| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Perspective |
1 Department of Radiology, School of Medicine, University of Washington, 1959 NE
Pacific St., Seattle, WA 98195-7987.
2 Department of Radiology, Harborview Medical Center, Seattle, WA.
Received September 18, 2006; accepted after revision February 5, 2007.
Address correspondence to B. K. Stewart
(bstewart{at}u.washington.edu).
OBJECTIVE. A data-mining program extracts computed radiography (CR) sensitivity-number (S-number) information from the PACS at our institution on a monthly basis as an ongoing quality assurance (QA) improvement project. These data are compared with the previous month's data and departmental S-number goals. The results are presented at monthly QA meetings. The S-number trends are then used by technologists to modify radiographic technique charts to reach the departmental S-number target range goals.
CONCLUSION. This cyclic QA improvement process shows that mining PACS data can be useful in reducing patient radiation dose and interexamination dose variance.
Keywords: ALARA concept • computed radiography • data-mining program • dosimetry • PACS • radiation dose • radiologic physics • S number
Since the commercial debut of computed radiography (CR) in 1983 [1–3], three problems that are associated with its use with respect to the administered patient dose, especially if CR is performed without phototiming (i.e., using manual techniques), have become well known.
The Need for CR Dose Monitoring
The first problem with CR is dose creep [4]. We have found that when the CR dose is not being monitored on a global or a detailed level, the dose administered to the CR imaging plate (and thus to the patient) creeps up over time. The second problem is that radiologists do not typically complain about image quality when the administered dose in a radiographic procedure is too high because this makes "pretty" images with low quantum mottle. Some radiologists are not cognizant of the relative dose administered and, thus, do not express a desire to lower the radiation dose à la the ALARA (as low as reasonably achievable) concept. The third problem is that the limited latitude of film-screen radiography requires technologists to adhere to strict limits on the radiographic technique applied to an examination. The wider latitude of the CR photostimulable phosphor imaging plate [5, 6] allows these limits to be relaxed, so marginal techniques can be used to produce a salvageable image and the best-technique proficiency may deteriorate.
The ongoing cyclic quality assurance (QA) project at our institution is an attempt to raise the consciousness not only of technologists, primarily, but also of radiologists and administrators about the importance of correct and adequate technique in reducing patient dose.
At Harborview Medical Center (HMC), which is part of the University of
Washington Medicine system, CR has been used for the past 14 years. Only CR
units manufactured by Fuji Medical Systems have been and are currently being
used (nine 5000, one 9501, and two SmartCR units). Fuji uses a parameter
termed the "S number" (for sensitivity number) that describes in
an approximate fashion the average absorbed dose to the imaging plate. The S
number (S) is defined as varying with exposure as follows: S
200 / exposure (mR) [7,
8]. The S number provides a
general indicator or estimate of proper CR exposure technique and radiation
dose [9] with some caveats,
which we address in the Discussion section.
Materials and Methods
Figure 1 shows the flow of information for the CR dose surveillance QA process from the acquisition of a CR image through monthly QA meetings to provide feedback to technologists about the necessity of modifying technique factors to meet the mutually agreed on S-number target goals. Multiple HMC CR systems acquire radiographic examinations that first undergo QA by technologists on a Fuji IIP QA (5000 and SmartCR units) workstation before being transferred to PACS (Centricity 2.0, GE Healthcare).
|
The data-mining program is automatically run in the early morning on the first day of every month for the data about the previous month's CR examinations. The output file is accessible from a Website running on the archive in a comma-separated value (CSV) format that is easily imported into a spreadsheet program for analysis and correlation.
The current data-mining program collects the S-number data for only the first image of each examination. Because some examinations with the examination codes that were surveilled contain more than five images, this clearly posed a gap in our S-number analysis. This gap was overcome through the use of the exposure result log function of the IIP QA workstations. Complete data about each imaging plate processed are stored on the IIP QA workstation's hard drive and are recorded monthly to a floppy diskette in a tab-delimited format text file. Although all the Fuji CR units provide DICOM work-list capabilities through the PACS (HL7/DICOM Broker, Mitra), the examination code is not one of the fields stored in the exposure result log. The data analysis, as we describe later in this article, is driven by the examination code. If these data were available, then the data-mining program on PACS would not be necessary.
Thus, there is a need to merge the results from the PACS data-mining program and the collection of exposure result log diskettes. We accomplished this task by writing a program using Visual Basic for Applications (VBA, Microsoft) that takes the CSV file output from the PACS data-mining program (> 12,000 rows: one for every unique CR examination / accession number that month) and the appropriate data range extracts from the exposure result log tab-delimited text files (> 25,000 rows: one for each imaging plate used). Data from both of these file types are imported into Excel (Microsoft) and stored as Excel workbooks. The VBA program is run through the Visual Basic Editor (Microsoft) in Excel and uses the accession number to correlate the two data sources. The merged output contains S numbers for all imaging plates for all examinations performed during the month in question.
Results
Because the S-number target goals are set anatomically, we decided to track the S numbers over time as a function of specific examination codes associated with anatomic delineations. Also, the S-number data can potentially vary over four orders of magnitude; thus, it is desirable to try to equalize the variance over this large range. A simple means of accomplishing this is through a logarithmic transform wherein simple statistics, such as log S-number average and SD, can be calculated for the number of examinations with specific examination codes. To remove outliers that would skew the analysis, all data outside 2 SDs of the mean are removed from the analysis. The new average value and those at ± the new SD are then anti–log transformed back into the S-number space. These values are then used to create bar charts for presentation (Fig. 2).
|
The previous month's data are compared with the one before and the baseline data, which are the data obtained when this mining process was initiated for that section. In addition, at the bottom of each bar in the chart, the percentage of excess dose is indicated to provide the technologists the difference between the imaging plate exposure associated with that examination code's current average S-number and the current target average S-number. The excess dose data are presented in an attempt to quantify, in a tangible way, the excess amount of radiation being administered, given that the percentage of increased dose to the imaging plate should be equal to the percentage of increased dose to the patient. The last feature of the bar charts is the number of red-flagged examinations, those with a CR plate S number in excess of 800 (quantum mottle interfering with diagnosis) and less than 50 (possibly large excess dose to the patient). The red-flagged examination information from the data-mining program is sent to the technologists' supervisors, so they can bring the errors to the attention of the appropriate technologists.
The S-number dashboard is copied and posted at key locations in each section (e.g., near X-ray equipment consoles and the CR QA workstations). The S-number dashboard trends are then used by the technologists to modify manual radiographic technique charts to reach the S-number target range. Although many examinations are phototimed in the department, many others are still performed using manual techniques. For example, in the HMC Trauma Center, some orthopedics and all portable radiography images are acquired using manually selected techniques. Each technologist has been trained differently and makes unique changes that have evolved into an "art" based on his or her experiences to compensate for patient body habitus and for backboards and metallic objects.
The mean S numbers for the main radiology department, orthopedic clinics, trauma center, and portable radiography examinations have increased by 61.7%, 65.8%, 47.4%, and 83.6%, respectively, since baseline, corresponding to an approximate reduction in dose to the CR imaging plate of between 32.2% and 45.5%.
The change in S number for each section is calculated by taking a weighted
average of the average S number (Si) for each specific
examination code under review for a section (i index) at baseline
(time [t] = 0) and at the current end point (t):
 |
200 /
. As a result of this process, we asked the radiographic equipment vendor to adjust our phototiming systems to reflect the S-number goals for specific anatomy. For example, as shown in Figure 2, there are two different S-number targets in orthopedics for different anatomic classes: For bone, spine, and extremity examinations, the target S number is 200 ± 15% and for general imaging, including the abdomen, is 300 ± 15%. We have not found their methodology or end result for these modifications to be entirely satisfactory, requiring us to adjust the anatomical program settings to between –1 and –3. However, our institution recently bought an electronic cassette (Radchex, Fluke Biomedical) that is instrumental in surveying and adjusting the S-number calibration of all the CR readers in the department and should, in the future, allow us to work with the radiographic equipment vendor to properly adjust the phototiming system to reflect the S-number goals for specific anatomy.
Discussion
In the section titled "The Need for CR Monitoring" at the beginning of this article, it was stated that the S number provides a general indicator of proper CR exposure technique and radiation dose [9] with some caveats. This statement raises several questions, the first of which is whether the S number is a good estimate of the absorbed dose to the CR storage phosphor imaging plate. The answer, of course, is yes and no. First, the S number cannot tell us how much of the radiation dose the patient absorbed because of varying body habitus and the possibility that radiation escaping from the patient is being partially absorbed by the imaging plate. Andoh and colleagues in Japan [10] found that the S number is a practical approximate indicator of image quality and is inversely correlated with skin dose, as might be expected from the numeric definition of the S number, but is only moderately so (r2 = –0.875).
Seibert and colleagues at the University of California Davis [4] found that a properly calibrated and functioning CR reader has the capability to estimate the incident exposure to the imaging plate within 20%. Also, the S number can be used as an indicator of the relative quantum noise, but not as an absolute guide, regarding image quality and radiation dose.
Christodoulou and colleagues at the University of Michigan [11] found that the S number may not necessarily represent the exposure to the imaging plate; factors other than the number of X-ray quanta that form the useful image such as technique, positioning, collimation, and histogram analysis may affect both the resulting S number and the image quality.
For radiographic (plastic and metal) or anthropomorphic phantoms for which positioning and collimation remain relatively constant, we have found that the output S number is highly correlated with the imaging plate absorbed dose. However, in the real world of clinical imaging, many factors can cause the output S number to deviate significantly from the dose absorbed by the imaging plate. To demonstrate this possibility, we must describe how the Fuji CR units derive the S number from an exposed imaging plate.
In the automatic (i.e., clinical) mode, a Fuji CR system generates an imaging plate absorbed dose histogram by subsampling on a 1.8 x 1.8 mm grid (pixel size for 14 x 17 inches is 0.2 x 0.2 mm) over the entire imaging plate or on a subregion based on collimation edge detection (CR Users Guide, Fuji Medical Systems). The measured histogram is "pattern matched" to empirically derived model histograms stored in the Fuji system specific to the anatomic region selected before the imaging plate is inserted into the CR reader. How well the measured histogram correlates with the model histogram depends on many factors such as technique, positioning, collimation, and the presence of unexpected foreign objects. The end result of this histogram matching is the calculation of specific absorbed dose breakpoints for the histogram—for instance, the demarcation of soft tissue and unattenuated background radiation in a chest image.
Typically two absorbed dose breakpoints (S1 and S2) result from the histogram analysis. These two breakpoints are empirically assigned by the CR system to digital values (10-bit range, 0–1,023)—for example, Q1(S1)= 818 and Q2(S2) = 205, where Q1 and Q2 are the digital values at the respective dose breakpoints. These two points set the slope and intercept of the sensitivity–latitude line that defines the absorbed dose to digital value conversion function. Sk is defined as the absorbed dose that provides a digital value of 511 (digital value midpoint) along this line. Given that the Sk value is proportional to the logarithm of the absorbed dose, the S number is calculated as follows: S number = 4 x 10(4 – Sk) [12].
Thus, the estimate of the S number is only as good as the histogram analysis that calculates its value, which, in turn, is affected by many factors including technique, positioning, collimation, and the presence of unexpected foreign objects. Changes in peak kilo-voltage (kVp) and tube current (mAs) settings do not alter the general shape of the histogram greatly but cause it to expand or contract, which the histogram analysis process can handle. However, changes in collimation and positioning can alter the measured histogram appreciably from the model histogram. To minimize these effects, Fuji has developed pattern recognizers of irradiated exposure field, which are called "PRIEFs," that analyze the image to determine whether more than one image is on the plate (e.g., two separate images of the feet on a single imaging plate) and where the collimation boundaries are located [12]. Image regions outside the detected collimation boundaries are not used in the histogram analysis process.
Fuji also uses neural network technology to cope with changes in positioning encountered in day-to-day radiography to determine the "best" S1 and S2 values for a particular measured histogram and selected anatomic region. However, unexpected foreign objects—for example, backboards and metallic objects, which are often used to image trauma patients—can cause miscalculation of the histogram breakpoints and thus an S number that does not reflect the geometric mean of the imaging plate absorbed dose as it is designed to do.
A remaining question is the following: What are appropriate S-number targets and ranges? Seibert et al. [4] have suggested a range of 125–275 (200 ± 37.5%) for portable chest radiography. Andoh et al. [10] have provided a range of 150–250 (200 ± 25%) for bone imaging and 260–400 (330 ± 21.2%) for renal imaging. Christodoulou et al. [11] have stated ranges for general imaging at 50–90 kVp of 400 ± 15% (i.e., 340–460) and chest imaging at 125 kVp of 200 ± 15% (i.e., 170–230). At HMC, the current S-number goals are (S number range = S number ± 15%) as follows: bone and extremities, 200 (170–230); portable and fixed chest, 300 (265–345); and general imaging including the abdomen, 300 (265–345).
This project has promoted a culture change in the way technologists manually perform CR. The development of new technique charts and methods for modifying techniques for the considerations we have discussed not only has shifted the resultant S numbers closer to the desired target values, but also has reduced the numeric and visual variances of the images, providing a greater uniformity in presentation to the radiologists on PACS. This cyclic QA improvement process shows that PACS data can be mined usefully to assist in reducing patient radiation dose and interexamination dose variance.
Acknowledgments
We thank Eric Feingold of GE Healthcare who wrote the PACS data-mining program that extracts the S-number data on a monthly basis. Also thanks to the technologists and administration of Harborview Medical Center (HMC) and of HMC's Department of Radiology for their encouragement and support and to the members of the HMC Main Radiology and Orthopedic and Trauma/OR Comprehensive Specialties Imaging Committees.
References
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
