Scatter estimation and removal of anti-scatter grid-line artifacts from anthropomorphic head phantom images taken with a high resolution image detector

Abstract

In radiography, one of the best methods to eliminate image-degrading scatter radiation is the use of anti-scatter grids. However, with high-resolution dynamic imaging detectors, stationary anti-scatter grids can leave grid-line shadows and moiré patterns on the image, depending upon the line density of the grid and the sampling frequency of the x-ray detector. Such artifacts degrade the image quality and may mask small but important details such as small vessels and interventional device features. Appearance of these artifacts becomes increasingly severe as the detector spatial resolution is improved. We have previously demonstrated that, to remove these artifacts by dividing out a reference grid image, one must first subtract the residual scatter that penetrates the grid; however, for objects with anatomic structure, scatter varies throughout the FOV and a spatially differing amount of scatter must be subtracted.

In this study, a standard stationary Smit-Rontgen X-ray grid (line density - 70 lines/cm, grid ratio - 13:1) was used with a high-resolution CMOS detector, the Dexela 1207 (pixel size - 75 micron) to image anthropomorphic head phantoms. For a 15 × 15cm FOV, scatter profiles of the anthropomorphic head phantoms were estimated then iteratively modified to minimize the structured noise due to the varying grid-line artifacts across the FOV.

Images of the anthropomorphic head phantoms taken with the grid, before and after the corrections, were compared demonstrating almost total elimination of the artifact over the full FOV. Hence, with proper computational tools, anti-scatter grid artifacts can be corrected, even during dynamic sequences.

1. INTRODUCTION

In digital radiographic imaging, the image quality is degraded substantially by the scatter produced when the primary beam passes through an object. Different techniques such as the use of air-gaps¹, scanning beams², collimation, moving grids³ and stationary grids may be implemented to control the amount of scatter. Due to its compact design and simplicity compared to scanning beams or moving grids, stationary grids have usually proved to be the most practical choice for fluoroscopy. The grid enables adequate control of scatter without increasing geometric un-sharpness as in the case of air-gap techniques and hence they are widely used in projection x-ray radiography to reduce scatter radiation and improve image contrast.

High resolution capabilities are also essential for an efficient, accurate, and successful endovascular interventional procedure⁴. However, when grids are used with these high resolution imaging detectors, the images contain a fixed-structure pattern noise or grid line artifacts and moiré patterns⁵. When these anti-scatter grids are used with a standard FPD, the grid lines are not very apparent because of the lower spatial resolution of the detector, which results in an averaging or blurring of the lines. However, for high resolution detectors, this fixed- pattern noise grid artifact must be removed from the images before being acceptable for clinical use.

Grid-line removal could be done by log subtracting a mask containing the fixed pattern grid-line artifact; however residual scatter prevents complete cancellation of the pattern. The scatter component must thus be estimated and linearly subtracted first.

The purpose of this work is to estimate the scatter distribution of complex realistic structured images where the amount of scatter can vary across the FOV so that the method of using the flat field obtained with the grid can be used to remove the anti-scatter grid-line artifacts in complex images, such as obtained during clinical interventions.

2. METHOD AND MATERIALS

In this study in the set-up shown in Figure 1, we used a stationary Smit-Rӧntgen x-ray grid with a high resolution Dexela 1207 CMOS X-ray detector (pixel size 75 μm and sensitivity area 11.5cm × 6.5cm) (Fig. 2) to image anthropomorphic head phantoms: SK-150, The Phantom Laboratory, NY (Fig 3) and RS-240T, Radiology Support Devices Inc., CA (Fig. 4).

Fig 1.

Fig 2.

External view of CMOS detector

Fig 3.

SK-150

Fig 4.

RS-240T

Studies previously reported by us involved a simulated artery block phantom with a uniform frontal head equivalent phantom used as the scattering source to investigate the effectiveness of the method to remove the anti-scatter grid-line artifacts when a grid is used with a Flat Panel Detector (FPD) and high resolution CMOS detector^{6, 7}.

Grid lines were quite prominent when anthropomorphic head phantoms, SK-150 and RS-240T, were imaged with the high resolution CMOS detector with grid placed in front of the detector. Dividing by the flat-field image obtained with the grid does not effectively remove those grid-line artifacts because there is residual scatter that is additive while primary x-ray attenuation is a multiplicative factor. Hence, it is important to estimate and subtract the scatter in the image before dividing by the flat-field image.

Scatter Estimation and Image Corrections

In order to estimate this residual scatter, we placed round lead markers (1 mm thick with 3 mm diameter) under the SK-150 anthropomorphic head phantom at different locations and imaged it using the high resolution CMOS detector with a grid placed in front of the detector. Theoretically, there should not be any signal on the image under the lead markers as these markers are thick enough to block all the incident x-rays. Hence, if any signal under the lead markers is observed, that is nothing but scatter. Grayscale values under all the lead markers were obtained at different location of the image. Since scatter has a very low spatial frequency distribution, these grayscale values under the lead markers were interpolated between markers to get an estimated scatter profile over the whole image. This estimated scatter profile is taken as the starting point for the iterative correction process.

In the iteration process of estimating the scatter, the image to be corrected and the estimated scatter profile is divided into 25 small parts. Then each part of the image is corrected individually by iteratively varying the scatter value in the respective section of the scatter profile, to minimize the standard deviation of the grayscale values in that part of the image. In the previous study, we showed that as the grid lines (structure pattern) are removed from the image, the standard deviation of pixel values in the image also goes down⁸. Once all the parts are corrected, they are combined back to form the corrected original image. Approximately, 8–9 iterations were required to get the corrected image and the time taken was approximately 0.7 seconds per iteration. However, by using powerful computing systems and parallel programming techniques, the iteration time can be reduced a lot and hence this method can be used effectively with dynamic and changing fields such as used in neuro-endovascular image-guided interventional procedures.

In order to further evaluate the effectiveness of this method, we corrected the images of an RS-240T anthropomorphic head phantom taken with the high resolution CMOS detector with grid in place. For correcting these images of the RS-240T phantom, we did not obtain the separate scatter profile by placing lead under the RS-240T phantom, but rather used the already estimated scatter profile (the one we obtained by placing lead in the beam for SK-150 anthropomorphic head phantom) as the starting point for the iteration method. Once again, we observed that the grid pattern was removed quite successfully, establishing the point that only one estimation of the scatter profile is needed and then the same profile can be used successfully as the starting point of iteration method for different heads. A total of 9 iteration were required to get the corrected image of second head phantom.

The grid used in this study is the standard grid presently used for the FPD. The following Table 1 shows the relevant specifications of the grid:

Table 1.

Grid specifications

Type number	989601061091
Line rate	70 line/cm
Grid ratio	13 : 1
Absorption material	Lead, 27.5 μ
Interspace material	Fiber

The images were taken in three different situations, explained below:

1. Table with grid (Grid mask image): We started first by “Table with grid” image acquisition. We kept the frontal head phantom (with the uniform section of the artery block inserted) on the x-ray tube (in order to keep filtration of the x-ray beam consistent). The grid is placed in the front of the CMOS detector and patient table is in the beam.

2. Object with grid (Image to be corrected): Second set of images taken were “Object with grid”. In this case, we carefully removed the frontal head phantom from the x-ray tube and placed one of the anthropomorphic head phantoms in the beam.

3. Object without grid: Then we took “Object without grid” images for the comparison purposes. In this case, we carefully removed the grid from the front of the CMOS detector keeping the rest of the set-up exactly the same as “Object with grid”.

Exposure parameters used for imaging were 100 kVp, 100 mA and 12 ms, and they were kept the same for all acquisitions. Hence, all the exposures were the same in all the images, with or without the grid. The FOV used was 15 cm × 15 cm at the image receptor.

3. RESULTS AND DISCUSSION

Fig 5 shows the image of the SK-150 anthropomorphic head phantom imaged with the FPD detector. This phantom shows the bone structure of the head. The dotted region shows the portion of the RS-240T head phantom which is imaged by the smaller FOV of the CMOS detector as shown in Fig 6. This region is around the base of the skull and the Circle of Willis is behind it.

Fig 5.

SK-150 anthropomorphic head phantom imaged with the FPD.

The dotted region shows the portion of the head phantom imaged by the CMOS detector as shown in Fig 6.

Fig 6.

SK-150 phantom imaged with grid attached to CMOS detector.

Grid lines were quite prominent when the anthropomorphic head phantoms, SK-150 and RS-240T, were imaged with the high resolution CMOS detector with grid placed in front of the detector (Fig 6). However, after following the earlier mentioned method, we successfully removed the residual grid lines.

Fig 6 shows the image of the SK-150 anthropomorphic head phantom obtained with the high resolution CMOS detector, with grid placed in front of the detector.

In order to show the effectiveness of the correction method, we zoomed in into three different regions of Fig. 6 and show them separately in Figs. 7, 8, 9 and 10. We tried to choose those regions of the image in which the grayscale value is varying throughout the region, so that the effectiveness of the method of removing the grid-line artifacts can be tested in those portions of the image where there are a lot of anatomical features/details.

Fig 7.

Final scatter distribution that was subtracted from Fig 6 to remove the residual anti-scatter grid lines (brighter indicates higher scatter).

Fig 8.

(a) Object with grid, (b) Object with grid: Table and Grid corrected, (c) Object with grid – Table, Grid & Scatter corrected (d) Object without grid

Fig 9.

(a) Object with grid, (b) Object with grid: Table and Grid corrected, (c) Object with grid – Table, Grid & Scatter corrected (d) Object without grid

Fig 10.

(a) Object with grid, (b) Object with grid: Table and Grid corrected, (c) Object with grid – Table, Grid & Scatter corrected (d) Object without grid

We have compared the following four cases in the zoomed regions.

1. Object with grid.

2. Object with grid: Table and Grid correction – In this case, we divided the “Object with grid” image by the “Grid reference image”. No scatter is subtracted in this case.

3. Object with grid: Table, Grid and Scatter correction – In this case, we first subtracted the scatter (Fig. 7) from the “Object with grid” image, before dividing it by the “Grid reference image”.

4. Object without grid.

We also calculated the contrast and contrast–to-noise ratio (CNR) in the zoomed-in images for quantitative comparison purposes using the following equations:

$$\mathit{\text{Contrast}}(\%)=\frac{I_{\mathit{\text{background}}}-I_{\mathit{\text{signal}}}}{I_{\mathit{\text{background}}}}\times 100,\mathit{\text{Noise}}(\%)=\frac{{\sigma }_{\mathit{\text{background}}}}{I_{\mathit{\text{background}}}}\times 100,\mathit{\text{CNR}}=\frac{\mathit{\text{Contrast}}}{\mathit{\text{Noise}}}$$

In order to calculate the contrast and CNR, we chose two regions in the image as ‘signal’ and ‘background’. A rectangular box (7 pixels × 9 pixels) was selected in small darker regions to serve as ‘signal’ and the same size box was selected in a somewhat lighter region close by as ‘background’. The solid box and dotted box represent the signal region and the background region respectively in the images of Figs. 8, 9, and 10. The Contrast and CNR is then calculated and compared in the same regions for all the above mentioned four cases. We observed that the improvement in contrast in all the images is quite large as compared to the respective CNR improvement. It is mainly because the exposure parameters for all the images were kept the same, irrespective of whether grid is used or not. Without the grid we get more scatter. However, when the grid is used, we not only reduce the scatter and increase structures noise but also increase quantum noise because of the fewer primary photons that can get through the grid.

Fig. 11 shows the image of the RS-240 anthropomorphic head phantom imaged taken with the FPD detector. This phantom shows the iodine-filled vessels in the brain. Some of the important arteries like the Anterior Cerebral Artery (ACA), Middle Cerebral Artery (MCA), Internal Carotid Artery (ICA) and Petro Cavernous Artery are identified in Fig. 11. The dotted region shows the portion of the RS-240T head phantom which is imaged by the CMOS detector as shown in Fig. 12.

Fig 11.

RS-240T anthropomorphic head phantom imaged with the FPD.

The dotted region shows the portion of the head phantom imaged by the CMOS detector as shown in Fig 12.

Fig 12.

RS-240T phantom imaged with the grid attached to the CMOS detector

Fig. 12 shows the image of the RS-240T anthropomorphic head phantom imaged with the CMOS detector with the grid in place. Figs. 13, 14 and 15 show the zoomed images of three different regions in the image. Fig 13 shows the portion of Internal Carotid artery. Fig 14 focuses on portion of the petro cavernous artery. Fig 15 shows the bone of the nose.

Fig 13.

(a) Object with grid, (b) Object with grid: Table and Grid corrected, (c) Object with grid – Table, Grid & Scatter corrected (d) Object without grid

Fig 14.

(a) Object with grid, (b) Object with grid: Table and Grid corrected, (c) Object with grid – Table, Grid & Scatter corrected (d) Object without grid

Fig 15.

(a) Object with grid, (b) Object with grid: Table and Grid corrected, (c) Object with grid – Table, Grid & Scatter corrected (d) Object without grid

4. CONCLUSIONS

We demonstrated that the images of anthropomorphic head phantoms taken with a grid using a high resolution CMOS detector can be corrected for anti-scatter grid line artifacts if we accurately estimate the scatter distribution present across the image. We estimated the scatter iteratively for the complex images of head phantoms by determining the scatter value providing a minimum pixel variance in selected regions in the image when the grid-correction method (i.e. division by the reference uniform field grid image) was used. We demonstrated an improvement in contrast and CNR of the images taken with the grid and corrected for the anti-scatter grid pattern, as compared to the images taken without a grid. Thus, with the use of the described method and advance computational tools, we should be able to correct the images for anti-scatter grid-line artifacts, even during dynamic sequences such as used in neuro-endovascular image-guided interventional procedures.