search
Contact Us
HOME

  • Crossref logo
  • Crossref Similarity Check logo
Journal Search Engine

to

Search Advanced Search Adode Reader(link)
Download PDF Export Citaion korean bibliography PMC previewer
ISSN : 2288-3509(Print)
ISSN : 2384-1168(Online)
Journal of Radiological Science and Technology Vol.47 No.3 pp.175-182
DOI : https://doi.org/10.17946/JRST.2024.47.3.175

Evaluation of Performance and No-reference-based Quality for CT Image with ADMIRE Iterative Reconstruction Parameters: A Pilot Study

Bo-Min Park1), Yoo-Jin Seo1), Seong-Hyeon Kang2), Jina Shim3), Hajin Kim4), Sewon Lim4), Youngjin Lee1)
1)Department of Radiological Science, Gachon University
2)Department of Biomedical Engineering, Eulji University
3)Department of Diagnostic Radiology, Severance Hospital
4)Department of Health Science, General School of Gachon University
Corresponding author: Youngjin Lee, Department of Radiological Science, Gachon University, 191 Hambakmoe-ro, Yeonsu-gu, Incheon 21936, Republic of Korea / Tel: +82-32-820-4362 / E-mail: yj20@gachon.ac.kr
03/04/2024 11/04/2024 13/05/2024

Abstract


Advanced modeled iterative reconstruction (ADMIRE) represents a repetitive reconstruction method that can adjust strength and kernel, each of which are known to affect computed tomography (CT) image quality. The aim of this study was to quantitatively analyze the noise and spatial resolution of CT images according to ADMIRE control factors. Patient images were obtained by applying ADMIRE strength 2 and 3, and kernel B40 and B59. For quantitative evaluations, the noise level, spatial resolution, and overall image quality were measured using coefficient of variation (COV), edge rise distance (ERD), and natural image quality evaluation (NIQE). The superior values for the average COV, ERD, and NIQE results were obtained for the ADMIRE reconstruction conditions of ADMIRE 2 + B40, ADMIRE 3 + B59, and ADMIRE3 + B59. NIQE, which represents the overall image quality based on no-reference, was about 6.04 when using ADMIRE 3 + B59, showing the best result among the reconstructed image acquisition conditions. The results of this study indicate that the ADMIRE strength and kernel chosen for use in ADMIRE reconstruction have a significant impact on CT image quality. This highlights the importance of adjusting to the control factors in consideration of the clinical environment.


Computed tomography , Advanced modeled iterative reconstruction (ADMIRE) , Iteration intensity and kernel , Noise level and spatial resolution evaluations , Blind image quality evaluation

ADMIRE 반복적 재구성 파라메터에 따른 CT 영상의 특성 및 무참조 기반 화질 평가: 선행연구

박보민1), 서유진1), 강성현2), 심지나3), 김하진4), 임세원4), 이영진1)
1)가천대학교 방사선학과
2)을지대학교 의료공학과
3)세브란스병원 영상의학과
4)가천대학교 일반대학원 보건과학과

초록

전산화단층영상검사 , Advanced modeled iterative reconstruction (ADMIRE) , 반복 강도 및 커널 , 노이즈 레벨 및 공간 분해능 평가 , Blind 영상 화질 평가

    Ⅰ. Introduction

    Computed tomography (CT) is a key method used for diagnosing lesions in the medical field[1-3]. However, with an increase in the usage of CT, issues have arisen regarding increased exposure to radiation dose among patients[4,5]. According to reports from the North Staffordshire Hospital in the United Kingdom, a correlation between exposure and the occurrence of cancer in spiral CT scans has been established. Thus, as the range of CT scans widens, the risk of cancer increases, with the risk of cancer varying based on the sensitivity of the organs in the exposed region[6]. Therefore, efforts to reduce the radiation dose of CT are necessary, with many researchers having recently proposed various methods.

    The most effective way to reduce the dose of radiation during CT scans is by lowering the scan conditions. However, reducing these doses gives rise to greater noise in the CT images, thereby affecting the image quality. This highlights the conflict between the goals of dose reduction and image quality, whereby achieving one goal often sacrifices the other[7]. As a result, advanced image reconstruction techniques have been developed to address the issue of radiation dose while maintaining CT image quality.

    One of the most representative CT image reconstruction methods is filtered back-projection (FBP)[8]. Although this method has been most widely used since the development of CT, it has many limitations in improving image quality when acquiring low-dose based images[9]. To overcome these drawbacks, various iterative reconstruction (IR) methods have been developed and are now widely used for CT image acquisition. IR methods can be classified into hybrid and model-based algorithms. Among these methods, advanced modeled iterative reconstruction (ADMIRE), a model-based reconstruction method developed by Siemens, is known to have various advantages[10,11]. According to research by Anja Örgel et al., the signal-to-noise ratio could be improved in CT angiography images using the ADMIRE method compared to the conventional FBP reconstruction method[12].

    Fundamentally, ADMIRE allows medical professionals to adjust the strength and kernel parameters to derive the desired CT result. This method can be used in five different strengths, significantly impacting the CT image quality. In general, although the noise of the image decreases as the strength increases, leading to improved quality, the temporal resolution also tends to decrease as a result[13,14]. Furthermore, the kernel is dependent on CT image quality based on ADMIRE. While CT images that use a relatively sharp kernel tend to have more noise, their spatial resolution is improved. In contrast, a smoother kernel can be used to obtain CT images with less noise, although this inevitably results in a reduced spatial resolution[15,16].

    As a result, the contributions of the ADMIRE strength and kernel parameters to image quality and CT images vary depending on the control factors[17]. In this study, we aimed to quantitatively analyze the noise and spatial resolution of CT images based on ADMIRE control factors mainly used in clinical settings. For that purpose, coefficient of variation (COV) and edge rise distance (ERD) were used as factors for the evaluation of the resulting noise and spatial resolution. Additionally, a no-reference based natural image quality evaluation (NIQE) was used to analyze the overall image quality.

    Ⅱ. Materials and methods

    1. CT system and image acquisition protocol

    This study was conducted following the Helsinki Declaration and received prior approval for exemption from the Institutional Review Board (IRB no. 4-2022- 0356). The image used in the study is a low-dose chest CT image of a 71-year-old woman, where the view of the pulmonary nodule is not obstructed by image artifacts.

    The SOMATOM Force CT system (Siemens Healthineers, Erlangen, Germany) was utilized for this study. The imaging acquisition conditions followed a protocol commonly used for low-dose chest CT, including 100 kVp tube voltage and Care Dose 4D automatic exposure control. Additionally, 2.5 pitch and 0.25 seconds of rotation time were applied, and tin materials were used as additional filters when obtaining images. The slice thickness and increment were each set at 1 mm, and the detector collimation was set at 192 × 0.6 mm2. The CT dose index volume and dose length product were 0.32 and 14.1, respectively. The reconstruction method adjusted CT parameters with the strength of ADMIRE 2 and 3, and the kernel of B40 (smooth) and B59 (sharp). For ADMIRE's strength setting, the popular 2 level was additionally used based on the 3 level, which is basically set to the default value[18].

    A total of three slice images were obtained for each ADMIRE parameter. Slice 1 acquired an Aortic arch image where only the trachea was visible, Slice 2 acquired an image showing the middle of heart, and Slice 3 acquired an image of the end of the heart.

    2. Quantitative evaluation of image quality

    Various quantitative evaluation factors were used to analyze the noise levels, spatial resolutions, and overall image quality in the obtained CT images. COV and ERD factors were used to evaluate the noise level and spatial resolution, respectively. The formula for calculating COV was as follows:

    C O V = σ A S A   Eq.
    (1)

    where σA and SA represent the standard deviation (SD) and mean value of intensity in target area, respectively. Additionally, the value for ERD indicates the distance between two points across the structural boundaries of the image at 10% and 90% of the maximum Hounsfield unit (HU), wherein a smaller ERD indicates a better spatial resolution[19]. The region of interest (ROI) used for COV and ERD measurement on the obtained CT sample image is shown in Fig. 1. The ROI used for the COV and ERD measurements are denoted as ROI_A and ROI_B, respectively. For the ROI setting, the same area was applied to all CT images, and ROI_A, which can measure the noise level in a uniform form, and ROI_B, which can clearly identify the edge area, were used.

    To assess the overall image quality, we utilized the no-reference based Natural Image Quality Evaluator (NIQE) proposed by Mittal et al. NIQE calculates a specific distance between distorted images and the original image data, with lower values indicating higher similarity to ideal images and better image quality[20].

    Ⅲ. Results and Discussion

    1. Results of visual evaluation

    The images obtained for each ADMIRE strength and kernel are shown in Fig. 2. Fig. 2(a), (b), and (c) show the CT results of the aortic arch, middle of heart, and end-of-heart regions, respectively. When visually evaluating the resulting images, noise was reduced in ADMIRE 3 compared to ADMIRE 2 with the same kernel. Additionally, we confirmed that the edges appeared slightly clearer in the image when ADMIRE 3 was used compared to the CT image acquired using ADMIRE 2. In addition, comparing the CT images according to the kernel at the same ADMIRE intensity showed a tendency for noise to increase at B59 compared to B40. In particular, we were able to confirm that the edge area was clearly observed in the CT image using the B59 kernel compared to the B40 kernel.

    A COV graph of the CT images obtained according to the ADMIRE reconstruction method used is provided in Fig. 3(a). The average COV values measured in the three slices were 0.313, 0.306, 0.413, and 0.379 under the reconstruction conditions of ADMIRE 2 + B40, ADMIRE 3 + B40, ADMIRE 2 + B59, and ADMIRE 3 + B59, respectively. These results indicate that the best value was obtained in those CT images achieved using the ADMIRE 3 + B40 kernel. The average COV result of CT images obtained using ADMIRE 3 was approximately 0.343, or an approximately 5.50% improvement compared to ADMIRE 2. In addition, the average COV value of the CT images acquired using the B40 kernel represented an improvement of about 21.75% compared to the B59 kernel. These findings indicate that the COV value, which can comprehensively evaluate the noise level, improves as the strength of ADMIRE increases and the kernel size decreases.

    Similarly, a graph of the ERD results corresponding to the CT images obtained according to the ADMIRE reconstruction method used is provided in Fig. 3(b). The average ERD values measured in three slices were 2.03, 1.96, 0.95, and 0.91 at the reconstruction conditions of ADMIRE 2 + B40, ADMIRE 3 + B40, ADMIRE 2 + B59, and ADMIRE 3 + B59, respectively. According to the results, an excellent ERD value was obtained in CT images using the ADMIRE 3 + B59 kernel. The average ERD result of CT images obtained using ADMIRE 3 was approximately 1.44, which showed an improvement of approximately 3.55% compared to ADMIRE 2. In addition, the average ERD value of CT images acquired using the B59 kernel showed an improvement of approximately 53.44% compared to the B40 kernel. These results demonstrate that the ERD value, which can be used to comprehensively evaluate spatial resolution, improves as the strength of ADMIRE increases and the kernel size increases.

    Fig. 4 presents a graph of NIQE, a blind quality evaluation, performed on the images obtained for each ADMIRE strength and kernel. The average of NIQE was 8.23, 8.60, 6.32, and 6.04 for ADMIRE 2 + B40, ADMIRE 3 + B40, ADMIRE 2 + B59, and ADMIRE 3 + B59, respectively. The best NIQE result was observed for CT images obtained using the ADMIRE 3 + B59 kernel. The average NIQE result of CT images using ADMIRE 2 was approximately 7.27, which showed an improvement of approximately 0.66% compared to ADMIRE 3. In addition, the average NIQE value of CT images acquired using the B59 kernel showed an improvement of approximately 26.55% compared to the B40 kernel.

    Among the CT image reconstruction methods, IR can be used to obtain images with less noise using a lower dose than the conventional FBP method. According to a preliminary study by Gordic et al., ADMIRE was found to reduce noise by up to 50% compared to FBP[21]. In addition, the researchers demonstrated that the border of the coronary artery can be made clearer by setting the strength of ADMIRE higher. Theoretically, when using a higher ADMIRE strength, the noise characteristics and edge preservation performance of CT images improve. As the size of the kernel used in the ADMIRE reconstruction method increases, the noise level increases and the edge area is clearly expressed. In the present study, we found that the COV and ERD results tended to be consistent with the theory of the basic ADMIRE reconstruction method.

    The noise and spatial resolution of the CT images vary depending on the strength of ADMIRE and the type of kernel used. This highlights the importance of evaluating the quality of the overall image. In addition to measuring the noise level and spatial resolution using the COV and ERD evaluation factors, methods for evaluating overall image similarity are also widely used in the medical imaging field. Representative parameters of similarity measurement that can evaluate overall image quality include the root mean square error and structural similarity index[22,23]. However, in order to apply both parameters in medical imaging including CT, an ideal reference image is required, which is very difficult due to the fact that images acquired using an actual CT device are inevitably deteriorated.

    In previous studies, the image quality of CT images changed according to ADMIRE parameters was analyzed using one evaluation factor [24,25]. Shin et al. used contrast-to-noise, COV, and noise power spectrum (NPS) evaluation factors [24], and Solomon et al. used filtered NPS to evaluate image quality according to changes in CT reconstruction factors [25]. In this study, the acquired CT images were analyzed using NIQE, a factor that can evaluate the overall image quality based on no-reference [20]. NIQE is widely used as a well-validated factor in evaluating the image quality of X-ray-based medical images [26]. Herein, the best NIQE value was obtained when ADMIRE 3 + B59 was used to acquire CT images using the ADMIRE reconstruction method. This result confirms the optimal reconstruction parameters when taking into consideration noise and spatial resolution simultaneously.

    One of the limitations of this study is that only one IR reconstruction method of CT images was used to analyze the noise level and spatial resolution. However, we expect more accurate CT image quality information to be available to users once additional analyses are conducted on IR methods developed and commercialized by manufacturers other than ADMIRE. In addition, we plan to experiment in the application of image-based deep learning technologies using CT images using the IR reconstruction method, the latter of which was found to result in the best image quality. In the field of CT imaging, we expect the diagnostic accuracy of various diseases to be greatly improved if images with an optimal image quality are used when constructing a data set for use by deep learning networks.

    Ⅳ. Conclusion

    This study analyzed image quality in CT images as a function of parameter adjustment in ADMIRE, a representative IR technology for CT image reconstruction. The results confirmed that the noise level and spatial resolution changed when adjusting the strength and kernel size of ADMIRE. Among the experimental conditions analyzed, the best results for NIQE were obtained when using the ADMIRE 3 + B59 reconstruction method. In conclusion, by analyzing the level at which the image quality changes according to the ADMIRE reconstruction method, we found that overall image quality improves as the strength and kernel size both increase.

    Figure

    JRST-47-3-175_F1.gif

    Acquired CT sample images including ROIs for calculating COV and ERD. ROI_A and ROI_B were used to calculate COV and ERD, respectively

    JRST-47-3-175_F2.gif

    (a) Aortic arch, (b) middle of heart, and (c) end-of-heart CT images acquired using various ADMIRE reconstruction methods and kernels

    JRST-47-3-175_F3.gif

    Representative graphs of the results for (a) COV and (b) ERD as function of the ADMIRE reconstruction method and kernel used.

    JRST-47-3-175_F4.gif

    Representative graphs of the results for NIQE as a function of the ADMIRE reconstruction method and kernel used.

    Table

    Reference

    1. Jeong HS, Kim IJ, Park SB, Park S, Oh Y, Lee WS, et al. Optimization of non-local means algorithm in low-dose computed tomographic image based on noise level and similarity evaluations. Journal of Radiological Science and Technology. 2024;47(1):39-48.
    2. Kim JW, Seoung YH. Development of biopsy assist device on computed tomography using 3D printing technology. Journal of Radiological Science and Technology. 2023;46(2):151-7.
    3. Yoon S, Yoo KH, Park SH, Kim H, Lee JH, Park J, et al. Low-dose abdominopelvic computed tomography in patients with lymphoma: An image quality and radiation dose reduction study. PLoS One. 2022;17(8):e0272356.
    4. Kataria B, Althen JN, Smedby O, Persson A, Sokjer H, Sandborg M. Image quality and potential dose reduction using advanced modeled iterative reconstruction (admire) in abdominal CT: A review. Radiation Protection Dosimetry. 2021;195(3-4): 177-87.
    5. Ellmann S, Kammerer F, Allmendinger T, Hammon M, Janka R, Lell M, et al. Advanced modeled iterative reconstruction (ADMIRE) facilitates radiation dose reduction in abdominal CT. Academic Radiology. 2018;25:1277-84.
    6. Richards PJ, George J, Metelko M, Brown M. Spine computed tomography doses and cancer induction. Spine. 2010;35(4):430-3.
    7. Naziroglu RE, van Ravesteijn VF, van Vliet LJ, Streekstra GJ, Vos FM. Simulation of scanner- and patient-specific low-dose CT imaging from existing CT images. Physica Medica. 2017;36:12-23.
    8. Hsieh J, Nett B, Yu Z, Sauer K, Thibault JB, Bouman CA. Recent advances in CT image reconstruction. Current Radiology Reports. 2013;1:39-51.
    9. Padole A, Khawaja RDA, Kalra MK, Singh S. CT radiation dose and iterative reconstruction techniques. AJR. 2015;204:W384-92.
    10. Geyer LL, Schoepf UJ, Meinel FG, Nance JW, Bastarrika G, Leipsic JA, et al. State of the Art: Iterative CT reconstruction techniques. Radiology. 2015;276(2):339-57.
    11. Solomon J, Mileto A, Ramirez-Giraldo JC, Samei E. Diagnostic performance of an advanced modeled iterative reconstruction algorithm for low-contrast detectability with a third-generation dual-source multidetector CT scanner: Potential for radiation dose reduction in a multireader study. Radiology. 2015; 275(3):735-45.
    12. Anja Ö, Georg B, Florian H, Richter H, Ulrike E, Till-Karsten H. Image quality of CT angiography of supra-aortic arteries. Clinical Neuroradiology. 2020;30:101-7.
    13. Gordic S, Desbiolles L, Stolzmann P, Gantner L, Leschka S, Husarik DB, et al. Advanced modelled iterative reconstruction for abdominal CT: Qualitative and quantitative evaluation. Clinical Radiology. 2014;69:e497-504.
    14. Rompel O, Glockler M, Janka R, Dittrich S, Cesnjevar R, Lell MM, et al. Third-generation dual-source 70-kVp chest CT angiography with advanced iterative reconstruction in young children: Image quality and radiation dose reduction. Pediatric Radiology. 2016;46:462-72.
    15. Weiss KL, Cornelius RS, Greeley AL, Sun D, Chang I-YJ, Boyce WO, et al. Hybrid convolution kernel: Optimized CT of the head, neck, and spine. AJR. 2011;196:403-6.
    16. Gierada DS, Bierhals AJ, Choong CK, Bartel ST, Ritter JH, Das NA, et al. Effects of CT section thickness and reconstruction kernel on emphysema quantification: Relationship to the magnitude of the CT emphysema index. Academic Radiology. 2010; 17:146-56.
    17. Gordic S, Morsbach F, Schmidt B, Allmendinger T, Flohr T, Husarik D, et al. Ultralow-dose chest computed tomography for pulmonary nodule detection. Investigative Radiology. 2014;49(7): 465-73.
    18. Denzinger F, Wels M, Breininger K, Taubmann O, Muhlberg A, Allmendinger T, et al. How scan parameter choice affects deep learning‑based coronary artery disease assessment from computed tomography. Scientific Reports. 2023;13.
    19. Tatsugami F, Higaki T, Sakane H, Fukumoto W, Kaichi Y, Iida M, et al. Coronary artery stent evaluation with model-based iterative reconstruction at coronary CT angiography. Academic Radiology. 2017;24:975-81.
    20. Mittal A, Soundararajan R, Bovik AC. Making a “Completely Blind” image quality analyzer. IEEE Signal Processing Letters. 2013;20(3):209-12.
    21. Gordic S, Desbiolles L, Sedlmair M, Manka R, Plass A, Schmidt B, et al. Optimizing radiation dose by using advanced modelled iterative reconstruction in high-pitch coronary CT angiograph. European Radiology. 2016;26(2):459-68.
    22. Hayashi R, Miyazaki K, Takao S, Yokokawa K, Tanaka S, Matsuura T, et al. Real-time CT image generation based on voxel-by- voxel modeling of internal deformation by utilizing the displacement of fiducial markers. Medical Physics. 2021;48: 5311-26.
    23. Joemai RMS, Geleijns J. Assessment of structural similarity in CT using filtered backprojection and iterative reconstruction: A phantom study with 3D printed lung vessels. The British Journal of Radiology. 2017;90:20160519.
    24. Shin JB, Yoon DK, Pak S, Kwon YH, Suh TS. Comparative performance analysis for abdominal phantom ROI detectability according to CT reconstruction algorithm: ADMIRE. Journal of Applied Clinical Medical Physics. 2020;21(1):136-43.
    25. Solomon JB, Christianson O, Samei E. Quantitative comparison of noise texture across CT scanners from different manufacturers. Medical Physics. 2012;39(10) :6048-55.
    26. Lim S, Park M, Kim H, Kang SH, Kim K, Lee Y. Optimization of median modified wiener filter for improving lung segmentation performance in low-dose computed tomography images. Applied Sciences. 2023;13(19):10679.
    Copyright © Korean Society of Radiological Science. All rights reserved.