Receiver Operating Characteristic Curve Analysis for Image Quality in the Detection of Solitary Pulmonary Nodules on High Resolution Paper Prints Versus Dry Laser Film.
Hu Xiao-yun,Fang Xiang-ming,Cao Yang,Hu Chun-hong,Yao Xuan-jun,Chen Hong-wei,Hu Su,Shao Ling,Hu Gang-feng
DOI: https://doi.org/10.3760/cma.j.issn.0366-6999.20130258
IF: 6.133
2013-01-01
Chinese Medical Journal
Abstract:A long with the development of digital X-ray imaging technology, a new, economical and practical paper-based output system gradually emerged. To reduce cost, more and more radiologic institutions began to change from film copies to lower-priced paper prints for documenting radiologic findings.1 Therefore, laser paper printers are frequently used as an alternative to duplicating film though the use of printed images is not yet widespread. The aim of this study was to compare the practical value of high-resolution paper printer with dry film copies for detecting small solitary pulmonary nodules (SPNs). METHODS Clinical data Sixty cases (28 cases of SPN, others were negative) of patients for thoracic plain CT scans were deliberately selected by 2 senior doctors in consensus via workstation or picture archiving and communication system (PACS) regarded as a gold standard, and all nodules were confirmed by surgery or clinical follow-up. The diameters of nodules were in the range of 3-20 mm. Images printing All cases were printed by both a dry laser printer (AGFA Drystar 5503, AGFA Gevaert N.V., Belgian) and a high resolution paper printer (BESWAY 8000, BESWAY, China), respectively. For each case, it was adjusted to best display mode with the same window level and width before printed by the same radiographer. All images for each patient were printed as usual and the format for each sheet was usually at 4×6, 5×6, 5×7, 5×8, 5×9, or 6×10, etc. Image interpretation Two diagnostic radiologists (Doctors A and B) interpreted film copies and paper prints of CT images independently of each other in a quiet atmosphere. The time interval was two weeks, and each time the observer interpreted only one type of the images. Paper prints were carefully examined under good light. The film copies were observed on a tested light box. For the receiver operating characteristic (ROC) methodology, each observer had to be familiar to define the probability of presence of pulmonary nodule for each of the test objects according to a 5-point rating scale: grade 1, certainly a nodule present; grade 2, probably a nodule present; grade 3, indeterminate; grade 4, probably no presence of a nodule; and grade 5, definitely no presence of a nodule. Statistical analysis Multi-observer ROC methodology was used to analyze the rating data. The SPSS 10.0 software (SPSS Inc., USA) was used to perform the calculations. Diagnostic accuracy was measured using the area under the binomial ROC curve. The significance level was set at P <0.05. ROC curves of all observers combined were obtained by computing mean values of the individual observers. RESULTS For some reconstructed images of SPNs, the paper prints were able to straightly demonstrate the lesions via the common sunlight rather than light box (Figure 1A). Because there was only grayscale display function for laser films, three-dimensional colored images of SPNs were nearly invisible on film copies (Figure 1B), and the anatomic structures were not clearly enough even with a light box (Figure 1C, 1D). On the contrary, they could be properly and directly showed on paper prints without the traditional light boxes (Figure 1A, 1E).Figure 1. A:: A case of benign solitary pulmonary nodule (SPN). It was confirmed as an arteriovenous malformation (AVM) by the contrasted CT scanning, particularly by the reconstructed colored images. The lesion was clearly and directly showed through the three dimensional reconstructed image on the paper print. B: The same image was nearly invisible on the film copy without the light box. C: The anatomic structures were very poor on the film copy even with the light box in comparison with paper print. D: A SPN located in the upper lobe of right lung was seemingly demonstrated on the soft copy via the monitor (arrows), but it was not clearly visible on either film or paper copies. E: The same SPN was quite directly displayed after three dimensional reconstruction with color-encoded method (arrow). It was truthfully showed only by paper prints. F: Graph showed area under receiver operating characteristic curve for paper prints (broken line) and film copies (solid line) for detection of SPNs. Statistically, there was no significant difference in observer's performance between paper prints (Az=0.940) and film copies (Az=0.973) (z=0.85, P=0.39).The result of image interpretation for detecting SPN with two types of hardcopies was independently recorded. The two observers had the almost same judgments for detecting the SPNs through either paper prints or film copies (Table 1). According to the results of two observers for interpretation of two hardcopies, the result of ROC analysis was drawn by the software of SPSS (Figure 1F). As to the area under the ROC curve in detection of SPNs, there was no significant difference between the two observers for paper prints (z=0.26, P=0.79) and film copies (z=1.02, P=0.31). The mean area under the ROC curve was 0.973 (standard error (SE), 0.019) for film copies and 0.940 (SE, 0.033) for paper prints and the difference between them was not statistically significant (z=0.85, P=0.39).Table 1: ROC analysis by two observers for interpretation of film copies and paper prints (n=60)DISCUSSION With the improvement of technology, medical imaging data have been converted to digital information which could be recognized, transmitted and communicated by personal computers. Once the images are converted into digital forms, they can be printed on paper either locally or at remote sites, thereby facilitating the transmission of medical information to referring doctors.1,2 To date, dry laser films with high image quality, to our knowledge, have overwhelmingly been the mainstream hardcopy. But the operating and handling cost of a film print system are comparatively high, and the film is extremely sensitive to environmental temperature and sunlight.3 Skyrocketing health care cost and pressure from managed care have combined to promote cost-cutting strategies in radiology department,4 which makes the paper print output system become a new tendency for hardcopies. As our results indicated, through the integration of high resolution laser printer and DICOM-based print servers, specialized laser paper, even plain paper, can deliver high quality images which are comparable to the original images and are acceptable for most medical uses. Generally, paper print system enjoys several advantages as follows: (1) Low-cost for each sheet of paper compared to a sheet of film. In our medical market, normal 35.56 cm × 43.18 cm (14 inches × 17 inches) specialized laser paper prints cost approximately $ 1 each, but for a film copy $3 at least. (2) Low operating and maintaining fees. (3) Simple installation and handling. (4) Highly effective, direct viewing. The new style of high resolution of paper prints deliver grey and color prints, which markedly meet the requirements of modern imaging. (5) Recyclable, readily biodegradable, thus eliminating the environmental, time, petroleum consumption, storage, and delivery problems associated with film. This is a great way to not only meet the “energy-saving, emission reduction” of the modern social requirements but markedly contribute to cost-cutting strategies in medical health care. Our initial experience shows that the laser paper prints were, in most cases, acceptable for the diagnosis of images of both CT and MRI, particularly for the reconstructed images, like three-dimensional multiplanar reconstruction and volume rendering. An estimate was made of the minimum size at which a CT scan could be printed without loss of detail. On the basis of the estimate of modulation transfer function made earlier and the assumption that the CT scan will probably be 512 × 512 pixels in size, 512/120 = 4.3 cm2.5 As a result, it is believed that most CT scans can be printed as “thumbnail” images less than 5 cm (2 inches) square without significant loss of detail, let alone the MR images (usually less than 512 × 512 pixels in size). More importantly, CT and MR have overwhelmingly become the main consumers for the use of hardcopies in daily work. Taken our department for example, the average 1300 sheets of hardcopies are needed for both CT and MR images. When the paper printer is normally used, most of the prints are comparable in quality to the original images and acceptable not only for radiologists' report-typing but for physicians' direct-viewing. In summary, for SPNs, the image quality obtained from a paper printer is comparable and similar to that from a dry laser printer. Consequently, we may safely draw the conclusion that the paper printer can be sufficiently used as a potential substitute. It is expected that paper prints will be honored as a “green helper”, which help to cut the cost of medical care and to facilitate the use and distribution of patients' imaging information.