Published online Apr 06, 2018.
https://doi.org/10.3348/kjr.2018.19.3.526
The Potential Role of Grid-Like Software in Bedside Chest Radiography in Improving Image Quality and Dose Reduction: An Observer Preference Study
Abstract
Objective
To compare the observer preference of image quality and radiation dose between non-grid, grid-like, and grid images.
Materials and Methods
Each of the 38 patients underwent bedside chest radiography with and without a grid. A grid-like image was generated from a non-grid image using SimGrid software (Samsung Electronics Co. Ltd.) employing deep-learning-based scatter correction technology. Two readers recorded the preference for 10 anatomic landmarks and the overall appearance on a five-point scale for a pair of non-grid and grid-like images, and a pair of grid-like and grid images, respectively, which were randomly presented. The dose area product (DAP) was also recorded. Wilcoxon's rank sum test was used to assess the significance of preference.
Results
Both readers preferred grid-like images to non-grid images significantly (p < 0.001); with a significant difference in terms of the preference for grid images to grid-like images (p = 0.317, 0.034, respectively). In terms of anatomic landmarks, both readers preferred grid-like images to non-grid images (p < 0.05). No significant differences existed between grid-like and grid images except for the preference for grid images in proximal airways by two readers, and in retrocardiac lung and thoracic spine by one reader. The median DAP were 1.48 (range, 1.37–2.17) dGy*cm2 in grid images and 1.22 (range, 1.11–1.78) dGy*cm2 in grid-like images with a significant difference (p < 0.001).
Conclusion
The SimGrid software significantly improved the image quality of non-grid images to a level comparable to that of grid images with a relatively lower level of radiation exposure.
INTRODUCTION
The use of bedside chest radiography has been increasing continually with the increase in aging population and the need for care of critically ill patients. Bedside chest radiography provides information that may not be obtained clinically such as malposition of invasive devices or detection of cardiopulmonary deterioration, which is crucial for disease management (1, 2). However, bedside examination has a major limitations involving image degradation by scattered radiation (3). Although anti-scatter grid improves the contrast by reducing the scattered radiation, many hospitals do not use grid in bedside examinations for various reasons. Using a grid in a bedside investigation is challenging due to the risk of misalignment and grid cutoff, which reduces diagnostic information. Additional radiation is required for repeat examinations if necessary. Furthermore, grid transportation and positioning is an additional technological burden, and grid usage typically entails a higher radiation dose. Eventual damage requires grid installation and replacement, and related costs. To overcome these limitations, researchers developed software that provides scatter correction without the physical use of a grid (4, 5, 6, 7). The technology enables estimation of the scatter signal and calculation of the grid effect. The recently introduced ‘SimGrid’ scatter correction software (Samsung Electronics Co. Ltd., Suwon, Korea) allows the estimation of the distribution and degree of scatter radiation using raw image data directly with pretrained Convolutional Neural Networks. These networks have been trained and optimized using tens of thousands of anthropomorphic phantoms and clinical images. Thus, it is robust under varying conditions of exposure, patient status, and positional changes. The estimated scatter image is subtracted from the raw image to produce a de-scattered image. Subsequently, the de-scattered raw image is processed by the post-processing software of the digital radiography system. Although the value of scatter correction in terms of physical properties using a phantom has been reported (6, 7), no test of observer preference has been performed to facilitate clinical application of this software.
Therefore, the purpose of our study was to compare the observer preference of image quality and radiation dose between non-grid and grid-like images generated by scatter correction software, and between grid-like and grid images.
MATERIALS AND METHODS
Support and Funding
This study was financially and technically supported by Samsung Electronics Corporation. However, the company had no control over any data or information submitted for publication or over any data or information included in this study.
Patients
This prospective study was approved by the Institutional Review Board of our institution and written consent was obtained from all participating patients. Patients who presented for scheduled routine follow-up for chest radiography at our institution were considered for inclusion. Patients who were unable to provide written informed consent or who were pregnant or expecting pregnancy in the near future were excluded. Patients who consented to chest radiography with and without grid were enrolled between January 17 2017 and February 17 2017. Our study population included a total of 38 patients (male-to-female ratio, 25:13; median age, 57 years; age range, 24–78 years).
Image Acquisition
Participants underwent chest radiography with the anteroposterior projection. All the investigations were performed with a mobile X-ray system GM85 (Samsung Electronics Co. Ltd.) using a 345 × 426 mm detector. Grid images were acquired using a 6:1 grid with 85 L/cm density and a focal distance of 130 cm. Three image sets were obtained for each patient. An image was acquired with grid (grid image), and another image without grid (non-grid image) immediately to minimize the positional change. The other image was obtained via post-processing with SimGrid software using non-grid image (grid-like image). Exposure parameters were as follows: source-to-object distance, 100 cm; focal spot size, 1.2 mm; tube voltage, 80–90 kVp; tube current 230–240 mA; and exposure time, 5 ms.
Radiation Dose Estimations
All the exposure factors were recorded. The dose area product (DAP [cGy·cm2]) was calculated using the S-DAP algorithm (Samsung Electronics Co. Ltd.), which enables calculation of the area dose under conditions such as tube voltage, tube current, the presence or absence of additional filters, filter thickness, and collimator size, instead of using DAP meter.
Image Evaluation
Two radiologists (Reader 1 and 2, with 26 and 5 years of experience in chest radiology, respectively) independently compared the paired images of (a) non-grid and grid-like images followed by (b) grid-like and grid images randomly. The readers were blinded to patient and clinical demographics. All the data were stored in Digital Imaging and Communications in Medicine format. The window width and window level of the images were automatically optimized by the customized program. Readers were allowed to adjust the brightness and contrast of the images. The anatomic landmarks used for image evaluation were similar to those of previous studies dealing with chest radiography (8). The 10 anatomic landmarks included the unobscured lung, hilum, minor fissure, retrocardiac lung, subdiaphragmatic lung, azygoesophageal recess, heart border, rib, proximal airway, and thoracic spines. The overall image quality was also evaluated. Images were graded using a five-point scale from one to five (1, strongly preferred A; 2, somewhat preferred A; 3, no preference; 4, somewhat preferred B; 5, strongly preferred B [A = initial image, B = second image]). The score was rearranged to A as non-grid images and B as grid-like images in test 1, and to A as grid images and B as grid-like images in test 2 since each pair of images was randomly arranged.
Sample Size Calculation
Sample size calculation was performed with the StatsDirect Statistical Software Version 3.1.11 (StatsDirect Ltd, Altrincham, UK), which revealed a sample size of 34 matched pairs of subjects to yield a power of 0.8, α = 0.05 for the detection of a mean difference of 11 with an estimated standard deviation of 22. The mean difference was determined based on the range rule for standard deviation. Considering a dropout rate of 10%, a sample size of 38 was required.
Statistical Analysis
Wilcoxon's signed rank test was used to assess the significance of differences between ratings. An analysis of each reviewer was performed. As normal distribution of DAP was rejected by the Kolmogorov-Smirnov test, Wilcoxon's signed rank test was used to compare the radiation dose. All the statistical tests were two-sided with a significance level of 0.05. The statistical analysis was performed using SPSS (SPSS version 23.0; IBM Corp., Armonk, NY, USA).
RESULTS
Preference Study
Non-Grid vs. Grid-Like Images
The composite data for both readers in each anatomic region are summarized in Table 1.
Table 1
Comparison of Non-Grid Images with Grid-Like Images for 11 Anatomic Regions
The interpreters strongly preferred grid-like images to non-grid images for all anatomic landmarks and overall image quality (p < 0.05) (Figs. 1, 2, 3). The most preferred region included unobscured lung (mean, 4.4), followed by thoracic spine (mean, 4.3), overall appearance (mean, 4.1), rib (mean, 4.1), retrocardiac lung (mean, 4.1), hilum (mean, 3.9), subdiaphragmatic lung (mean, 3.8), proximal airway (mean, 3.8) and heart border (mean, 3.7) and azygoesophageal recess (mean, 3.5). The minor fissure was the least, but also preferred region (mean, 3.4).
Fig. 1
Comparison of (A, D) non-grid, (B, E) grid-like, and (C, F) grid images.
A-C. Chest radiography of 57-year-old male patient who underwent multiple-wedge resection in both lungs for metastasis associated with undifferentiated pleomorphic sarcoma. Note superior image contrast of grid-like image to non-grid image, and similarity in contrast appearance between grid-like and grid images. D-F. Magnified non-grid, grid-like and grid images: surgical materials (black arrows) and multiple linear opacities (white arrows) are clearly demonstrated in grid-like image, as well as grid image.
Fig. 2
Magnified anteroposterior chest radiographs of 59-year-old female with ovarian cancer.
Compared with (A) non-grid image, delineation of trachea and left main bronchus (arrows) is improved in (B) grid-like image, as well as (C) grid image.
Fig. 3
Magnified anteroposterior chest radiographs of 63-year-old male with hepatocellular carcinoma.
Compared with (A) non-grid image, delineation of vascular shadows in retrocardiac area is improved in (B) grid-like image, as well as (C) grid image.
Grid vs. Grid-Like Images
Neither reader indicated preference for any of the seven regions (unobscured lung, hilum, minor fissure, rib, heart border, subdiaphragmatic lung, azygoesophageal recess) (Fig. 1). Grid images were preferred to grid-like images in proximal airways by both readers (p = 0.002 and 0.001, respectively) (Fig. 2). For retrocardiac lung and thoracic spine, Reader 1 preferred grid-images to grid-like images (p < 0.001 and 0.014, respectively) (Fig. 3). Reader 1 rated equally for overall appearance (p = 0.317), while Reader 2 preferred grid images (p = 0.034). The composite data for both readers in each anatomic region are summarized in Table 2.
Table 2
Comparison of Grid Images with Grid-Like Images for 11 Anatomic Region
Radiation Dose
The median DAP were 1.48 (range, 1.37–2.17) dGy*cm2 in grid images and 1.22 (range, 1.11–1.78) dGy*cm2 in non-grid and grid-like images. Wilcoxon's signed rank test demonstrated a significant difference between grid and grid-like images (p < 0.001).
DISCUSSION
Grid-like images were highly preferred to non-grid images in all the 10 anatomic landmarks and overall appearance. It is noteworthy the quality of grid-like images was equivalent to that of grid images.
The highest scoring regions included unobscured lungs, followed by thoracic spines and retrocardiac lungs using SimGrid compared with non-grid images. Contrast improvement using an anti-scatter grid is known to be associated with local scatter fraction. The scatter fraction is the ratio of the intensity of scatter radiation to that of total (scattered plus unscattered) radiation recorded on the image. Scatter fraction values were higher in the mediastinum and retrocardiac regions than the lungs (9), which explains our findings indicating that retrocardiac lungs and thoracic spines were among the most preferred regions. A greater scatter occurring in these regions suggested higher chances of restoration of the contrast using scatter correction software. Interestingly, unobscured lungs were the most preferred areas despite a relatively low scatter fraction. Unobscured lungs occupy the largest area in a chest radiograph and, under high natural contrast, readers may recognize significant improvements in this region.
Scatter radiation has been reduced traditionally by modifying the image acquisition process using hardware. The introduction of a grid for mobile applications improved the image quality (10, 11, 12, 13). However, only 12% of radiologists and technologists used anti-scatter grids for bedside chest radiography according to Wandtke's survey (3). The most frequent suggestion to improve the image quality of bedside chest radiography was to use grids regularly and consistently. Laser-alignment and automatic grid alignment systems were developed to improve the control of grid positioning (14, 15). The recently developed bedside digital radiography is equipped to display the current detector angle on the screen for proper alignment. However, despite such developments facilitating tube detector alignment, the grid has yet to be used in bedside examinations widely. Precise positioning and alignment are still difficult because patients are still the focus of technologists rather than equipment screen when determining the grid position. Further, cumbersome workflow parameters include grid attachment and detachment, as well as disinfection after usage. Concerns related to additional radiation dosage when using grid or performing repeated examinations due to grid misalignment also need to be addressed.
With the introduction of digital imaging, an alternate approach has been developed to compensate for image degradation through digital image processing. Recently, several vendors have developed new software that estimates the scattered radiation and provides scatter correction without the usage of grids (4, 5, 7, 16). The technology saved 15 seconds to 34 seconds per examination on average compared with grid-based workflow (5, 17), resulting in workflow improvement by 28% (17). Mentrup et al. (6) conducted experiments investigating the performance of a software-based scatter correction and demonstrated contrast improvement by software consistent with the results obtained using the grid quantitatively. It is in accordance with our results showing that the quality of grid-like images is comparable to that of grid images while reducing radiation dose by 18.7%. The findings suggest that SimGrid software facilitates reduction in dose as well as workload.
There are several limitations in this study. First, non-grid images were easily identified due to conspicuous differences even though they were blinded, which could lead to a bias. However, it was not easy to differentiate grid-like from grid images. Second, the study population is composed of patients who visited the outpatient clinic, suggesting differences under a clinical setting. Most patients in bedside settings do not cooperate with chest examination such as positioning and breath-holding. However, these conditions pose a bigger challenge in grid images than in grid-like images. Third, as this is an observer preference study, it is unknown whether a better preference can be interpreted as better observer performance. Further studies investigating the impact of SimGrid software on diagnostic performance are needed. Most importantly, under optimal conditions excluding bedside radiography, grid-based radiography is still superior to non-grid radiography. Therefore, the results do not indicate if this application can replace radiography using a grid under all circumstances.
In conclusion, SimGrid software improved the image quality of non-grid images significantly to a level comparable to that of grid images under a relatively lower level of radiation exposure.
This study was supported in part by grant from the Samsung Electronics Research Fund and in part by a grant of the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (grant number : HI15C1532).
References
-
Kawamura T, Naito S, Okano K, Yamada M. Improvement in image quality and workflow of x-ray examinations using a new image processing method, “Virtual Grid Technology”. Web site. 2015 [Accessed July 13, 2017].http://www.fujifilm.com/about/research/report/060/pdf/index/ff_
rd060_ 004_ en.pdf.
-
-
Mentrup D, Specht S, Spieckermann J, Zorba G. Gridless imaging with SkyFlow: time savings and workflow improvements. 2015 Jun; [Accessed July 13, 2017].
-
-
Lee K, Lee K, Lee S, Kim H. Workflow improvement with Samsung SimGrid SW. Web site. 2016 Feb 22; [Accessed July 13, 2017].
-
Figures
1 / 3
Tables
1 / 2
