Dose reduction in digital chest radiography and ... - BIR Publications

0 downloads 0 Views 156KB Size Report
ABSTRACT. The aim of the study was to investigate whether radiologists can rank the image quality of digital radiographs with different doses; a preliminary ...
The British Journal of Radiology, 80 (2007), 984–988

Dose reduction in digital chest radiography and perceived image quality 1

L J M KROFT, MD, 1 J GELEIJNS, PhD

PhD,

1

W J H VELDKAMP,

PhD,

2

B J A MERTENS,

PhD,

1

J-P A VAN DELFT,

BSc

and

Departments of 1Radiology and 2Medical Statistics, Leiden University Medical Centre, Albinusdreef 2, 2333 ZA Leiden, The Netherlands

ABSTRACT. The aim of the study was to investigate whether radiologists can rank the image quality of digital radiographs with different doses; a preliminary study investigated whether reduced dose images provide sufficient diagnostic quality. Raw data of 40 chest radiographs (posteroanterior (PA) and lateral) obtained with a fullfield slot-scan charge-coupled device system in 20 patients with chest pathology were used. Noise was added to simulate reduced dose levels to 50%, 25% and 12%. Four observers ranked the quality of the corresponding images and judged the diagnostic quality. Linear regression analysis was performed. Differences were found in image quality at the different dose levels for both PA (p#0.0001) and lateral images (p,0.002). The quality was graded sufficient in 159/160 100% dose-level observations and in 159/160 50% dose-level observations. At lower doses, substantially fewer images were graded as having sufficient quality: for PA images, 69/80 for the 25% dose and 63/ 80 for the 12% dose; for lateral images, 74/80 for the 25% dose and 63/80 for the 12% dose images. In conclusion, observers recognized the reduced quality of lower dose radiographs, although 50% dose imaging was regarded as equal to the 100% dose in having sufficient quality to answer clinical questions. Preliminary findings suggest that a 50% dose reduction seems feasible in a variety of chest pathologies, whereas further dose reduction reduces the diagnostic quality.

Since the introduction of digital radiography, several patient-based studies have shown equal-to-superior perceived imaging performance of digital radiography compared with film-screen radiography, even with substantial dose reductions of up to 59%. These observer preference studies used one dose level for each system and the visibility of various anatomical structures was then compared between the systems [1–4]. Lesion detection studies have shown similar results [5–8]. This has resulted in the expectation that the radiation dose for digital chest radiography could be substantially reduced [1–7, 9]. It is unclear what effect various degrees of dose reduction within the same digital imaging system would have on observers’ opinions with regard to depicting chest pathology and answering clinical questions relating to the investigations performed. Such knowledge would be helpful in deciding to what level doses could be reduced for digital chest imaging. Here, an observer preference study was performed to investigate whether radiologists can rank image quality of radiographs that represent acquisitions at substantially reduced doses, with a preliminary study investigating whether these images provide sufficient quality for answering common clinical questions.

Address correspondence to: L J M Kroft, Department of Radiology, C2-S, Leiden University Medical Centre, Albinusdreef 2, 2333 ZA Leiden, The Netherlands. E-mail: [email protected]

984

Received 29 September 2006 Revised 7 February 2007 Accepted 22 February 2007 DOI: 10.1259/bjr/80232832 ’ 2007 The British Institute of Radiology

Methods Patients and imaging Raw data of 40 digital chest radiographs (posteroanterior (PA) and lateral projection), obtained in 20 consecutive patients who had their exam reported as abnormal (i.e. with intra-thoracic abnormalities), were used for the study. Imaging had been performed for clinical reasons. The clinical history provided by the referring clinician, as well as the clinical question needing to be answered for each investigation, is listed in Table 1. Institutional review board approval was not required for this retrospective study. In these patients, PA and lateral chest images had been obtained with a digital chest system (charge-coupled device system with slot-scan technology: ThoraScan; Delft Imaging Systems, Veenendaal, the Netherlands). With this system, the full imaging process takes about 3.5 s, which includes a pre-scan and the actual acquisition (which takes approximately 1.3 s). The image is built up line-by-line with a local exposure time of approximately 20 ms to avoid movement artefacts. The slot-scan technique results in low scattered radiation and a grid is not needed. Further technical details relating to this system have been described [10, 11]. Standard acquisition settings were used (automated exposure control, tube voltage 133 kV, focus to detector distance 183 cm, 3 mm Al equivalent (inherent plus additional aluminium filtration) and 0.3 mm Cu The British Journal of Radiology, December 2007

Dose reduction in digital chest radiography Table 1. Clinical history and clinical question for each patient No.

Clinical history provided

Clinical question to answer

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

COPD, mammaca metastases, tachypnea. Small cell lung ca after chemotherapy. COPD, mitral valve surgery. Fever. Pleural empyema in the past. Pleuritis carcinomatosa. Small cell lung ca after 4th chemotherapy. History of decompensatio cordis, now cardiac arythmia. Lobectomy right-upper lobe. Empyema. Re-CABG preoperatively. Fever. Back pain. Pain and shortness of breath. Prostate ca with pleuritis carcinomatosa. Persistent coughing. Rhonchi. SOB on exercise. Wheezing. Lung ca after radiation therapy. Coughing. Rheumatoid arthritis. Fever. Coughing for months. Pre-operative. SOB with temperature rise. SOB on exercise. COPD. Adrenal ca. Pulmonary metastases.

Pneumonia? Follow-up pleura effusion? Follow up; tumour progression? Pneumonia? Pleural empyema? Follow up; pleural effusion? Follow-up tumour? Decompensatio cordis? Pleural effusion? Abnormalities? Pneumonia? Pneumothorax? Follow-up pleura effusion? Other pathology? Abnormalities? COPD signs? Radiation pneumonitis? Pneumonia? Other abnormalities? Abnormalities? Pneumonia? Pulmonary abnormalities? Size of metastases before chemotherapy?

COPD, chronic obstructive pulmonary disease; ca, carcinoma; CABG, coronary artery bypass graft; SOB, short of breath.

additional filtration). The image pixel size was 162 mm with an image matrix of 273662736 pixels. The effective dose calculated for imaging patients with this system is intermediate compared with other digital chest systems and in the range of that of speed class 400 screen-film techniques [11, 12].

Dose reduction simulation Reduced dose images were simulated digitally. This was achieved by adding noise to the raw data radiographs on a pixel-by-pixel basis [13]. Images with reduced doses representing 50%, 25% and 12% of the original 100% dose that had been used for routine clinical chest imaging under standard conditions were obtained. Identical postprocessing techniques were used for all dose levels. The 100%, 50%, 25%, and 12% dose levels of PA and lateral radiographs were printed as hard copies for each patient. An example of the appearance of PA images at the four different dose levels is shown in Figure 1. The images were anonymized and encoded in such a way that each of the four corresponding radiographs was randomly assigned a character with a unique number added per character. This encoding allowed us to randomize the presentation of data per patient for PA and lateral images (Excel 2000; Microsoft, Redmond, WA).

Image reading Four senior radiologists performed the image reading. The images were presented on a film viewing-box (Rotolux Planilux; Philips Medical Systems, Best, The Netherlands) with a light output of 4000 candelas per m2 and good homogeneity. All eight images for each patient were presented at the same time; one row with the four PA images and one row with the four lateral images, representing the 100%, 50%, 25% and 12% doses. For each patient, the sequence of the presented doses was The British Journal of Radiology, December 2007

randomized, and the clinical history provided by the referring clinician, as well as the clinical question to be answered, was presented to the observer. Observers were informed of the presence of differences in image quality but not about the nature or magnitude of these differences. The observers were then asked to rank the quality of the four images for each patient and each projection by assigning, in a forced choice experiment, a number to each image, ranging from one to four (representing best to least quality, respectively). Furthermore, the observers were asked to decide for each individual image whether the quality was sufficient for recognizing chest pathology and answering the clinical question. The observers worked independently and had no restriction in reading time.

Statistical analysis The dose applied was always expressed as the percentage of the initial 100% dose for that patient and for PA and lateral images separately. To evaluate to what extent observers were able to recognize a reduced dose in digital chest images, the mean dose level of the four observers was calculated separately for each of the four quality categories (ranging from one to four) for each patient and for each PA and lateral image. By doing this, the ‘‘observer effect’’ was eliminated. For PA and lateral images, the mean dose levels for the second, third and fourth quality choices were compared with mean dose levels observed for the first choice by linear regression analysis. 80 observations were made for each dose level and projection. The number of observations resulting in the judgment that perceived image quality was sufficient for assessing chest pathology and answering clinical questions was calculated.

Results Observers were able to recognize images that corresponded with a reduced dose. Linear regression analysis 985

L J M Kroft, W J H Veldkamp, B J A Mertens et al

Figure 1. (a) Digital posteroanterior chest radiograph obtained with a normal clinical dose that was taken as the 100% reference dose. (b–d) Simulated reduced dose images that were obtained with 50%, 25% and 12% of the standard of reference (100%) dose. Differences in image quality between the dose levels can be recognized visually as differences in noise levels. This is best appreciated at locations projecting rather radio-opaque structures, such as the area of the spine and diaphragm.

showed that the average dose for the second, third and fourth quality category differed significantly from the first choice (best quality) for PA images (p#0.0001) and lateral images (p,0.002). The dose reduction was much larger from the first to the third and fourth choices than from the first to second choice (Table 2). The original images (100% dose) were rated as the images with best image quality in only 51/80 (64%) of PA and 43/80 (54%) of lateral images. The four dose level images presented and their allocated quality category are displayed in Figure 2. These results indicate that, although observers were able to depict images with reduced dose, the 100% reference dose was not recognized as the best quality image in almost half of the cases. Larger dose reductions to 25% and 12% were usually recognized as either the third and fourth quality category (Figure 2). Variation was found among the radiologists’ decisions on whether images provided sufficient perceived image quality to assess chest pathology and answer clinical questions (Table 3). The quality of the PA images was graded as sufficient in all 80 observations (80/80) at the 100% dose, in 79/80 at the 50% dose, in 69/80 at the 25%

dose and in 63/80 at the 12% dose level. For lateral images, sufficient image quality was found in 79 out of 80 observations (79/80) at the 100% dose, in 80/80 at the 50% dose, in 74/80 at the 25% dose, and in 63/80 at the 12% dose. Images reduced to 50% patient dose were thus graded equal to the 100% dose images in having sufficient quality for depicting chest pathology and answering clinical questions. The 25% and 12% dose images were substantially less often graded as having sufficient imaging quality.

Discussion The main finding of this study is that, although observers were able to recognize quality differences between images corresponding to various dose levels by direct comparison, PA and lateral chest images reduced to a patient dose of 50%, as well as the original images (100% dose), were graded as having sufficient quality to answer clinical questions relating to the investigations performed in a variety of chest pathologies. At the 25% and 12% dose levels, substantially fewer observations

Table 2. The amount of dose reduction by forced choice

Second vs first choice Third vs first choice Fourth vs first choice

PA

PA p-value

Lateral

Lateral p-value

217 % 251 % 269 %

0.0001 ,0.0001 ,0.0001

213 % 238 % 263 %

0.0012 ,0.0001 ,0.0001

For each choice and for each patient, we first calculated the mean dose across observers (where dose is expressed as a percentage of the initial 100% dose) for posteroanterior (PA) and lateral images separately. Mean differences in these dose levels for the second, third and fourth choice compared with the first choice were then expressed and evaluated.

986

The British Journal of Radiology, December 2007

Dose reduction in digital chest radiography

Figure 2. Allocated first, second, third and fourth choices as a function of doses used for (a) posteroanterior and (b) lateral chest imaging.

resulted in the judgment of sufficient image quality; however, a considerable number of these 25% and 12% dose level images were still deemed of sufficient quality to answer clinical questions. Our preliminary study results suggest that imaging with doses reduced by 50% would be sufficient for depicting chest pathology and answering usual clinical questions. Dose levels lower than 50% could result in a reduction in diagnostic accuracy. Substantial variation has been found between digital systems in the doses used for clinical imaging [14, 15]. The relationship between digital imaging dose and image quality is complex. Comparative studies have shown that the contrast-detail performance, as well as diagnostic performance, depends upon technical differences between individual digital systems rather than on dose [12, 14]. With the introduction of digital radiography systems, settings that were traditionally used for screen-film techniques have been implemented for digital techniques. These conventional system settings may not be optimal for digital radiography. Various attempts are Table 3. Number of images rejected Radiologist

A B C D

PA images rejected (all doses together)

1 5 (in 3 patients) 4 (in 2 patients) 19 (in 12 patients)

Lateral images rejected (all doses together)

0 0 4 (in 2 patients) 20 (in 15 patient)

Number of images per radiologist graded as not having sufficient image quality to answer the clinical question. Each radiologist judged posteroanterior (PA) and lateral images of 20 patients with four different doses (a total of 160 images per radiologist).

The British Journal of Radiology, December 2007

made to optimize digital imaging techniques, such as improving the signal and noise characteristics of the image by applying dedicated Z filtration [16] and applying lower kV settings for improving perceived imaging quality [17]. Because of differences in technical design between the various digital systems, it may be that optimal acquisition techniques and optimal dose settings have to be determined for each digital system type. It has been suggested that careful consideration should be made when lowering the radiation dose for digital radiography used to detect subtle nodules [18]. Lesion detection studies have shown that the ability to detect subtle lesions projecting over the mediastinum decreased after halving the radiation dose of digital radiography when compared with a 100% dose that was equivalent to regular screen-film techniques; however, no significant loss of lesion detection was found in the lung fields [9, 18]. In the majority of patients referred for chest radiography, pathology such as pneumonia is sought for, or follow-up of manifest pathology is aimed for rather than detecting subtle nodules as in cases of possible metastatic disease. It may be that application of reduced dose techniques will depend on the clinical question that has to be answered. The digital image quality that was obtained in radiographs corresponds to the 100% standard-of-reference dose, and the effect of simulated substantial dose reduction was investigated from that point. One advantage of simulating reduced dose images was that patients did not need to undergo repeated imaging, and were not exposed to extra doses. Another advantage was that reduced dose simulation allowed the comparison of identical images, rather than multiple images obtained sequentially. Acquiring multiple radiographs at different dose levels introduces variations owing to positional changes as well as changes in anatomical noise [19]. Simulating reduced dose images, therefore, provided for an absolute standard-of-reference. Our study may have had some limitations. Images were only used for evaluation after chest abnormalities had been reported. Therefore, the study group represents patients with manifest chest pathology referred for chest imaging, rather than a representation of the general population. This approach was chosen as the aim was to determine in patients with chest pathology to what extent images with reduced dose still provided sufficient perceived imaging quality to answer clinical questions related to the investigations performed. We have observed that radiologists vary in their opinions when deciding whether images have sufficient quality for answering clinical questions, although rejected images were almost always in the low doseranges: 32% of the rejections were in the 25% dose level and 64% in the 12% dose level category. The variation between the observers underlines the complexity between dose, image quality and decision making in clinical radiography. In this study, preliminary results are presented for a small study group that may not have been fully representative of the large variety of different chest pathologies. Larger studies are needed to fully explore the possibility of reducing dose for various clinical questions. 987

L J M Kroft, W J H Veldkamp, B J A Mertens et al

In conclusion, although observers can recognize quality differences between imaging doses by direct comparison, chest images reduced to a patient dose of 50% were graded equal to the 100% dose in having sufficient quality to answer clinical questions. As such, preliminary findings suggest that a dose reduction of 50% seems feasible in a variety of chest pathologies, whereas further dose reduction reduces the diagnostic quality.

Acknowledgments We greatly acknowledge radiologists S J G C Frerichs, MD, A de Roos, MD, PhD, G J Vielvoye, MD, PhD and H M Zonderland, MD, PhD from the Leiden University Medical Centre, the Netherlands, for participating as observers in the study.

References 1. Strotzer M, Vo¨lk M, Fru¨nd R, Hamer O, Zorger N, Feuerbach S. Routine chest radiography using a flat-panel detector: Image quality at standard detector dose and 33% dose reduction. AJR Am J Roentgenol 2002;178:169–71. 2. Fink C, Hallscheidt PJ, Noeldge G, Kampschulte A, Radeleff B, Hosch WP, et al. Clinical comparative study with a large-area amorphous silicon flat-panel detector: Image quality and visibility of anatomic structures on chest radiography. AJR Am J Roentgenol 2002;178:481–6. 3. Ganten M, Radeleff B, Kampschulte A, Daniels MD, Kauffmann GW, Hansmann J. Comparing image quality of flat-panel chest radiography with strorage phosphor radiography and film-screen radiography. AJR Am J Roentgenol 2003;181:171–6. 4. Bacher K, Smeets P, Bonnarens K, De Hauwere A, Verstraete K, Thierens H. Dose reduction in patients undergoing chest imaging: Digital amorphous silicon flatpanel detector radiography versus conventional film-screen radiography and phosphor-based computed radiography. AJR Am J Roentgenol 2003;181:923–9. 5. van Heesewijk HPM, van der Graaf Y, de Valois JC, Feldberg MAM. Effects of dose reduction on digital chest imaging using a selenium detector: A study of detecting simulated diffuse interstitial pulmonary disease. AJR Am J Roentgenol 1996;167:403–8. 6. Strotzer M, Gmeinwieser JK, Vo¨lk M, Fru¨nd R, Seitz J, Feuerbach S. Detection of simulated chest lesions with normal and reduced radiation dose. Invest Radiol 1998;33:89–103.

988

7. Rapp-Bernhardt U, Roehl FW, Gibbs RC, Schmidl H, Krause UW, Bernhardt TM. Flat-panel x-ray detector based on amorphous silicon versus asymmetric screen-film system: phantom study of dose reduction and depiction of simulated findings. Radiology 2003;227:484–92. 8. Bernhardt TM, Otto D, Reichel G, Ludwig K, Seifert S, Kropf S, et al. Detection of simulated interstitial lung disease and catheters with selenium, storage phosphor, and film-based radiography. Radiology 1999;213:445–54. 9. Metz S, Damoser P, Hollweck R, Roggel R, Engelke C, Woertler K, et al. Chest radiography with a digital flatpanel detector: experimental receiver operating characteristic analysis. Radiology 2005;234:776–84. 10. Samei E, Saunders RS, Lo JY, Dobbins JT III, Jesneck JL, Floyd CE, et al. Fundamental imaging characteristics of a slot-scan digital chest radiographic system. Med Phys 2004;31:2687–98. 11. Veldkamp WJH, Kroft LJM, Mertens BJA, Geleijns J. Digital slot-scan charge-coupled device radiography versus amber and bucky screen-film radiography: comparison of image quality in a phantom study. Radiology 2005;235:857–66. 12. Veldkamp WJH, Kroft LJM, Boot MV, Mertens BJA, Geleijns J. Contrast-detail evaluation and dose assessment of eight digital chest radiography systems in clinical practice. Eur Radiol 2006;16:333–41. 13. Veldkamp WJH, Kroft LJM, van Delft JP, Mertens BJA, Geleijns J. Techniques for feasibility of dose reduction with a digital chest radiography system. Biomedizinische Technik 2005;50:425–6. 14. Kroft LJM, Veldkamp WJH, Mertens BJA, Boot MV, Geleijns J. Comparison of eight different digital chest radiography systems: variation in detection of simulated chest disease. AJR Am J Roentgenol 2005;185:339–46. 15. Pascoal A, Lawinski CP, McKenzie A, Tabakov S, Lewis CA. Chest radiography: a comparison of image quality and effective dose using four digital systems. Radiat Prot Dosimetry 2005;114:273–7. 16. Samei E. Dobbins JT 3rd, Lo JY, Tornai MP. A framework for optimizing the radiographic technique in digital X-ray imaging. Radiat Prot Dosimetry 2005;114:220–9. 17. Uffmann M, Neitzel U, Prokop M, Kabalan N, Weber M, Herold CJ, et al. Flat-panel-detector chest radiography: effect of tube voltage and image quality. Radiology 2005;235:642–50. 18. Kroft LJ, Veldkamp WJ, Mertens BJ, van Delft JP, Geleijns J. Dose reduction in digital posterior-anterior chest radiography: decreased detection of simulated nodules in clinical radiographs. Radiology 2006;241:392–8. 19. Ha˚kansson M, Ba˚th M, Bo¨rjesson S, Kheddache S, Grahn A, Ruschin M, et al. Nodule detection in digital chest radiography: summary of the radius chest trial. Radiat Prot Dosim 2005;114:114–20.

The British Journal of Radiology, December 2007