- Email: [email protected]
Low-cost digital teleradiology
EUROPEAN JOURNAL OF RADIOLOGY
ELSEVIER
European Journal of Radiology 19 (1995) 226-231
Low-cost digital teleradiology J. Reponen*, S. L/ihde, O. Tervonen, E. Ilkko, T. Rissanen, I. Suramo Department of Diagnostic Radiology, University Hospital of Oulu, Kajaanintie 50, FIN -- 90220 Oulu, Finland
Received 22 September 1994; revision received 22 November 1994;accepted 13 December 1994
Abstract
A teleradiology link based on standard personal computers and a flat-bed CCD scanner was tested. A 64 kbit/s dial-up digital ISDN telephone line was used for transmission. A total of 254 films (174 uncompressed, 80 compressed) were sent. Ninety-six per cent of the uncompressed images and 98% of the compressed images were considered technically acceptable. The total diagnostic agreement between the acceptable transmitted images and the original films was 98%. Image quality was sufficient for diagnosis in CT and conventional chest and bone radiographs. However, a 256-step gray scale of the scanner was not sufficient for demanding situations, such as overexposed images with a high contrast gradient. The average speed of transmission was 60 kbit/s, which was considered adequate. The tested system suggests that a teleradiology link based on standard personal computers and programs works in situations where instant consultation is needed. However, the image digitization time with the prototype system was quite long, and a better user interface is under development. Keywords: Computers, diagnostic aid; Radiography, technology; Video systems
1. Introduction
Many recent teleradiology experiments have been made with high-capacity equipment, which have been built for frequent transmission of large image sets. Various authors suggest the use of laser scanners with 2048 x 2048 pixels resolution together with 12 bits dynamics, high-end workstations with 2048 x 2048 pixel monitors and fast connection lines up to 1.544 Mbit/s speed [1-5]. The establishment of such teleradiology links may be non-profitable if only a small number of images are transmitted. The capacity of personal computers (PC) has improved, while computers have simultaneously become less expensive. It is also possible to apply inexpensive commercial programs, which can process images under standard graphical interfaces (MS-Windows by Microsoft Corporation, Seattle, WA, USA or Macintosh Finder by Apple Computer, Cupertino, CA, USA). Many of these programs have been primarily developed for print industry, but can display and modify any image in a standard image file format. * Corresponding author.
The purpose of this study was to find out whether teleradiology can be performed using low-cost personal computers together with a flat-bed image scanner and a digital dial-up telephone line. 2. Material and methods
The hardware consisted of a standard PC computer with an Intel (Santa Clara, CA, USA) 33-MHz 30486 processor. The images were scanned using a 300 dots per inch (DPI) flat-bed charged coupled device (CCD) image scanner (model XRS CX3 by X-Ray Scanner Corporation, Torrance, CA, USA) with a 21 x 35 cm scanning area thus giving a maximum of a 2478 × 4130 pixel matrix. In practice the image digitization matrix sizes varied from 512 x 512 to 2048 x 2048 pixels depending on the selected resolution and image size (Table 1). Each pixel had a 256-step (8 bits) gray scale. Large images (e.g. 35 x 35 cm films) were scanned in two parts and connected together. A commercial imageprocessing program was used for scanning and viewing (PhotoStyler by Aldus Corporation, Seattle, USA). The computer screen was 47 cm in diameter. Its resolution was 1024 x 768 pixels, but the image zooming and
0720-048X/95/$09.50 © 1995 ElsevierScience Ireland Ltd. All rights reserved SSD1 0720-048X(94)00602-Z
J. Reponen et al. /European Journal of Radiology 19 (1995) 226-231
227
Table i Scanning density, average sizes and average on-line transmission time for uncompressed images sent from a health centre (Phase 1) and a central hospital (Phase 2) and for compressed images sent from a regional hospital (Phase 3) Image
Scanning density (DPI)
Size (average, bytes)
Transmission time
Uncompressed
Compressed
Uncompressed
Compressed
150 300
3 897 973 2 981 370
NA NA
10 min 50 s 8 min 20 s
NA NA
100 100 150 250 150
840 390 842 210 1 653 345 1 883 547 2 273 610
NA NA NA NA NA
2 min 2 min 4 min 5 min 6 min
NA NA NA NA NA
129 165 150
3 608 680 2 034 236 2 537 205
180 438 184 685 172 733
NA NA NA
Health centre
Thorax Bone Central hospital
Head CT film Body CT film Angiography film Mammography film Ultrasound film Regional hospital Thorax Bone Mammography
20 s 20 s 32 s 14 s 20 s
30 s 31 s 29 s
NA, not applicable.
scrolling functions in the viewing program made it possible to interpret larger digitized images. The image data were transmitted via a dial-up digital 64 kbit/s one-channel ISDN (Integrated Services Digital Network) telephone line with the help of a communication program (Telelmage by Telecom Finland). The procedure of sending the images consisted of the following steps: (1) selecting the images, (2) prescanning and setting up parameters for the first image, (3) scanning, (4) saving the files, (5) making a connection, (6) sending the images. Image receiving was automatic. Both the online sending time and the total image transmission time were recorded. The PC-based teleradiology system was tested in three different system configurations (phases). The software interface and the use of compression differed in these three phases. In the first and second phases, all the programs had to be used separately and the scanning parameters were adjusted manually. No common user interface was used. In the first and second phases, moreover, all the images were sent uncompressed. Before the actual clinical cases, a small pilot series of 10 images was made to determine appropriate settings for image digitization. In the third phase, the effect of data compression and a new easier graphical user interface (programmed with Microsoft Visual Basic) on the total transfer time was tested. All the programs operated in a Microsoft Windows environment. No more than five buttons were available at each screen. Presets for the scanning parameters were used for different images. These presets included a possibility to decrease excessive contrast in chest and bone images. Image resolution for chest and
bone images was now set to the same level as in the commercial image plate system Fuji AC-I (Fuji Corporation, Tokyo, Japan). In this series, images were sent using lossy DCT (discrete cosine transform) image compression with approximately 1:10 compression results (Joint Photographer Experts Group, JPEG compression). (In the literature, DCT compression methods with slight compression between 1:10 and 1:25 depending on the quality of the original have given acceptable results with no statistical difference in diagnostic quality [6-8]. JPEG compression has been proposed for use in PACS environments [9].) The requests and reports were written with the Windows Write program (Microsoft Corporation, Seattle, USA) in a separate program window. A total of 254 films were sent, all to a university hospital. In the first phase 48 uncompressed images were sent from a rural health centre, and in the second phase 126 sets of uncompressed images were sent from a central hospital. In the third phase, 80 compressed images were sent from a regional hospital. The selection of patients was chosen so as to simulate as accurately as possible the real teleconsultation needs when radiological expertise is not available locally (Table 2). In the health centre, the images were scanned and sent by a computer operator, while in the central hospital and the regional hospital this was done by Xray technicians. At the university hospital, the images were interpreted by two staff radiologists on a computer screen. They first scored the images as acceptable or unacceptable in the technical respect. After that, a consensus diagnosis was made from the technically acceptable images. A report was given and returned to the sender via telefax in
228
J. Reponen et al./ European Journal of Radiology 19 (1995) 226-231
Table 2 Image file types, user interfacetype, reporting mode, examination type, selection of patients and total number of films sent in three different phases Image file type
User interface
Request and report
Examination
Selection
Separate interfaces Separate interfaces
Telefax Telefax
Thorax Bone
Clinically problematic cases Clinically problematic cases
38
Separate interfaces Separate interfaces Separate interfaces Separate interfaces Separate interfaces
Telefax Telefax Telefax Telefax Telefax
Head CT Consecutive emergencies Body CT Consecutive emergencies Angiography Consecutiveemergencies Mammography Consecutivescreening cases U l t r a s o u n d Consecutiveemergencies
55 38 25 4 4
Common graphical interface Common graphical interface Common graphical interface
Randomized emergencies Computerfile Thorax Randomized emergencies Computerfile Bone Computerfile Mammography Consecutive screening cases
20 20 40 254
10
Health center (Phase 1)
Uncompressed Uncompressed
Number of films
Central hospital (Phase 2)
Uncompressed Uncompressed Uncompressed Uncompressed Uncompressed Regional hospital (Phase 3)
Compressed Compressed Compressed Total
Phases 1 and 2 and as a computer file in Phase 3. The same radiologists inspected the original images 2 weeks later and again made a consensus diagnosis. The diagnoses were compared for possible differences. 3. Results
3.1. Timefactor The average total image transfer procedure time in Phase 1 (from the health centre) was 15 min 50 s (range 15 min to 18 min 20 s) for an uncompressed single chest X-ray image (average size 3897 bytes). The longest single task in the transmission process was the on-line sending of the image (average 10 min 50 s), which was followed by scanning (64 s, range 55-74 s) and saving the images on a hard disk (57 s, range 53-83 s). The average throughput of the digital line was 60 kbit/s (range 56-61 kbit/s). The average total image transfer procedure time for a CT film (four to six images each) with an average size of 840 kbytes in Phase 2 (from the central hospital) was 6 min 20 s (range 5 min 10 s to 7 min 20 s) per uncompressed film. The longest single step was sending the film (2 min 20 s), the next longest step was scanning (35 s, range 23-42 s), and then came saving the films on a hard disk (26 s, range 19-35 s). The average total image transfer procedure time for a 1:10 compressed chest X-ray film in Phase 3 (from the regional hospital) was 5 min for a single view (36-49% of the time for uncompressed images). Of this, only 30 s on average was on-line time (5-9% of that for uncompressed images). The average time required for the preparatory steps preceding the scanning has decreased from 3 min to 2 min 30 s. The on-line transmission times for the different image files are shown in Table 1.
3.2. Technical quality Of the 48 uncompressed films transmitted in Phase 1 (from the health centre), two chest X-rays and five bone images were considered technically unacceptable. This was due to faulty scanner settings, one chest X-ray had an excessively small scanning area and the other was initially digitized too light. Two thoracic and three cervical spine images were scanned with excessive contrast settings and details lost in both light and dark areas. All the uncompressed transmitted images in Phase 2 (from the central hospital) were considered technically high-quality. Of the 80 compressed films sent in Phase 3 (from the regional hospital), the only low-quality film was an image of the thoracic spine, which was digitized with excessive contrast settings and a consequent loss of details in the highlight parts. 3.3 Diagnostic ability The diagnostic agreement between the transmitted and original images was made in all of the technically acceptable chest (n = 8) and bone (n = 33) images transmitted uncompressed in Phase 1 (from the health center). A lesion was missed in two cases of the 126 image sets (2%) transmitted uncompressed in Phase 2 (from the central hospital). Both cases were brain CT scans (n = 55): the first was a small infarction in the cerebral pons (diameter < 5 mm), while the second was a thickening of the right nervus opticus (difference of 5 mm compared to the other side). All the 38 body CT sets as well as all the 25 angiography film sets, four mammographies and four ultrasound images were correctly interpreted from the transmitted images.
J. Reponen et al. / European Journal of Radiology 19 (1995) 226-231
A discrepant finding was made in four (5%) out of 79 image sets satisfactorily transmitted compressed in Phase 3 (from the regional hospital). All of them were mammography images (n = 40). After interpreting the original images, an enlargement mammography was considered necessary in two cases, while in two cases no further examinations were considered necessary even though an enlargement mammography was suggested after reading the transmitted images. All 20 chest images and all 19 bone images were interpreted correctly from the transmissions. 4. Discussion The earliest generation of teleradiology was based on the transmission of analog TV images, but the image quality was inadequate compared with the quality of film images [10,11]. Modern teleradiology began to develop after the introduction of digitized images. The early digitized systems captured video images with a 512 x 512 pixel matrix and an 8-bit dynamic range, while modern scanners enable resolution up to 4096 x 4096 pixel and a 12-bit gray scale [4,12,13]. Images scanned with 2048 x 2048 × 12 bit matrix have already been proposed for primary diagnosis [4]. The early digitized systems used telephone lines with a speed up to 9600 bits/s. Nowadays speeds from 28.8 kbit/s (phone line) through 64 kbit/s (single line ISDN) up to 1.544 Mbit/s (T-l links or Asyncronous Transfer Mode links) are possible [4,5]. In the cases of high-standard systems with resolution equal to the original films and a full 12-bit (4096 steps) gray scale with instant image transmission, the price for a single-site unit can be as high as 196 000 USD and a central referral unit may cost as much as 343 000 USD. Network services make for additional costs if permanent connections are used [ 1,2,14]. The cost of the equipment in this study was 20 000 USD for the transmitter and 15 000 USD for the receiver unit. This cost level makes teleradiology feasible for health centres and smaller hospitals. The type of dial-up one-channel ISDN lines used in this study is available in all parts of Finland. Their fixed monthly costs are only 20-100 USD and the connections are paid by use, the price being that of normal long distance calls. Only a few reports on the use of low-cost digital teleradiology have been published. Lear et al. [15] built a system based on prototype high-resolution CCD camera, a personal computer and an ISDN link. Their equipment and software were proprietary and not commercially available. Dohrmann et al. [16] used personal computers to establish a low-cost teleradiology link in Australia. He, however, used a video camera and a frame grabber to capture the images from printed films. A slow-speed (2400 bit/s) ordinary telephone line modem was used to transfer images. This solution limits
229
the maximum resolution and the transfer speed and is mostly suited to CT images. Yamamoto et al. [17] built their low-cost teleradiology equipment out of personal computers and hand-held scanners. Their scanning area was limited to 6.1 x 7.0 cm and an ordinary telephone line was used for transmission. The on-line sending time for a 10:1 compressed chest image (size before compression < 4 Mb) in this study was at the same 30-s level as reported by Goldberg et al. [4] for an uncompressed 7 Mb image over a fast 1.544 Mbit/s network. If efficient compression can be used, even ISDN networks can serve effectively. Only a few studies, however, discuss the total procedure time [13,15,18]. In our experience, if efficient compression is used, the digitization together with preparations took more of the total procedure time than the actual sending. The transmitted images were scored as technically acceptable with eight exceptions (3%). In seven cases the inadequacy was due to improper scanner settings or insufficient gray scale dynamics. Because our scanner limited the image gray scale to 8 bits (256 levels), such parameters like brightness and contrast are critical for good image quality. If the original images present excessive contrast, details in both the dark and the light areas may be lost. This means that such images should be sent in duplicate for the dark and light areas. Another solution is an image scanning program which can decrease image contrast by making the highlights darker and the shadows lighter. This method was used in Phase 3 (from the regional hospital). True 12-bit (4096 levels) scanners, which are not so operator dependent, are three to four times more expensive than our unit. Image spatial resolution was sufficient, except in mammograms. The maximum scanning density of the tested system is 300 dots per inch (DPI), which equals 11.8 pixels/mm. This exceeded the scanning density of the Fuji AC-I image plate system, for example. In practice, we finally set the scanning density to the same level as in the image plate system for chest (129 DPI = 5.1 pixels/mm) and bone (165 DPI = 6.5 pixels/mm) images. If we had used the maximum scanning density of the tested scanner all the time, the image sizes would have been too large for our equipment. A 35 x 35 cm chest X-ray with 300 DPI (11.8 pixels/mm) would require >17 Mb and could not be loaded into the 8-Mb memory of the computer. There were no significant differences in the diagnoses of technically acceptable chest or bone images. In two CT images, a small pons infarction and a slight thickening of nervus opticus were missed in the original screen reading. In retrospect, however, these lesions were detected. It seems that the errors were due to the reader's unfamiliarity with the computer screen. The opportunity of film printing might be an advantage. In mammograms, the error frequency was markedly great-
230
J. Reponen et al. / European Journal of Radiology 19 (1995) 226-231
er (10%), which suggests that the scanning density used in this study was not enough for a transmission of the most detailed films. Our results, with their total 98% agreement on the technically acceptable images, correspond well to the resuits by Carey et al. [18], who sent 489 laser-digitized conventional and US examinations with a 98% agreement between the original and transmitted images. It is notable that they still had 12 patients in whom disease was not correctly identified even though the original films were digitized with 12-bit gray scale. Slovis et al. [13] also had an agreement of 98% in their work with 4200 conventional pediatric images. Goldberg et al. [4] had discrepant interpretations in 18 (2.6%) of 685 cases while using a 12-bit scanner. When standard computer programs are adapted to for clinical work, some limitations are encountered. The programs have not been specifically designed for medical work and require many adjustments. In our first two phases, the participants themselves had to make all the preparatory steps and the changes from the viewing program to the transmission program. This was the reason why they let a computer operator send the images from the health center. This caused some quality problems because the computer operator was not accustomed to deciding on the technical quality of scanned images. The easier graphical interface, the "shell" used in Phase 3, resulted in more convenient image transmission. It eliminated the complicated steps and required less teaching. It made it possible for X-ray technicians to transmit the images as part of their normal daily routine without being too much concerned of the technology. This is important, because X-ray technicians cannot spend too much time on teleradiology efforts. The X-ray technicians were also able to keep a constant image quality. This is important for the cost-effectiveness of the system. If a teleradiology system is built on standard personal computers, one has to be aware of the limitations. Depending on the scanner, the gray scale dynamics and scanning density might not be enough for the most detailed images like mammograms. Also the smaller display unit necessitates the use of image zooming. If these problems could be resolved, the same central unit can be used for office automation, Medline literature searches and scientific image processing. Radiologists who are familiar with computers do not need to learn a new user interface, but can utilize the skills they already have. Images are also transferable into teaching programs and archives, for example. Our investigation suggested that a good image quality can be restored in the transmission of images with tested standard personal computers and a commonly available commercial software configuration. This could make it possible to build effective teleradiology links for special purposes with lower costs compared to the commercial
high end systems. According to our experience, the scanner is the most critical unit and its 8-bit gray scale should be extended. In our system, the total sending procedure still took a relatively long time and thus manoeuvrability needs to be further developed. The suitability of this system to mammography films also warrants further studies.
Acknowledgements The authors wish to thank the staff of the Kuusamo health centre, the Rovaniemi central hospital and the Raahe regional hospital for giving their valuable time for this study. The authors recognize the support given by Telecom Finland for this work in the form of computer equipment and communication programs.
References [1] Batnitzky S, Rosenthal SJ, Siegel EL, Wetzel LH, Murphey MD, Cox GG, McMillan JH, Templeton AW, Dwyer SJ 1II. Teleradiology: an assessment. Radiology 1990; 177:11-17. [2I Dwyer SJ III, Templeton AW, Batnitzky S. Teleradiology: costs of hardware and communications. AJR 1991; 156: 1279-1282. [3] Stewart BK, Dwyer SJ III, Kangafloo. Design of a high-speed, high resolution teleradiology network. J Digit Imaging 1992; 5: 144-155. [4] Goldberg MA, Rosenthal DI, Chew FS, Blickman JG, Miller SW, Mueller PR. New high resolution teleradioiogy system: prospective study of diagnostic accurary in 685 transmitted clinical cases. Radiology 1993; 186: 429-434. [5] Felix R, Kleinholz L, Ohly M, Lutzeb~ck D, Mahr B, V6ge KH, Hosten N, Zendel W. State of the art in telemedicine: the Berlin experience. Eur Radiol 1993; 3: 295-303. [6] Lo SC, Huang HK. Compression of radiological images with 512, 1024, and 2048 matrices. Radiology 1986; 161: 519-525. [7] lshigaki T, Sakuma S, Ikeda M, Itoh Y, Suzuki M, Iwai S. Clinical evaluation of irreversible image compression: analysis of chest imaging with computed radiography. Radiology 1990; 175: 739-743. [8] MacMahon H, Doi K, Sanada S, Montner SM, Giger ML, Metz CE, Nakamori N, Yin FF, Xu XW, Yonekawa H, Takeuchi H. Data compression: effect on diagnostic accurary in digital chest radiography. Radiology 1991; 178: 175-179. [9] Kajiwara K. JPEG compression for PACS. Comput Methods Programs Biomed 1992; 37: 343-351. [10I Jelaso DV, Southworth G, Purcell LH. Telephone transmission of radiographic images. Radiology 1978; 127: 147-149. [11] Page G, Gregoire A, Galand C, Sylvestre J, Chahlaoui J, Fauteux P, Dussault R, Seguin R, Roberge FA. Teleradiology in northern Quebec. Radiology 1981; 140: 361-366. [12] Franken EA Jr., Driscoll CE, Berbaum KS, Smith Wl, Sato Y, Steinkraus LW, Kao SCS. Teleradiology for a family practice center. J Am Med Assoc 1989; 261: 3014-3015. [13] Slovis TL, Guzzardo-Dobson PR. The clinical usefulness of teleradiology of neonates: expanded services without expanded staff. Pediatr Radiol 1991; 21: 333-335. [14] Templeton AW, Dwyer SJ III, Rosenthal SJ, Eckard DA, Harrison LA, Cook LT. A dial-up digital teleradiology system: technical considerations and clinical experience. AJR 1991; 157: 1331-1336. [15] Lear JL, Manco-Johnson M, Feyerabend A, Anderson G, Robinson D. Ultra-high-speed teleradiology with ISDN technology. Radiology 1989; 171: 862-863.
J. Reponen et al. / European Journal of Radiology 19 (1995) 226-231 [16] Dohrman PJ. Low-cost teleradiology for Australia. Aust NZ J Surg 1991; 61: 115-117. [17] Yamamoto LG, Wiebe RA, Long DC, Dimauro R. Scanned computed tomography image transmission by modem: feasibility for emergency department transfers. Am J Emerg Med 1992; 10: 226-229.
231
[18] Carey LS, O'Connor BD, Bach DB, Hobbs BB, Hutton LC, Lefcoe MS, Lyons RO, Munro TG, Paterson RG, Rankin RN, Rutt BK. Digital teleradiology: Seaforth-London network. Can Assoc Radiol J 1989; 40: 71-74.
ELSEVIER
European Journal of Radiology 19 (1995) 226-231
Low-cost digital teleradiology J. Reponen*, S. L/ihde, O. Tervonen, E. Ilkko, T. Rissanen, I. Suramo Department of Diagnostic Radiology, University Hospital of Oulu, Kajaanintie 50, FIN -- 90220 Oulu, Finland
Received 22 September 1994; revision received 22 November 1994;accepted 13 December 1994
Abstract
A teleradiology link based on standard personal computers and a flat-bed CCD scanner was tested. A 64 kbit/s dial-up digital ISDN telephone line was used for transmission. A total of 254 films (174 uncompressed, 80 compressed) were sent. Ninety-six per cent of the uncompressed images and 98% of the compressed images were considered technically acceptable. The total diagnostic agreement between the acceptable transmitted images and the original films was 98%. Image quality was sufficient for diagnosis in CT and conventional chest and bone radiographs. However, a 256-step gray scale of the scanner was not sufficient for demanding situations, such as overexposed images with a high contrast gradient. The average speed of transmission was 60 kbit/s, which was considered adequate. The tested system suggests that a teleradiology link based on standard personal computers and programs works in situations where instant consultation is needed. However, the image digitization time with the prototype system was quite long, and a better user interface is under development. Keywords: Computers, diagnostic aid; Radiography, technology; Video systems
1. Introduction
Many recent teleradiology experiments have been made with high-capacity equipment, which have been built for frequent transmission of large image sets. Various authors suggest the use of laser scanners with 2048 x 2048 pixels resolution together with 12 bits dynamics, high-end workstations with 2048 x 2048 pixel monitors and fast connection lines up to 1.544 Mbit/s speed [1-5]. The establishment of such teleradiology links may be non-profitable if only a small number of images are transmitted. The capacity of personal computers (PC) has improved, while computers have simultaneously become less expensive. It is also possible to apply inexpensive commercial programs, which can process images under standard graphical interfaces (MS-Windows by Microsoft Corporation, Seattle, WA, USA or Macintosh Finder by Apple Computer, Cupertino, CA, USA). Many of these programs have been primarily developed for print industry, but can display and modify any image in a standard image file format. * Corresponding author.
The purpose of this study was to find out whether teleradiology can be performed using low-cost personal computers together with a flat-bed image scanner and a digital dial-up telephone line. 2. Material and methods
The hardware consisted of a standard PC computer with an Intel (Santa Clara, CA, USA) 33-MHz 30486 processor. The images were scanned using a 300 dots per inch (DPI) flat-bed charged coupled device (CCD) image scanner (model XRS CX3 by X-Ray Scanner Corporation, Torrance, CA, USA) with a 21 x 35 cm scanning area thus giving a maximum of a 2478 × 4130 pixel matrix. In practice the image digitization matrix sizes varied from 512 x 512 to 2048 x 2048 pixels depending on the selected resolution and image size (Table 1). Each pixel had a 256-step (8 bits) gray scale. Large images (e.g. 35 x 35 cm films) were scanned in two parts and connected together. A commercial imageprocessing program was used for scanning and viewing (PhotoStyler by Aldus Corporation, Seattle, USA). The computer screen was 47 cm in diameter. Its resolution was 1024 x 768 pixels, but the image zooming and
0720-048X/95/$09.50 © 1995 ElsevierScience Ireland Ltd. All rights reserved SSD1 0720-048X(94)00602-Z
J. Reponen et al. /European Journal of Radiology 19 (1995) 226-231
227
Table i Scanning density, average sizes and average on-line transmission time for uncompressed images sent from a health centre (Phase 1) and a central hospital (Phase 2) and for compressed images sent from a regional hospital (Phase 3) Image
Scanning density (DPI)
Size (average, bytes)
Transmission time
Uncompressed
Compressed
Uncompressed
Compressed
150 300
3 897 973 2 981 370
NA NA
10 min 50 s 8 min 20 s
NA NA
100 100 150 250 150
840 390 842 210 1 653 345 1 883 547 2 273 610
NA NA NA NA NA
2 min 2 min 4 min 5 min 6 min
NA NA NA NA NA
129 165 150
3 608 680 2 034 236 2 537 205
180 438 184 685 172 733
NA NA NA
Health centre
Thorax Bone Central hospital
Head CT film Body CT film Angiography film Mammography film Ultrasound film Regional hospital Thorax Bone Mammography
20 s 20 s 32 s 14 s 20 s
30 s 31 s 29 s
NA, not applicable.
scrolling functions in the viewing program made it possible to interpret larger digitized images. The image data were transmitted via a dial-up digital 64 kbit/s one-channel ISDN (Integrated Services Digital Network) telephone line with the help of a communication program (Telelmage by Telecom Finland). The procedure of sending the images consisted of the following steps: (1) selecting the images, (2) prescanning and setting up parameters for the first image, (3) scanning, (4) saving the files, (5) making a connection, (6) sending the images. Image receiving was automatic. Both the online sending time and the total image transmission time were recorded. The PC-based teleradiology system was tested in three different system configurations (phases). The software interface and the use of compression differed in these three phases. In the first and second phases, all the programs had to be used separately and the scanning parameters were adjusted manually. No common user interface was used. In the first and second phases, moreover, all the images were sent uncompressed. Before the actual clinical cases, a small pilot series of 10 images was made to determine appropriate settings for image digitization. In the third phase, the effect of data compression and a new easier graphical user interface (programmed with Microsoft Visual Basic) on the total transfer time was tested. All the programs operated in a Microsoft Windows environment. No more than five buttons were available at each screen. Presets for the scanning parameters were used for different images. These presets included a possibility to decrease excessive contrast in chest and bone images. Image resolution for chest and
bone images was now set to the same level as in the commercial image plate system Fuji AC-I (Fuji Corporation, Tokyo, Japan). In this series, images were sent using lossy DCT (discrete cosine transform) image compression with approximately 1:10 compression results (Joint Photographer Experts Group, JPEG compression). (In the literature, DCT compression methods with slight compression between 1:10 and 1:25 depending on the quality of the original have given acceptable results with no statistical difference in diagnostic quality [6-8]. JPEG compression has been proposed for use in PACS environments [9].) The requests and reports were written with the Windows Write program (Microsoft Corporation, Seattle, USA) in a separate program window. A total of 254 films were sent, all to a university hospital. In the first phase 48 uncompressed images were sent from a rural health centre, and in the second phase 126 sets of uncompressed images were sent from a central hospital. In the third phase, 80 compressed images were sent from a regional hospital. The selection of patients was chosen so as to simulate as accurately as possible the real teleconsultation needs when radiological expertise is not available locally (Table 2). In the health centre, the images were scanned and sent by a computer operator, while in the central hospital and the regional hospital this was done by Xray technicians. At the university hospital, the images were interpreted by two staff radiologists on a computer screen. They first scored the images as acceptable or unacceptable in the technical respect. After that, a consensus diagnosis was made from the technically acceptable images. A report was given and returned to the sender via telefax in
228
J. Reponen et al./ European Journal of Radiology 19 (1995) 226-231
Table 2 Image file types, user interfacetype, reporting mode, examination type, selection of patients and total number of films sent in three different phases Image file type
User interface
Request and report
Examination
Selection
Separate interfaces Separate interfaces
Telefax Telefax
Thorax Bone
Clinically problematic cases Clinically problematic cases
38
Separate interfaces Separate interfaces Separate interfaces Separate interfaces Separate interfaces
Telefax Telefax Telefax Telefax Telefax
Head CT Consecutive emergencies Body CT Consecutive emergencies Angiography Consecutiveemergencies Mammography Consecutivescreening cases U l t r a s o u n d Consecutiveemergencies
55 38 25 4 4
Common graphical interface Common graphical interface Common graphical interface
Randomized emergencies Computerfile Thorax Randomized emergencies Computerfile Bone Computerfile Mammography Consecutive screening cases
20 20 40 254
10
Health center (Phase 1)
Uncompressed Uncompressed
Number of films
Central hospital (Phase 2)
Uncompressed Uncompressed Uncompressed Uncompressed Uncompressed Regional hospital (Phase 3)
Compressed Compressed Compressed Total
Phases 1 and 2 and as a computer file in Phase 3. The same radiologists inspected the original images 2 weeks later and again made a consensus diagnosis. The diagnoses were compared for possible differences. 3. Results
3.1. Timefactor The average total image transfer procedure time in Phase 1 (from the health centre) was 15 min 50 s (range 15 min to 18 min 20 s) for an uncompressed single chest X-ray image (average size 3897 bytes). The longest single task in the transmission process was the on-line sending of the image (average 10 min 50 s), which was followed by scanning (64 s, range 55-74 s) and saving the images on a hard disk (57 s, range 53-83 s). The average throughput of the digital line was 60 kbit/s (range 56-61 kbit/s). The average total image transfer procedure time for a CT film (four to six images each) with an average size of 840 kbytes in Phase 2 (from the central hospital) was 6 min 20 s (range 5 min 10 s to 7 min 20 s) per uncompressed film. The longest single step was sending the film (2 min 20 s), the next longest step was scanning (35 s, range 23-42 s), and then came saving the films on a hard disk (26 s, range 19-35 s). The average total image transfer procedure time for a 1:10 compressed chest X-ray film in Phase 3 (from the regional hospital) was 5 min for a single view (36-49% of the time for uncompressed images). Of this, only 30 s on average was on-line time (5-9% of that for uncompressed images). The average time required for the preparatory steps preceding the scanning has decreased from 3 min to 2 min 30 s. The on-line transmission times for the different image files are shown in Table 1.
3.2. Technical quality Of the 48 uncompressed films transmitted in Phase 1 (from the health centre), two chest X-rays and five bone images were considered technically unacceptable. This was due to faulty scanner settings, one chest X-ray had an excessively small scanning area and the other was initially digitized too light. Two thoracic and three cervical spine images were scanned with excessive contrast settings and details lost in both light and dark areas. All the uncompressed transmitted images in Phase 2 (from the central hospital) were considered technically high-quality. Of the 80 compressed films sent in Phase 3 (from the regional hospital), the only low-quality film was an image of the thoracic spine, which was digitized with excessive contrast settings and a consequent loss of details in the highlight parts. 3.3 Diagnostic ability The diagnostic agreement between the transmitted and original images was made in all of the technically acceptable chest (n = 8) and bone (n = 33) images transmitted uncompressed in Phase 1 (from the health center). A lesion was missed in two cases of the 126 image sets (2%) transmitted uncompressed in Phase 2 (from the central hospital). Both cases were brain CT scans (n = 55): the first was a small infarction in the cerebral pons (diameter < 5 mm), while the second was a thickening of the right nervus opticus (difference of 5 mm compared to the other side). All the 38 body CT sets as well as all the 25 angiography film sets, four mammographies and four ultrasound images were correctly interpreted from the transmitted images.
J. Reponen et al. / European Journal of Radiology 19 (1995) 226-231
A discrepant finding was made in four (5%) out of 79 image sets satisfactorily transmitted compressed in Phase 3 (from the regional hospital). All of them were mammography images (n = 40). After interpreting the original images, an enlargement mammography was considered necessary in two cases, while in two cases no further examinations were considered necessary even though an enlargement mammography was suggested after reading the transmitted images. All 20 chest images and all 19 bone images were interpreted correctly from the transmissions. 4. Discussion The earliest generation of teleradiology was based on the transmission of analog TV images, but the image quality was inadequate compared with the quality of film images [10,11]. Modern teleradiology began to develop after the introduction of digitized images. The early digitized systems captured video images with a 512 x 512 pixel matrix and an 8-bit dynamic range, while modern scanners enable resolution up to 4096 x 4096 pixel and a 12-bit gray scale [4,12,13]. Images scanned with 2048 x 2048 × 12 bit matrix have already been proposed for primary diagnosis [4]. The early digitized systems used telephone lines with a speed up to 9600 bits/s. Nowadays speeds from 28.8 kbit/s (phone line) through 64 kbit/s (single line ISDN) up to 1.544 Mbit/s (T-l links or Asyncronous Transfer Mode links) are possible [4,5]. In the cases of high-standard systems with resolution equal to the original films and a full 12-bit (4096 steps) gray scale with instant image transmission, the price for a single-site unit can be as high as 196 000 USD and a central referral unit may cost as much as 343 000 USD. Network services make for additional costs if permanent connections are used [ 1,2,14]. The cost of the equipment in this study was 20 000 USD for the transmitter and 15 000 USD for the receiver unit. This cost level makes teleradiology feasible for health centres and smaller hospitals. The type of dial-up one-channel ISDN lines used in this study is available in all parts of Finland. Their fixed monthly costs are only 20-100 USD and the connections are paid by use, the price being that of normal long distance calls. Only a few reports on the use of low-cost digital teleradiology have been published. Lear et al. [15] built a system based on prototype high-resolution CCD camera, a personal computer and an ISDN link. Their equipment and software were proprietary and not commercially available. Dohrmann et al. [16] used personal computers to establish a low-cost teleradiology link in Australia. He, however, used a video camera and a frame grabber to capture the images from printed films. A slow-speed (2400 bit/s) ordinary telephone line modem was used to transfer images. This solution limits
229
the maximum resolution and the transfer speed and is mostly suited to CT images. Yamamoto et al. [17] built their low-cost teleradiology equipment out of personal computers and hand-held scanners. Their scanning area was limited to 6.1 x 7.0 cm and an ordinary telephone line was used for transmission. The on-line sending time for a 10:1 compressed chest image (size before compression < 4 Mb) in this study was at the same 30-s level as reported by Goldberg et al. [4] for an uncompressed 7 Mb image over a fast 1.544 Mbit/s network. If efficient compression can be used, even ISDN networks can serve effectively. Only a few studies, however, discuss the total procedure time [13,15,18]. In our experience, if efficient compression is used, the digitization together with preparations took more of the total procedure time than the actual sending. The transmitted images were scored as technically acceptable with eight exceptions (3%). In seven cases the inadequacy was due to improper scanner settings or insufficient gray scale dynamics. Because our scanner limited the image gray scale to 8 bits (256 levels), such parameters like brightness and contrast are critical for good image quality. If the original images present excessive contrast, details in both the dark and the light areas may be lost. This means that such images should be sent in duplicate for the dark and light areas. Another solution is an image scanning program which can decrease image contrast by making the highlights darker and the shadows lighter. This method was used in Phase 3 (from the regional hospital). True 12-bit (4096 levels) scanners, which are not so operator dependent, are three to four times more expensive than our unit. Image spatial resolution was sufficient, except in mammograms. The maximum scanning density of the tested system is 300 dots per inch (DPI), which equals 11.8 pixels/mm. This exceeded the scanning density of the Fuji AC-I image plate system, for example. In practice, we finally set the scanning density to the same level as in the image plate system for chest (129 DPI = 5.1 pixels/mm) and bone (165 DPI = 6.5 pixels/mm) images. If we had used the maximum scanning density of the tested scanner all the time, the image sizes would have been too large for our equipment. A 35 x 35 cm chest X-ray with 300 DPI (11.8 pixels/mm) would require >17 Mb and could not be loaded into the 8-Mb memory of the computer. There were no significant differences in the diagnoses of technically acceptable chest or bone images. In two CT images, a small pons infarction and a slight thickening of nervus opticus were missed in the original screen reading. In retrospect, however, these lesions were detected. It seems that the errors were due to the reader's unfamiliarity with the computer screen. The opportunity of film printing might be an advantage. In mammograms, the error frequency was markedly great-
230
J. Reponen et al. / European Journal of Radiology 19 (1995) 226-231
er (10%), which suggests that the scanning density used in this study was not enough for a transmission of the most detailed films. Our results, with their total 98% agreement on the technically acceptable images, correspond well to the resuits by Carey et al. [18], who sent 489 laser-digitized conventional and US examinations with a 98% agreement between the original and transmitted images. It is notable that they still had 12 patients in whom disease was not correctly identified even though the original films were digitized with 12-bit gray scale. Slovis et al. [13] also had an agreement of 98% in their work with 4200 conventional pediatric images. Goldberg et al. [4] had discrepant interpretations in 18 (2.6%) of 685 cases while using a 12-bit scanner. When standard computer programs are adapted to for clinical work, some limitations are encountered. The programs have not been specifically designed for medical work and require many adjustments. In our first two phases, the participants themselves had to make all the preparatory steps and the changes from the viewing program to the transmission program. This was the reason why they let a computer operator send the images from the health center. This caused some quality problems because the computer operator was not accustomed to deciding on the technical quality of scanned images. The easier graphical interface, the "shell" used in Phase 3, resulted in more convenient image transmission. It eliminated the complicated steps and required less teaching. It made it possible for X-ray technicians to transmit the images as part of their normal daily routine without being too much concerned of the technology. This is important, because X-ray technicians cannot spend too much time on teleradiology efforts. The X-ray technicians were also able to keep a constant image quality. This is important for the cost-effectiveness of the system. If a teleradiology system is built on standard personal computers, one has to be aware of the limitations. Depending on the scanner, the gray scale dynamics and scanning density might not be enough for the most detailed images like mammograms. Also the smaller display unit necessitates the use of image zooming. If these problems could be resolved, the same central unit can be used for office automation, Medline literature searches and scientific image processing. Radiologists who are familiar with computers do not need to learn a new user interface, but can utilize the skills they already have. Images are also transferable into teaching programs and archives, for example. Our investigation suggested that a good image quality can be restored in the transmission of images with tested standard personal computers and a commonly available commercial software configuration. This could make it possible to build effective teleradiology links for special purposes with lower costs compared to the commercial
high end systems. According to our experience, the scanner is the most critical unit and its 8-bit gray scale should be extended. In our system, the total sending procedure still took a relatively long time and thus manoeuvrability needs to be further developed. The suitability of this system to mammography films also warrants further studies.
Acknowledgements The authors wish to thank the staff of the Kuusamo health centre, the Rovaniemi central hospital and the Raahe regional hospital for giving their valuable time for this study. The authors recognize the support given by Telecom Finland for this work in the form of computer equipment and communication programs.
References [1] Batnitzky S, Rosenthal SJ, Siegel EL, Wetzel LH, Murphey MD, Cox GG, McMillan JH, Templeton AW, Dwyer SJ 1II. Teleradiology: an assessment. Radiology 1990; 177:11-17. [2I Dwyer SJ III, Templeton AW, Batnitzky S. Teleradiology: costs of hardware and communications. AJR 1991; 156: 1279-1282. [3] Stewart BK, Dwyer SJ III, Kangafloo. Design of a high-speed, high resolution teleradiology network. J Digit Imaging 1992; 5: 144-155. [4] Goldberg MA, Rosenthal DI, Chew FS, Blickman JG, Miller SW, Mueller PR. New high resolution teleradioiogy system: prospective study of diagnostic accurary in 685 transmitted clinical cases. Radiology 1993; 186: 429-434. [5] Felix R, Kleinholz L, Ohly M, Lutzeb~ck D, Mahr B, V6ge KH, Hosten N, Zendel W. State of the art in telemedicine: the Berlin experience. Eur Radiol 1993; 3: 295-303. [6] Lo SC, Huang HK. Compression of radiological images with 512, 1024, and 2048 matrices. Radiology 1986; 161: 519-525. [7] lshigaki T, Sakuma S, Ikeda M, Itoh Y, Suzuki M, Iwai S. Clinical evaluation of irreversible image compression: analysis of chest imaging with computed radiography. Radiology 1990; 175: 739-743. [8] MacMahon H, Doi K, Sanada S, Montner SM, Giger ML, Metz CE, Nakamori N, Yin FF, Xu XW, Yonekawa H, Takeuchi H. Data compression: effect on diagnostic accurary in digital chest radiography. Radiology 1991; 178: 175-179. [9] Kajiwara K. JPEG compression for PACS. Comput Methods Programs Biomed 1992; 37: 343-351. [10I Jelaso DV, Southworth G, Purcell LH. Telephone transmission of radiographic images. Radiology 1978; 127: 147-149. [11] Page G, Gregoire A, Galand C, Sylvestre J, Chahlaoui J, Fauteux P, Dussault R, Seguin R, Roberge FA. Teleradiology in northern Quebec. Radiology 1981; 140: 361-366. [12] Franken EA Jr., Driscoll CE, Berbaum KS, Smith Wl, Sato Y, Steinkraus LW, Kao SCS. Teleradiology for a family practice center. J Am Med Assoc 1989; 261: 3014-3015. [13] Slovis TL, Guzzardo-Dobson PR. The clinical usefulness of teleradiology of neonates: expanded services without expanded staff. Pediatr Radiol 1991; 21: 333-335. [14] Templeton AW, Dwyer SJ III, Rosenthal SJ, Eckard DA, Harrison LA, Cook LT. A dial-up digital teleradiology system: technical considerations and clinical experience. AJR 1991; 157: 1331-1336. [15] Lear JL, Manco-Johnson M, Feyerabend A, Anderson G, Robinson D. Ultra-high-speed teleradiology with ISDN technology. Radiology 1989; 171: 862-863.
J. Reponen et al. / European Journal of Radiology 19 (1995) 226-231 [16] Dohrman PJ. Low-cost teleradiology for Australia. Aust NZ J Surg 1991; 61: 115-117. [17] Yamamoto LG, Wiebe RA, Long DC, Dimauro R. Scanned computed tomography image transmission by modem: feasibility for emergency department transfers. Am J Emerg Med 1992; 10: 226-229.
231
[18] Carey LS, O'Connor BD, Bach DB, Hobbs BB, Hutton LC, Lefcoe MS, Lyons RO, Munro TG, Paterson RG, Rankin RN, Rutt BK. Digital teleradiology: Seaforth-London network. Can Assoc Radiol J 1989; 40: 71-74.