in this study. Generally, contrast of LCD is inferior to that of CRT, because black reproducibility by LCD is lower than that by CRT. Oschatz et al. [7] compared the detection performance of a monochrome CRT (2,560 2,048 lines) versus a greyscale LCD (1,536 2,048 lines) for simulated subtle pulmonary lesions on chest radiographs. They found no significant difference in the ability to detect subtle pulmonary lesions on the two displays. Based on the results of the study of Oschatz et al. [7], Pärtan et al. [5], and ours, LCD does not appear inferior to CRT in diagnostic performance on CR and CT. Medical image displays are commonly required to display greyscale according to the Greyscale Standard Display Function (GSDF) defined by DICOM Part 14 [15]. In our study, the household LCD was not adjusted to GSDF, while the medical high-resolution LCD was adjusted to it. It may cause the different greyscale tones between the two monitors. However, we believed that the greyscale tones did not differ significantly between them, because both had a gamma-correction function and were corrected by one of the authors using the SMPTE test pattern [16] before the reading examination. This study was designed to clarify whether household LCDs can be used to read CT studies. Although a variety of diagnostic actions are included in CT reading, e.g., detecting hypodense liver lesions or detecting small pulmonary nodules [5, 8], examining many diagnostic actions would not be realistic for a study. Therefore, we examined the ability to detect a subtle hypodense lesion in the liver on unenhanced CT studies, which is one of the most frequent diagnostic actions requiring careful observation in our institution, and found that a household LCD was sufficient for this purpose. In this study, we did not examine diagnostic actions using magnetic resonance (MR) imaging, in which a large number of images have also been generated, because of the development of parallel imaging. However, a household LCD may also be applicable for reading MR images
2153 because they have a similar number of pixels as CT images and generally have higher greyscale contrast than CT. Nevertheless, we do not believe that an extrapolation to digital mammography would be possible, because resolving fine flecks of calcification on mammograms requires much higher resolution, perhaps 5 megapixels [21]. There were several limitations to our study. First, the order of the reading sessions was not randomized. All the observers used the medical high-resolution LCD in their first rating session, followed by the household LCD. This order might have improved the outcome for the latter LCD if the observers had become accustomed to assessing the images. Second, the size of the image shown in each display differed because the pitch of the pixels differed for each display. In fact, some observers felt more comfortable observing the images on the household LCD than on the medical high-resolution LCD because of the larger size of the images. The two above-mentioned factors could possibly have conferred an advantage to the household LCD. In this study, the A z (area under the ROC curve) value of the household LCD was higher than that of the medical high-resolution LCD though there was no significant difference. A third factor is that the luminance of the two LCDs was not accurately measured. However, as described before, the two LCDs were on the factory default setting for luminance and had been used for less than 1 month, so it was guessed that the luminance of two LCDs was very near the manufacturer s specified luminance. Fourth, the display performance specifications were not compared independently, as has also been the case in reported studies. It would be practically impossible to set only one specification of the displays to differ while keeping the others identical owing to the limitations in the mechanical adjustment of currently available LCDs. In conclusion, we found no significant difference in the diagnostic performance of detecting liver SOLs on CT images between a household LCD and a medical highresolution LCD. The results imply that efficient household LCDs could be used without any difficulty for the diagnosis of CT images, eliminating the use of expensive high-resolution LCDs. If efficient household LCDs were used as monitors, PACS would become much more economical. References 1. Reiner BI, Siegel EL, Hooper FJ et al (2001) Radiologists productivity in the interpretation of CT scans: a comparison of PACS with conventional film. Am J Roentgenol 176:861 864 2. Wu TC, Lee SK, Peng CH et al (1999) An economical, personal computerbased picture archiving and communication system. RadioGraphics 19:523 530 3. Kolodny GM, Raptopolous V, Simon M et al (1999) A low-cost, full-function picture archiving and communication system using standard PC hardware and the traditional 4-over-4 display format. Am J Roentgenol 172:591 594 4. Doyle AJ, Fevre JL, Anderson GD (2005) Personal computer versus workstation display: observer performance in detection of wrist fractures on digital radiographs. Radiology 237:872 877 5. Partan G, Mayrhofer R, Urban M et al (2003) Diagnostic performance of liquid crystal and cathode-ray-tube monitors in brain computed tomography. Eur Radiol 13:2397 2401 6. Usami H, Ikeda M, Ishigaki T et al (2006) The influence of liquid crystal display (LCD) monitors on observer performance for the detection of nodular lesions on chest radiographs. Eur Radiol 16:726 732 7. Oschatz E, Prokop M, Scharitzer M et al (2005) Comparison of liquid crystal versus cathode ray tube display for the detection of simulated chest lesions. Eur Radiol 15:1472 1476 8. Ko JP, Rusinek H, Naidich DP et al (2003) Wavelet compression of lowdose chest CT data: effect on lung nodule detection. Radiology 228:70 75 9. Otto D, Bernhardt TM, Rapp-Bernhardt U et al (1998) Subtle pulmonary abnormalities: detection on monitors with varying spatial resolutions and maximum luminance levels compared with detection on storage phosphor radiographic hard copies. Radiology 207:237 242 10. Herron JM, Bender TM, Campbell WL et al (2000) Effects of luminance and resolution on observer performance with chest radiographs. Radiology 215:169 174 11. O Connor PJ, Davies AG, Fowler RC et al (1998) Reporting requirements for skeletal digital radiography: comparison of soft-copy and hard-copy presentation. Radiology 207:249 254 12. Kim AY, Cho KS, Song KS et al (2001) Urinary calculi on computed radiography: comparison of observer performance with hard-copy versus soft-copy images on different viewer systems. Am J Roentgenol 177:331 335 13. Slone RM, Muka E, Pilgram TK (2003) Irreversible JPEG compression of digital chest radiographs for primary interpretation: assessment of visually lossless threshold. Radiology 228:425 429 14. The World Medical Association (2004) World Medical Association Declaration of Helsinki Ethical Principles for Medical Research Involving Human Subjects. http://www.wma.net/e/policy/ pdf/17c.pdf Accessed 11 Oct 2004 15. Digital Imaging and Communications in Medicine (DICOM) (2007) DICOM Part14: Grayscale Standard Display Function (GSDF). http://medical.nema. org/dicom/2007/07_14pu.pdf Accessed 10 Jan 2007 16. Nawfel RD, Chan KH, Wagenaar DJ et al (1992) Evaluation of video gray-scale display. Medical Physics 19:561 567
2154 17. Metz CE (1986) ROC methodology in radiologic imaging. Invest Radiol 21:720 733 18. Obuchowski NA (2000) Sample size tables for receiver operating characteristic studies. Am J Roentgenol 175:603 608 19. Kurt Rossmann Laboratories for Radiologic Image Reserch, The University of Chicago Department of Radiology (2007) ROC SOFTWARE. http://www-radiology.uchicago.edu/krl/ roc_soft6.htm Accessed 27 Feb 2007 20. American College of Radiology (2007) Acr technical standard for electronic practice of medical imaging. http://www. acr.org/secondarymainmenucategories/ quality_safety/guidelines/med_phys/ electronic_practice.aspx Accessed 01 Oct 2007 21. Pisano ED, Cole EB, Kistner EO et al (2002) Interpretation of digital mammograms: comparison of speed and accuracy of soft-copy versus printedfilm display. Radiology 223:483 488