W. Dennis Foley, MD
Diagnosis
with CT at an Electronic
I N this issue of Radiology, Straub et al have demonstrated comparable di agnostic performance by radiologists in the recognition or exclusion of single abnormalities on abdominal computed tomographic (CT) studies interpreted either on a single-screen workstation or on multiformat film (1). This is an en couraging result and differs from the results of a recent report by Foley et al (2). However, in the study by Foley et al, a more demanding interpretative task (recognition of multiple abnormal ities on all CT studies, one-half of which involved comparisons with pri or examinations) and use of a less func tional cathode ray tube (CRT) display probably affected the results. In the study of Straub et al, a 19-inch diameter, 2,000 X 1,500-line CRT capa ble of a 12:1 display of 512 X 512 matrix images was used. The display was flick er-free, with rapid interactive window level control. More important, the disk to-display writing time for a full screen of images was 0.35 seconds, which a!lowed the interpreting radiologist to rapidly “¿page― through each study. Presuming an average of 20—30images per abdominal CT examination, only two to three screen changes were nec essary to review all the information. In addition, each radiologist understood that only one abnormality would be found in each positive case. This com bination of a rapid display and a rela tively simple diagnostic task allowed the single-screen CRT system to be
Index terms: Computed tomography (CT), computer programs, 70.1211 •¿ Computed to mography (CT), image display and recording, 70.1211 •¿ Computed tomography (CT), image processing, 70.1211 •¿ Editorials •¿ Diagnostic ra diology, observer performance •¿ Picture ar chiving and communication system (PACS), 70. 1211 Radiology
1991; 178:631-632
‘¿Fromthe Department waukee County Medical 8700 W Wisconsin Aye, Received December 13, ber 14. Address reprint ‘¿c RSNA,
of Radiology, Mil Complex, Box 151, Milwaukee, WI 53226. 1990; accepted Decem requests to the author.
1991
See also the article by Straub et al (pp 739743) in this issue.
used as efficiently and accurately as a conventional multifilm, multiformat display. By comparison, in the muitiscreen
study by Foley et al, each of the two CRT screens used in the study could display only four 512 X 512 matrix im ages, which allowed eight CT images to be displayed simultaneously at full res olution. More important, the disk-to display
writing
time
for each
set of
eight images was 11 seconds. The in ability to rapidly page through a CT study was a relative distraction about which each of the observers comment ed adversely. In addition, the observa tion task was more complex, requiring detection of between one and eight ab normalities per case. Thus, an observer could not be satisfied when only one abnormality had been observed. The necessity to remember multiple obser vations during the slow “¿pagination― process increased the difficulty of the diagnostic task. In addition, use of a two-screen display to evaluate interval changes in the observed abnormalities in one-half of the cases further in creased the complexity of the study; not only did the observer have to de tect multiple abnormalities during rela tively
slow
pagination,
but he or she
also had to arrange the display on the second screen in such a way as to re view scan sections at comparable 1ev els. The article by Straub et al indicates that as current technology is imple mented to make the convenience, mul tiplicity, and resolution of the electron ic environment comparable to those associated with multiformat film, ra diologists' diagnostic skills should not be adversely affected. Another interest ing facet of the study by Straub et al was the use of a sequential mode, la beled the “¿click-and-move― mode, to review an examination on the CRT dis play. Radiologists' performance when reviewing single images in rapid se quence was equivalent to their perfor mance with the mosaic 12:1 display of both CRT images and multiformat film. Interestingly, when reviewing in the sequential mode, radiologists were more comfortable with a larger image than with a smaller image that was equal in size to images used in the mo saic 12:1 display; nevertheless, there
Workstation' was no gain, but even a slight drop, in diagnostic accuracy when the larger image was used. To my knowledge, radiologists have not received in-depth exposure to use of either sequential or cine modes for the interpretation
of multiple,
contigu
ous, cross-sectional CT images. Inter preting a cross-sectional imaging study is, in essence, an exercise in the visual ization of three-dimensional anatomy and pathology. On the basis of the re suits of the study by Straub et al, it ap pears that radiologists are capable of performing this task, at least when they are looking for single abnormalities, with use of either the sequential or mo saic mode of display. However, one cannot necessarily conclude from this single experience with the detection of single abnormalities that radiologists will be equally facile when detecting multiple abnormalities of different pathologic types (eg, infarction, fluid collection, organ contour abnormality) with
use of the sequential
viewing
mode when compared with use of the mosaic configuration. The necessity to search for only a sin gle abnormality in the study of Straub et al was dictated by the receiver oper ating characteristic methodology. In addition, cases were categorized as ei ther “¿subtle― or “¿typical.― A lesion was labeled “¿subtle― if it was the subjective impression of the radiologist selecting the case that the abnormality would likely be missed “¿at least 20% of the time if read by an experienced radiolo gist in a clinical
environment.―
This
judgment is neither objective nor quan tifiable. A similar problem has existed for many years regarding observer de tection experiments that made use of clinical
images,
specifically
chest
radio
graphs of interstitial lung disease (3—5). Some studies have demonstrated lower performance in the detection of inter stitial lung disease at resolutions less than 2,000 X 2,000 (0.2-mm pixel size) than at 2,000 X 2,000 or larger matrix size (5). However, other studies have come to the opposite conclusion (3,4). These discrepant results are probably due to differences in case material and observers, as well as in the viewing en vironments. Whether observers in the future can be as accurate in detecting subtle abnormalities on a CRT display 631
with use of either a mosaic or sequen tial mode as they are with use of film requires further corroboration. Although the viewing environment in the study by Straub et al was superi or in terms of disk-to-display writing time when compared with that in the study by Foley et al, some important
features of the CRT performance—lu minance, contrast scale, and 12-bit depth of image memory—were similar. In general, the luminance of a conven tional CRT is an order of magnitude less than that of a multiformat film viewed on a conventional view box. Despite the well-known concept of ret ma! adaptation to lower luminance 1ev els, there are no clinical studies that demonstrate whether low-contrast le sions can be detected as accurately on CT studies displayed on a film panel al ternator as they can be detected on a CRT. However, the ability to vary im age brightness and contrast on a CRT display through the use of window level controls may well compensate for any deficiencies in the CRT mode and may possibly improve a radiologist's ability to detect low-contrast lesions. Room light reflected from the face plate of a CRT display is illuminance, which degrades the ability to perceive contrast. This means that radiologists who use electronic workstations for primary diagnosis should be aware that their eyes will adapt to a lower light level and that interactive window-level controls should be used as required. In addition, incident light on the face plate of the monitor should be reduced to a minimum. Studies comparing the use of CRT displays with film displays should not only state the luminance and contrast scale but also the average illuminance of the device.
632 •¿ Radiology
Accurate interpretation through the use of electronic workstations is the prime requirement for implementation of a successful picture archiving and communications system (PACS). The use of efficient and user-friendly com puter devices, with primary emphasis on speed, resolution, and multiplicity, either in mosaic or sequential mode, will be required. However, these dis play stations must be linked to imaging devices that allow on-line review of current cases and reports. Efficient in teraction of imaging systems and net work interface devices, as well as a mu tually agreed-on data format for access to images from many different yen dors, is required. I believe that use of the American College of Radiology, National Electrical Manufacturers As sociation data format is a prerequisite to achieving this goal. Efficient retriev al from long-term and intermediate ar chives of comparison examinations and accompanying reports will be neces sary. Image reformatting and distance and region-of-interest measurements are simple analytic tools that should be available to the image readers. The number and orientation of the CRT screens required for the interpretation of complex examinations such as mag netic resonance (MR) imaging, particu larly for comparison with prior studies, are other issues. In addition, the use of multiple minified images as a reference pictorial index when interpreting stud ies with use of only a limited number of display screens should be explored (6). In the specialty of radiology, the pur pose of performing diagnostic tasks ac curately and efficiently is to provide a faster conduit for consultations by re ferring clinicians by sharing both the radiologic images and the text of re ports. Interactive consultations be tween clinicians and radiologists and
use of artificial intelligence techniques should add to the flexibility and utility of any PACS. If PACS is to develop into a viable system, the major components of dis play, networking, and archiving will need to be tested initially in the con trolled “¿mini-PACS― environment of CT and MR imaging. The article by Straub et al illustrates that the use of current computer technology at an electronic workstation to implement speed, resolution, and multiplicity in both mosaic and sequential modes can be the first step in the development of such a system. U
References 1. Straub WH, Cur D, Good WF, et al. Prima ry CT diagnosis of abdominal masses in a PACS environment. Radiology 1991; 178:739-743. 2. Foley WD, Jacobson DR. Taylor AJ, et a!. Display of CT studies on a two-screen elec tronic workstation versus a film panel a!ternator: sensitivity and efficiency among radiologists. Radiology 1990; 174:769-773. 3. Seeley CW, Fisher HO, Stempski MO, eta!. Total digital radiology department: special resolution requirements. AJR 1987; 148: 421-426. 4. Lams PN, Cocklin ML. Spatial resolution requirements with digital chest radio graphics: ROC study of observer perfor mance in selected cases. Radiology 1986; 158:11—19. 5. MacMahon H, Vyborny CJ, Metz CE, Doi K, Sabegi V, Solomon S. Digital radiography of subtle pulmonary abnormalities: ROC study of the effect of pixel size on observer performance. Radiology 1986; 158:21-26. 6. Beard 0. Critical characteristics of radiolo gy workstation navigation. In: Arenson RU, Friedenberg RM, eds. Computer applica tions to assist radiology. Carlsbad, Calif: Symposia Foundation, 1990; 324-330.
March 1991
Diagnosis
with CT at an Electronic
I N this issue of Radiology, Straub et al have demonstrated comparable di agnostic performance by radiologists in the recognition or exclusion of single abnormalities on abdominal computed tomographic (CT) studies interpreted either on a single-screen workstation or on multiformat film (1). This is an en couraging result and differs from the results of a recent report by Foley et al (2). However, in the study by Foley et al, a more demanding interpretative task (recognition of multiple abnormal ities on all CT studies, one-half of which involved comparisons with pri or examinations) and use of a less func tional cathode ray tube (CRT) display probably affected the results. In the study of Straub et al, a 19-inch diameter, 2,000 X 1,500-line CRT capa ble of a 12:1 display of 512 X 512 matrix images was used. The display was flick er-free, with rapid interactive window level control. More important, the disk to-display writing time for a full screen of images was 0.35 seconds, which a!lowed the interpreting radiologist to rapidly “¿page― through each study. Presuming an average of 20—30images per abdominal CT examination, only two to three screen changes were nec essary to review all the information. In addition, each radiologist understood that only one abnormality would be found in each positive case. This com bination of a rapid display and a rela tively simple diagnostic task allowed the single-screen CRT system to be
Index terms: Computed tomography (CT), computer programs, 70.1211 •¿ Computed to mography (CT), image display and recording, 70.1211 •¿ Computed tomography (CT), image processing, 70.1211 •¿ Editorials •¿ Diagnostic ra diology, observer performance •¿ Picture ar chiving and communication system (PACS), 70. 1211 Radiology
1991; 178:631-632
‘¿Fromthe Department waukee County Medical 8700 W Wisconsin Aye, Received December 13, ber 14. Address reprint ‘¿c RSNA,
of Radiology, Mil Complex, Box 151, Milwaukee, WI 53226. 1990; accepted Decem requests to the author.
1991
See also the article by Straub et al (pp 739743) in this issue.
used as efficiently and accurately as a conventional multifilm, multiformat display. By comparison, in the muitiscreen
study by Foley et al, each of the two CRT screens used in the study could display only four 512 X 512 matrix im ages, which allowed eight CT images to be displayed simultaneously at full res olution. More important, the disk-to display
writing
time
for each
set of
eight images was 11 seconds. The in ability to rapidly page through a CT study was a relative distraction about which each of the observers comment ed adversely. In addition, the observa tion task was more complex, requiring detection of between one and eight ab normalities per case. Thus, an observer could not be satisfied when only one abnormality had been observed. The necessity to remember multiple obser vations during the slow “¿pagination― process increased the difficulty of the diagnostic task. In addition, use of a two-screen display to evaluate interval changes in the observed abnormalities in one-half of the cases further in creased the complexity of the study; not only did the observer have to de tect multiple abnormalities during rela tively
slow
pagination,
but he or she
also had to arrange the display on the second screen in such a way as to re view scan sections at comparable 1ev els. The article by Straub et al indicates that as current technology is imple mented to make the convenience, mul tiplicity, and resolution of the electron ic environment comparable to those associated with multiformat film, ra diologists' diagnostic skills should not be adversely affected. Another interest ing facet of the study by Straub et al was the use of a sequential mode, la beled the “¿click-and-move― mode, to review an examination on the CRT dis play. Radiologists' performance when reviewing single images in rapid se quence was equivalent to their perfor mance with the mosaic 12:1 display of both CRT images and multiformat film. Interestingly, when reviewing in the sequential mode, radiologists were more comfortable with a larger image than with a smaller image that was equal in size to images used in the mo saic 12:1 display; nevertheless, there
Workstation' was no gain, but even a slight drop, in diagnostic accuracy when the larger image was used. To my knowledge, radiologists have not received in-depth exposure to use of either sequential or cine modes for the interpretation
of multiple,
contigu
ous, cross-sectional CT images. Inter preting a cross-sectional imaging study is, in essence, an exercise in the visual ization of three-dimensional anatomy and pathology. On the basis of the re suits of the study by Straub et al, it ap pears that radiologists are capable of performing this task, at least when they are looking for single abnormalities, with use of either the sequential or mo saic mode of display. However, one cannot necessarily conclude from this single experience with the detection of single abnormalities that radiologists will be equally facile when detecting multiple abnormalities of different pathologic types (eg, infarction, fluid collection, organ contour abnormality) with
use of the sequential
viewing
mode when compared with use of the mosaic configuration. The necessity to search for only a sin gle abnormality in the study of Straub et al was dictated by the receiver oper ating characteristic methodology. In addition, cases were categorized as ei ther “¿subtle― or “¿typical.― A lesion was labeled “¿subtle― if it was the subjective impression of the radiologist selecting the case that the abnormality would likely be missed “¿at least 20% of the time if read by an experienced radiolo gist in a clinical
environment.―
This
judgment is neither objective nor quan tifiable. A similar problem has existed for many years regarding observer de tection experiments that made use of clinical
images,
specifically
chest
radio
graphs of interstitial lung disease (3—5). Some studies have demonstrated lower performance in the detection of inter stitial lung disease at resolutions less than 2,000 X 2,000 (0.2-mm pixel size) than at 2,000 X 2,000 or larger matrix size (5). However, other studies have come to the opposite conclusion (3,4). These discrepant results are probably due to differences in case material and observers, as well as in the viewing en vironments. Whether observers in the future can be as accurate in detecting subtle abnormalities on a CRT display 631
with use of either a mosaic or sequen tial mode as they are with use of film requires further corroboration. Although the viewing environment in the study by Straub et al was superi or in terms of disk-to-display writing time when compared with that in the study by Foley et al, some important
features of the CRT performance—lu minance, contrast scale, and 12-bit depth of image memory—were similar. In general, the luminance of a conven tional CRT is an order of magnitude less than that of a multiformat film viewed on a conventional view box. Despite the well-known concept of ret ma! adaptation to lower luminance 1ev els, there are no clinical studies that demonstrate whether low-contrast le sions can be detected as accurately on CT studies displayed on a film panel al ternator as they can be detected on a CRT. However, the ability to vary im age brightness and contrast on a CRT display through the use of window level controls may well compensate for any deficiencies in the CRT mode and may possibly improve a radiologist's ability to detect low-contrast lesions. Room light reflected from the face plate of a CRT display is illuminance, which degrades the ability to perceive contrast. This means that radiologists who use electronic workstations for primary diagnosis should be aware that their eyes will adapt to a lower light level and that interactive window-level controls should be used as required. In addition, incident light on the face plate of the monitor should be reduced to a minimum. Studies comparing the use of CRT displays with film displays should not only state the luminance and contrast scale but also the average illuminance of the device.
632 •¿ Radiology
Accurate interpretation through the use of electronic workstations is the prime requirement for implementation of a successful picture archiving and communications system (PACS). The use of efficient and user-friendly com puter devices, with primary emphasis on speed, resolution, and multiplicity, either in mosaic or sequential mode, will be required. However, these dis play stations must be linked to imaging devices that allow on-line review of current cases and reports. Efficient in teraction of imaging systems and net work interface devices, as well as a mu tually agreed-on data format for access to images from many different yen dors, is required. I believe that use of the American College of Radiology, National Electrical Manufacturers As sociation data format is a prerequisite to achieving this goal. Efficient retriev al from long-term and intermediate ar chives of comparison examinations and accompanying reports will be neces sary. Image reformatting and distance and region-of-interest measurements are simple analytic tools that should be available to the image readers. The number and orientation of the CRT screens required for the interpretation of complex examinations such as mag netic resonance (MR) imaging, particu larly for comparison with prior studies, are other issues. In addition, the use of multiple minified images as a reference pictorial index when interpreting stud ies with use of only a limited number of display screens should be explored (6). In the specialty of radiology, the pur pose of performing diagnostic tasks ac curately and efficiently is to provide a faster conduit for consultations by re ferring clinicians by sharing both the radiologic images and the text of re ports. Interactive consultations be tween clinicians and radiologists and
use of artificial intelligence techniques should add to the flexibility and utility of any PACS. If PACS is to develop into a viable system, the major components of dis play, networking, and archiving will need to be tested initially in the con trolled “¿mini-PACS― environment of CT and MR imaging. The article by Straub et al illustrates that the use of current computer technology at an electronic workstation to implement speed, resolution, and multiplicity in both mosaic and sequential modes can be the first step in the development of such a system. U
References 1. Straub WH, Cur D, Good WF, et al. Prima ry CT diagnosis of abdominal masses in a PACS environment. Radiology 1991; 178:739-743. 2. Foley WD, Jacobson DR. Taylor AJ, et a!. Display of CT studies on a two-screen elec tronic workstation versus a film panel a!ternator: sensitivity and efficiency among radiologists. Radiology 1990; 174:769-773. 3. Seeley CW, Fisher HO, Stempski MO, eta!. Total digital radiology department: special resolution requirements. AJR 1987; 148: 421-426. 4. Lams PN, Cocklin ML. Spatial resolution requirements with digital chest radio graphics: ROC study of observer perfor mance in selected cases. Radiology 1986; 158:11—19. 5. MacMahon H, Vyborny CJ, Metz CE, Doi K, Sabegi V, Solomon S. Digital radiography of subtle pulmonary abnormalities: ROC study of the effect of pixel size on observer performance. Radiology 1986; 158:21-26. 6. Beard 0. Critical characteristics of radiolo gy workstation navigation. In: Arenson RU, Friedenberg RM, eds. Computer applica tions to assist radiology. Carlsbad, Calif: Symposia Foundation, 1990; 324-330.
March 1991
