Digital chest radiography: quality assurance.

Abstract

INTRODUCTION Although the requirements and philosophy of quality assurance in a digital environment are very similar to the ones used in film based methods, the advent of digital radiography and especially of PACS has made it necessary to rethink the way radiologists have to control quality of the work they are performing. Quality assurance very often focuses on equipment performance and resulting image quality and is often referred to as quality control. However, quality assurance should not only include the control of the image quality but also the equipment used to produce and maintain this image quality. Quality assurance should also examine the quality of each step in the “imaging process” starting with the request for a radiologic test from the referring physician and not ending before the images and the report are available to this referring clinician. This means that quality assurance also deals with human performance and organizational issues such as training, efficiency of the work, and availability of correct policies and procedures which should be known to everybody involved in the process. 1 This concept of “total” quality assurance is even more mandatory in a digital environment. While, for example, in a film based department, image distribution and availability of previous examinations to the radiologist were mostly considered as beyond the scope of quality control, because they had little dependence on equipment, filmless radiology departments continue to depend on equipment performance also for delivery of images to the referring physician and to the radiologist. 2 In general, quality assurance requests the imaging physician to ensure that: (a) every imaging procedure performed is necessary and appropriate to the problem at hand, (b) equipment performance is optimal, (c) the images generated contain information critical to the solution of the problem, (d) the recorded information is correctly interpreted and made available in a timely fashion to the patient's physician, and (e) the examination results in the lowest possible radiation exposure, cost, and inconvenience to the patient. 3 In most hospitals chest radiography represents an important part of the daily radiologic work. This is not only related to the high value of the chest radiograph in diagnosis and follow-up but also to the fact that the technique is easy and fast and induces only a very small amount of radiation to the patient. In many intensive care units, a chest radiograph belongs to the daily routine. Although bedside radiography can be a difficult task for the radiologic technologist and although interpretation of a chest radiograph can be difficult, workload induced by chest radiographs is more related to the number of examinations performed daily than to the complexity of each individual examination. In addition, especially in hospitals with large intensive care units, timely management of imaging information and fast image accessibility for both the radiologist and the referring clinician are mandatory. 4 Finally, because findings on a chest radiograph are often nonspecific, availability of clinical information and comparison with previous chest radiographs are necessary to deliver a good report. For these reasons quality assurance in chest radiography should not only concentrate on image- and equipment quality but also on operational efficiency and adequate and timely management of imaging information. As will be discussed further on in this text, the digitalization process and especially the development of PACS and hospital and radiologic information systems (HIS/RIS) can offer unique improvements in the total quality assurance of chest radiography. 5,6 This paper will focus on how quality should be reached and maintained at 4 levels of the total process of obtaining and delivering imaging information using digital chest radiography: (1) image acquisition, (2) image presentation to and manipulation by the radiologist, (3) image storage and retrieval, and (4) image distribution and presentation to the clinical user. Image Acquisition Computed radiography (CR), also known as storage phosphor radiography, is currently the most widely used commercial solution for the digital acquisition of conventional radiographs. With this technology, a reusable phosphor plate is placed in an imaging cassette. X-ray exposure produces an image on the phosphor plate which is extracted using a CR plate reader that converts the pattern of radiation on the phosphor plate into a matrix corresponding to the amount of radiation that fell on each picture element. In flat-panel digital radiography, a flat-panel x-ray detector or array of detectors is used to directly capture the radiographic image numerically. In this way the step of reading out the detector (phosphor plate) in another location, as necessary in CR, is eliminated. An alternative way of producing digital chest radiographs is to digitize an analogue film using a “digitizer,” a method predominantly and successfully used during the process of transition from conventional film practice to digital radiology, or to put films from outside the hospital into a digital archive. 7,8 However, because film is non-linear, part of the information is lost and cannot be recovered after digitization. Major advantages of CR over the conventional film-based system are its wide dynamic range, its linearity, and its suitability for computer image processing. Images have a higher and more constant quality, resulting in a reduction of retake rate, which is of major importance in areas such as the emergency room and the intensive care unit. 9 A major limitation, however, is the need for separately reading the phosphor plate after exposure, which is time consuming not only because of the process itself but also because of the necessity to bring the cassette to CR plate readers. The lower spatial resolution of CR compared with conventional film seems less important in clinical practice and is compensated by the contrast resolution, which is greater than in conventional film. In addition, the greater availability of (high resolution) computed tomography and of low dose CT imaging of the lung has increased the use of these techniques to study diffuse and interstitial lung disease, a pathology that with projection radiography is diagnosed less accurately when spatial resolution decreases. Finally, the limited life expectancy of the phosphor plates is an important issue in the assessment of image quality and should also be considered when calculating working costs of the system. Flat-panel digital radiography eliminates the time-consuming step of reading the phosphor plate and has the potential to increase productivity. In our experience this increase in productivity is not only related to the lack of time loss because the detector is read immediately but is also related to the almost immediate availability of the image after exposure allowing the technologist to decide immediately whether image quality is sufficient or whether additional images should be taken, an advantage that probably is more important in bone radiology than in chest imaging. The major disadvantages of this technology are its high costs, the relatively high vulnerability of the system to damage to the detector system, the progressive deterioration, and the lack of portability. Nevertheless, it can be expected that DR, also because of its excellent image quality, will replace CR over the next 5 to 10 years. 10 An important issue in quality assurance during acquisition of digital radiographs is radiation dose. In contrast to the early systems, modern digital, especially flat panel, detectors can reduce dose in comparison with conventional film-screen systems 11–16 both directly and indirectly by eliminating the need for retakes. The latter is related to the reduction of acquisition failures and to the increased amount of information present, allowing production of images that give information on both high and low density areas without a second exposure to radiation. Nevertheless, it is obvious from physical reasons that major dose reduction will reduce image quality. 17 Today most investigators agree that exposure indices should be optimized to produce images that address the clinical information needed and that are not necessarily geared toward producing the best image quality. According to this “ALARA principle” radiation dose should be kept “As Low As Reasonably Achievable.”18–20 While important underexposure usually becomes evident because of deterioration of image quality by increasing quantum mottle, it is more difficult to detect overexposure. 21 Whereas an overexposed conventional film is easily recognized because the film becomes dark, the wider dynamic range of digital detectors together with the image processing software can accommodate overexposure. In this way there is danger for systematic overexposure, a danger that is especially high in bedside chest radiographs and if the examination is performed out of hours by less experienced technologists. 20 Indeed, because of auto ranging, changing (increasing) the x-ray beam current (mAs) has little effect on the overall appearance of the image. Unfortunately, this is the primary method used by technologists to modify the appearance of the conventional film-screen radiographs. Regular monitoring of exposure indices is necessary and manufacturers provide derived indices of exposure based on the results of the normalization process. To have a good radiation management, a report of this index of exposure should always accompany the image. These indices differ among manufacturers but all give an idea about the amount of signal harvested from the imaging plate, in this way providing indirect information about the quantity of radiation exposed to the plate and the patient (Fig. 1). Although the relationship between these exposure indices and radiation exposure to the patient is clear, neither the appropriate target values nor the range of acceptable values is well established.FIGURE 1.: Index of exposure, based on the results of the image normalization process, provides indirect information about the quantity of radiation exposure to the patient.Of course, as with the film-screen system, other indices of image quality should be checked. These include correct patient positioning, correct cessation of respiration, adequate image collimation, absence of jewelry worn by the patient, and checking for foreign matter or debris on the cassette or imaging plate. An artifact specific for CR is caused by mechanical damage to the imaging plate. Scuffs or cracks can mimic signs of pneumothorax. When planning for a filmless department and filmless hospital, it is necessary that together with the other acquisition devices the chest device can communicate directly with the PACS. A perhaps less obvious but equally important aspect is that this device, like the other acquisition devices, must also be connected to the hospital or radiology information system (HIS/RIS). The reason is that the images must be correctly linked to the right patient and to the right examination information. In a film- and paper-based organization, this link is made in an informal way, by putting the films in the right jacket and by classifying that jacket in the right file. While this traditional system is prone to errors at any time (everyone can reinsert a film into the wrong jacket at some time), it does tolerate small errors (a spelling error on the film does not matter as long as the human operator knows which patient is meant and puts that film into the correct jacket). A totally digital system requires a more strict and rigid procedure. Images are filed according to the patient ID and the examination ID (or “accession number”) present in the data sets that are sent to the PACS. The PACS should check whether these IDs correspond to procedures expected to be done in this timeframe, and signal any exceptions for manual correction. When the IDs are entered manually at the modality, typing errors can occur and will require expensive manual correction at a later stage. These typing errors can be eliminated when a “worklist” is sent from the RIS to the modality. In that case the operator only needs to pick an entry from the list. However, these worklists, or any other mechanism for modality-HIS/RIS connection, do not eliminate all errors. Most importantly, it is still possible to select the wrong patient or the wrong examination from the list. The PACS will not be able to automatically detect such errors; it will just assume that some additional images have been generated for the other examination. A consequence of a modality worklist is the false sense of security the operator may get. Especially in large departments, such errors will not be detected until either the radiologist notices that some examination contains unexpected images, or after the clinician complains that a certain examination “has not been performed.” An essential part of quality assurance then becomes to regularly check whether, for example, all examinations that have been marked performed in the RIS indeed are present in the PACS. This requires management tools that go beyond the individual PACS and RIS. Image Display and Manipulation by the Radiologist A major advantage of digital radiography over conventional film-screen is that, once acquired, the digital image data can readily be manipulated. This image processing consists of several sequential events and is intended to improve image appearance and to produce the “optimum image” for a certain anatomic view. Contrast rescaling or contrast enhancement, for example, optimizes contrast and density either to enhance conspicuity of clinical features or simply to mimic the appearance of a conventional film-screen radiograph. Frequency processing, often in the form of edge enhancement using techniques such as unsharp masking or more elaborate ones, changes the appearance of the image to accentuate or suppress details according to their spatial frequency. The radiologist, working for the first time with a digital system, should take some time to explore the different processing techniques and to examine how they change image appearance. Then, the optimum image processing should be chosen. This choice depends very much on the anatomic region that is examined but within 1 anatomic region more than 1 processing algorithm can be chosen. For example in the portable chest, compared with the erect chest, a different frequency processing can be chosen to better visualize lines and tubes. It is important to understand, however, that radiologists interpret images in relation to a “mental reference frame” of what they have learned to be normal or abnormal patterns. In particular in chest radiography such a “mental reference frame” is extremely important to recognize subtle changes in density seen, for example, when a mediastinal mass or interstitial lung disease develops. In our experience, radiologists reading digital chest radiographs for the first time often take the more pronounced visualization of vascular markings in a digital chest for increased interstitial markings or interstitial lung disease. Quality assurance should not only focus on obtaining the best image processing technique but should also concentrate on trying to keep the “mental reference frame” image constant over time, and when more than 1 chest system is available make and keep image quality similar between these different systems. This should enable correct comparison between different chest examinations and will reduce the danger of erroneously perceiving or, in contrast, missing pathology because of image processing. Image processing can indeed have a significant effect on the perceived image quality. 22 This does not, however, mean that once the image is displayed, manipulation should be avoided and that image processing cannot be optimized for an individual case. If necessary, individualized image processing should be performed, but this should then be indicated on the image (Fig. 2). In this way, misinterpretation of a quality change induced by image manipulation can be avoided.FIGURE 2.: Index (2B, arrow) indicating that individualized image manipulation has been performed and that image quality might be changed in comparison with the routinely performed image processing (2A).If no additional image processing was performed and image quality has changed, an extensive objective technical quality control, including measurement of basic imaging characteristics and examination of the associated image processing system, should be performed. Examination of basic imaging characteristics includes control of linearity, repeatability, uniformity of response, modulation transfer function (MTF), noise power spectrum, detective quantum efficiency (DQE), contrast sensitivity, signal to noise ratio, and scatter content, and can be performed using phantoms. 10,23–25 Digital radiographs can be displayed as softcopy images on workstation monitors or as hardcopy produced by film printers. Image processing for softcopy and hardcopy display should be individually optimized to the display method that is used (Fig. 3). Aspects that may influence output quality are size of the image (particularly important with film), spatial resolution, and contrast (particularly important with monitors).FIGURE 3.: Image displayed on a clinical workstation. Despite adequate acquisition, image quality is insufficient because of failure in the optimization process.Several reports have been published on the display size of digital radiographs and suggest that life-sized to two-thirds life-sized images are usually sufficient to detect disease 26–29. Schaefer et al compared digital radiographs laser-printed in a full-size conventional format with radiographs printed at image lengths of two-thirds, one-half, and five-elevenths of the conventional format and found that the detectability of lines and micronodular opacities decreased with declining image format size. 30 Other difficulties experienced by radiologists using images displayed at other than life size are the fact that these images must be viewed from a closer working distance and that measurements become more difficult. The latter is especially pronounced when life-sized and reduced-size images are compared. Digital images can be displayed on film using either wet laser imagers or a dry system. The latter can be a feasible alternative for producing digital chest images for decentralized units such as the intensive care unit. 31 Images can also be displayed on computer monitors. A simple use of these computer monitors is in or close to the x-ray room to give the technologist an impression on image quality. Since this quality check only includes control of patient position and of major failures in acquisition and display, quality is not critical. Workstations on the other hand, are used to perform primary diagnosis and to render a report, and therefore quality of the monitors is very important. A first aspect is spatial resolution, most often expressed as the number of points that can be discerned in either dimension. Radiologic monitors usually have a resolution of 2K or 1K depending on the type of images to be reviewed. A higher number indicates a better quality for this measure, but also a higher cost price. In our experience, spatial resolution tends to be overrated as a measure of quality, and even for chest radiographs 1K is sufficient provided that brightness and contrast are sufficient. A lower resolution may somewhat more often result in the radiologist zooming into a part of the image to check whether the impression of a pattern being present was indeed right. A more important quality measure with computer screens is contrast, that is, the ease with which our visual system can discern slightly different gray values. This in turn depends very much on the amount of light, the brightness of the picture. Conventional monitors emit less light than lightboxes. Much of what makes a radiologic monitor expensive currently is the technology to ensure a high brightness. It should also be emphasized that this brightness diminishes over time, requiring that the monitor should have somewhat more brightness capacity to start with even when that reserve is not used initially. Appreciation of contrast is quite a difficult issue, and is to a large degree related to the “mental reference frame” mentioned earlier. Much more than with a lightbox, ambient light that reflects on the monitor will decrease contrast. For this reason, it is important to control the level of ambient light in the reading room and to calibrate the monitors accordingly. This calibration is important and should be performed regularly to keep display quality at its optimum. Also, to keep image quality constant, sensitometric and densitometric tests, similar to the ones performed on the film processor and lightbox when conventional film-screen systems are used, should be performed on the monitors of the workstation. 32,33 Radiologists who are familiar with PACS may argue that they can effectively read images from lower quality monitors as well. They then do more effort to continuously adapt their frame of reference and to zoom into suspicious regions. Much of the motivation of investing in higher-quality monitors for radiologic diagnosis is reading efficiency, including recognition of the fact that radiologists have to work intensively with monitors during many hours at a time. For this reason, it is important to provide the least fatiguing viewing conditions. 34 Workstations can have 2 to 4 monitors. In our experience, 2 monitors is sufficient for the interpretation and comparison of chest x-rays and even images from CT and MRI can easily be interpreted and compared when they are electronically stacked together and reviewed in sequential or cine fashion. Workstations must be able to retrieve images rapidly and must have an efficient and intuitive navigation system to facilitate retrieval and comparison of older and related exams. It must be possible to personalize the interface between the workstation and the user by providing tools that link to often used functions. 35 In this way speed and quality of work and service will improve. Pavlicek et al compared the use of a RIS and PACS integrated digital chest unit with that of a previous analogue chest unit and dedicated wet processor and found several improvements in quality of service including: (1) immediate release of the patient, (2) decreased rate of repeat radiographs, (3) improved image quality, (4) decreased time for the examination to be available for interpretation, (5) automatic hanging of current and previous images, (6) ad-hoc availability of images, (7) decreased time for clinicians to view results, and (8) increased capacity of examinations per room. 36 However, a workstation should not only be an “electronic lightbox.” Tools for image enhancement and measurements should also be available, and the system should allow the development of tools for computer-aided diagnosis. Image Storage and Retrieval Once the images are acquired, they must be stored and archived for subsequent image review by the radiologist and by the clinician. In a fully digitized hospital, this image database, storage, and distribution system replaces not only the film file room but also the people that are responsible for filing the films and the people or system responsible for distributing the images and for tracking the status of the radiologic records within the department and within the hospital. In this section we focus on storage and retrieval, while in the next section we go into distribution to clinical users. The image storage system often has 2 parts: the short-term storage and the long-term storage. The key point of short-term storage is that access to the stored images is rapid, at the expense of this subsystem being quite expensive. Exams that belong in this short-term storage are those for which an approved report has not yet been rendered, older exams necessary for rendering a report on a new examination, and exams necessary during scheduled clinical appointments. This short-term storage system should permit retention of images for a period of a few weeks and optimally should have the capacity to store 2 to 3 months of images. The long-term archive is designed to store the imaging studies for a longer period using less expensive and slower technology. It is necessary to ensure that retrieval times from the long-term storage system are less than a minute in order not to interrupt the reporting process. Therefore, most often “prefetching” is used, which is automatically transferring from long- to short-term storage those images that are likely to be needed. Prefetching should be triggered automatically by the HIS/RIS, for example to retrieve relevant older studies when a request for a new examination is made, or to retrieve all previous images when a patient is admitted to the hospital or has a scheduled appointment. A well-designed and good functioning prefetched system avoids that the radiologist or clinician has to slow down or stop when viewing images and in this way increases workflow. 5 With technological advances, combined with the desire to obtain a simple storage system that can be shared with other applications, more and more imaging systems abandon the use of hierarchical storage systems and hold all images available in the fast archive level. The desire to hold images available in fast storage for an extended period of time motivates the use of image compression. Compression techniques, that enable the original image to be reconstructed exactly, only attain a compression factor of about 2. The interesting compression schemes in this context are those in which substantially higher compression factors can be obtained, at the expense of slight modification of the pixel values. Slone et al found that at x2 magnification, images compressed with either JPEG or WTCQ algorithms were indistinguishable from unaltered original images for most observers at compression ratios between 8:1 and 16:1, indicating that 10:1 compression is acceptable for primary image interpretation. 37 Others concluded that compression ratios as high as 25:1 or even 30:1 can be acceptable for primary diagnosis in chest radiology. 38,39 An aspect that must not be neglected with storage systems is backup to assure that no data are lost when any of the data-carrying media fails. For this reason a copy of the data should be available on another medium. But conventional backup strategies, in which the complete database is copied, do not work with the amount of data in a PACS, so in practice each new image is automatically written to another medium. It is also important that operations can resume in a reasonable period of time when the archive gets completely destroyed. A prerequisite is that the backup media are stored in a completely different location. Unless the recovery procedure is well thought over, it may take a long time before the system is operational again. Finally, it is necessary that the most important operations can continue when the storage system temporarily fails. Image Distribution and Presentation to the Clinical Users An important advantage of PACS is the possibility of a rapid distribution of images not only within the radiology department but also within the entire hospital. This allows the radiologist to view the images and make the report, and the clinician to view the images and consult the report very shortly after these images have been acquired, an advantage that is especially valuable at the intensive care and emergency departments. 40 Some discussion can arise about whether the images should become available to the referring clinician before or after the radiologist has reported them. When images are made available immediately after the examination and before the radiologist reports them, extra attention should be given to the integrity, correctness, and diagnostic quality of these transmitted images (Fig. 4).FIGURE 4.: Obvious acquisition failures. Early quality control should avoid automatic transmission of these images to the PACS.Besides the aspect of timely and efficient availability, there also is the aspect of presentation quality for clinical viewing. As was discussed in a previous section, providing high display quality on computer monitors is expensive. It can be argued that more clinical value can be achieved by providing viewing on many workstations, albeit with plain (low cost) monitors, than on a low number of workstations with dedicated (high cost) monitors that offer the highest display quality. In addition, it is usually unrealistic to control lightning conditions in the working space of the clinical users, because of conflicting requirements that do not exist within the radiology department. Taking into account that a radiology report is available in which the radiologist can point to the subtle details, and taking into account that the number of cases in which clinicians need to spend somewhat more time and effort to study such details themselves is limited, a common practice is to provide plain monitors for clinical viewing. Likewise, the clinical viewing software often is “lighter,” as the emphasis is on ease of use, also by occasional users, and on ease of maintenance more than on availability of sophisticated tools for experienced users. However, differentiation in display quality and viewing software may be necessary to fulfill the specific needs of the referring clinician. As a clinician has to integrate results of different diagnostic tests (of which diagnostic imaging is one) and as it can be assumed that these results should be available in the clinical workstation (or electronic patient record), a very important aspect of clinical viewing software is its ability to be integrated within an overall clinical workstation. 41,42 Availability of all kinds of diagnostic information to all actors that cooperate in the treatment process of the patient indeed is very important. This process of information sharing will benefit the patient not only because it improves the diagnostic process in which imaging often plays an important role, but also because it allows the radiologist to improve the quality of his work since, at the moment when the examination is planned or when the report is made, the entire clinical information is now available. That is why the radiologist is not only responsible for the intrinsic quality of the images but should also help to optimize the information sharing process, not only by making the images and reports timely available, but also by using modern technology to more clearly point to or highlight relevant details in the imaging dataset. 43 To an increasing extent, quality of the radiologist's work will indeed not only be judged by the intrinsic quality of the images and the radiologic interpretation, but also by his ability to efficiently distribute the right information into the complete process of medical decision making. CONCLUSIONS Quality assurance in chest radiography should not only focus on equipment performance and resulting image quality, but should also include quality check of each step in the imaging process from the request to perform a chest radiograph to the moment the images and the report are available to the referring physician. This concept of “total” quality assurance is even more mandatory in a digital environment where there is a continuous and complex interaction between the imaging system and the radiology and hospital information systems.