Abstract
This study aims to investigate the influence of adaptive statistical iterative reconstruction-V (ASIR-V) and deep learning image reconstruction (DLIR) on CT radiomics feature robustness. A standardized phantom was scanned under single-energy CT (SECT) and dual-energy CT (DECT) modes at standard and low (20 and 10 mGy) dose levels. Images of SECT 120 kVp and corresponding DECT 120 kVp-like virtual monochromatic images were generated with filtered back-projection (FBP), ASIR-V at 40% (AV-40) and 100% (AV-100) blending levels, and DLIR algorithm at low (DLIR-L), medium (DLIR-M), and high (DLIR-H) strength levels. Ninety-four features were extracted via Pyradiomics. Reproducibility of features was calculated between standard and low dose levels, between reconstruction algorithms in reference to FBP images, and within scan mode, using intraclass correlation coefficient (ICC) and concordance correlation coefficient (CCC). The average percentage of features with ICC > 0.90 and CCC > 0.90 between the two dose levels was 21.28% and 20.75% in AV-40 images, and 39.90% and 35.11% in AV-100 images, respectively, and increased from 15.43 to 45.22% and from 15.43 to 44.15% with an increasing strength level of DLIR. The average percentage of features with ICC > 0.90 and CCC > 0.90 in reference to FBP images was 26.07% and 25.80% in AV-40 images, and 18.88% and 18.62% in AV-100 images, respectively, and decreased from 27.93 to 17.82% and from 27.66 to 17.29% with an increasing strength level of DLIR. DLIR and ASIR-V algorithms showed low reproducibility in reference to FBP images, while the high-strength DLIR algorithm provides an opportunity for minimizing radiomics variability due to dose reduction.
Similar content being viewed by others
Explore related subjects
Discover the latest articles, news and stories from top researchers in related subjects.Avoid common mistakes on your manuscript.
Introduction
Radiomics converts medical imaging data into high-dimensional minable features for constructing diagnostic, prognostic, or predictive models to aid clinical decision-making [1,2,3,4,5,6]. However, the issue of robustness should be evaluated before applying radiomics as a daily tool in clinical practice [6,7,8,9,10]. It is of interest to better understand the influence of acquisition and reconstruction parameters on radiomics robustness [11,12,13,14,15,16,17,18,19]. It has been shown that single-energy CT (SECT) and dual-energy CT (DECT) scan modes and discrepancy in technique setups among platforms impact the reproducibility of radiomic features [13,14,15,16,17,18,19]. Owing to developments in CT acquisition technique and reconstruction algorithms [20,21,22], low-dose CT has been realized with comparable and even better image quality [23,24,25,26,27,28]. Compared to the traditional filtered back projection (FBP), iterative reconstruction (IR) algorithms and deep learning image reconstruction (DLIR) algorithm have been successively introduced to provide better image quality with lower radiation dose [20,21,22,23,24,25,26,27,28], but their influence on the radiomics features has not been fully investigated.
The application of low-dose CT scan protocols and new reconstruction algorithms becomes a potential source of radiomics variability. It is necessary to find a way to reduce radiomics variability due to the use of scan protocols at different dose levels, and to allow the translation of radiomics models derived at high-dose level to lower ones. On one hand, CT radiomics features are sensitive to diverse reconstruction algorithms and present significant variation when comparing different strength levels [12, 29,30,31]. On the other hand, the difference between images acquired at distinct low and ultra-low dose levels can be bridged by careful adjusting of strength levels of IR algorithms [29,30,31]. Nevertheless, it is unclear how reconstruction algorithms, such as the DLIR algorithm, impact the underlying minable information in images [14]. Although the deep learning reconstruction (DLR) algorithm by Canon Medical Systems showed potential for improving radiomics reproducibility in SECT images [32], the potential of the DLIR algorithm by GE Healthcare for reducing SECT and DECT radiomics variability is still unknown.
Therefore, the aim of this study was to investigate the influence of DLIR on the robustness of radiomics features and to find out whether DLIR provided an opportunity for minimizing CT radiomics variability at different dose levels.
Materials and Methods
Phantom
The workflow of the present study is presented in Fig. 1. The ethics approval was not required because this was a phantom study. A 330-mm diameter Gammex phantom (Gammex Inc.) made of water-equivalent material was scanned. The phantom has sixteen 28-mm diameter holes for holding interchangeable inserts with various densities. To mimic contrast media in blood vessel, five iodine inserts with concentrations from 2.0 to 15.0 mg/mL were selected. To simulate a wide range of densities in the human body, 11 tissue inserts with densities of 0.44 to 1.69 g/cm3 were chosen. The inserts were placed to avoid beam-hardening artifacts, and their positions remained the same across the scans in the study.
Image Acquisition and Reconstruction
All images were acquired on a 256-slice CT scanner with dual-energy CT imaging capability (Revolution Apex CT, GE Healthcare). The acquisition parameters are presented in Table 1. The SECT and DECT scans were performed with conventional 120 kVp and the rapid kVp switching dual-energy imaging technology (80/140 kVp), respectively, at two dose levels (volume CT dose indexes, CTDIVol 20 and 10 mGy). The 20 mGy dose level was selected as the reference level for an adult abdomen based on the clinical practice in our institution [33]. The 10 mGy level was selected to simulate low-dose protocol, following the previous study results indicating that using a high strength level of the DLIR algorithm could potentially reduce half of the radiation dose [23,24,25,26,27,28]. Tube currents and rotation time were modified to obtain these two dose levels. The scan field of view (500 mm × 500 mm), matrix size (512 × 512), and section thickness (5 mm) were kept the same across all scans. All the scans were repeated, several minutes after the first scan after reposition, to allow repeatability analysis.
The conventional 120-kVp images were generated for SECT scans. The virtual monochromatic images (VMI) were reconstructed at 70 keV to create 120-kVp-like images for DECT scans, since the CT numbers of images at the energy level of 70 keV were used as a clinical standard of reference at our institution and were reported to be comparable to those of 120-kVp images [34]. A standard reconstruction kernel was used for all reconstructions. Six reconstruction algorithms were applied: FBP, adaptive statistical iterative reconstruction-V (ASIR-V, GE Healthcare) at 40% (AV-40) and 100% (AV-100) blending levels, and DLIR (TrueFidelity™, GE Healthcare) at low (DLIR-L), medium (DLIR-M), and high (DLIR-H) strength levels. Therefore, 24 sets of images were generated in total.
Segmentation and Feature Extraction
The circular regions of interest (ROIs) were plotted using an open-source ITK-SNAP software version 3.6.0 (http://www.itksnap.org/pmwiki/pmwiki.php). We copied the ROIs from one scan to another with rigid registrations to minimize extra variations due to segmentation. Sixteen 25-mm- (26-pixel) diameter circular ROIs were drawn at the center of each insert, covering each insert as much as possible while avoiding touching its edge. To present the true difference among platforms, we did not employ any image pre-processing steps. The radiomics features were extracted via Python version 3.7.6 (https://www.python.org) with Pyradiomics version 3.0 (https://pyradiomics.readthedocs.io/en/latest/) from each ROI on original images. Since the shape and size of ROIs were fixed, the 26 shape-based features were excluded. Within 94 extracted features, 19 were first-order features, and 75 were texture features. The details of radiomics analysis methods are presented in Supplementary Note S1.
Radiomics Robustness Analysis
The robustness of radiomics features was evaluated in terms of repeatability and reproducibility. Test–retest repeatability of features was assessed by Bland–Altman analysis for repeated scans, with a cutoff value of 90% [35]. To test the hypothesis that the obtained biases of the radiomics feature values between the scan and re-scan were equal to zero, a one-sample t-test was performed. The reproducibility of radiomic features was estimated using intraclass correlation coefficient (ICC) with single rater, absolute agreement, two-way random effects model [36], and concordance correlation coefficient (CCC) [37, 38]. The reproducibility of features was calculated between the standard and low-dose levels for each reconstruction algorithm. The reproducibility of features was also estimated between reconstruction algorithms in reference to the FBP images for each scan mode per dose level, to find out whether the reconstruction algorithm alters information in images. The FBP images were used as reference, as they were considered the original version of the images. We further evaluated the reproducibility across all 24 image sets, to identify potential opportunity for minimizing radiomics variability. The reproducibility of features was considered excellent if ICC or CCC was > 0.9, good if ICC or CCC was > 0.75 and ≤ 0.9, moderate if ICC or CCC was > 0.5 and ≤ 0.75, and poor if ICC or CCC was ≤ 0.5 [39,40,41].
Statistical Analysis
The statistical analysis was performed with R language version 4.1.3 (https://www.r-project.org/) within RStudio version 1.4.1106 (https://www.rstudio.com/) [42]. The proportions of reproducible radiomic features as nominal variables are presented as the percentage and were compared among different reconstruction algorithms using Cochran’s Q test. ICC and CCC values as continuous variables are presented as average value and were compared among different reconstruction algorithms using the Friedman test. A P value less than 0.05 was considered statistically significant. The significance threshold for adjusted P values was set at 0.05, applying the Bonferroni method for post hoc pairwise multiple-comparison correction. The details of data analysis methods are presented in Supplementary Note S2.
Results
Test–Retest Repeatability Analysis
The average percentages of features that met the criteria of repeatability in SECT scans and DECT scans were 91.31% and 95.04% at the 10 mGy dose level, and 90.60% and 96.81% at the 20 mGy dose level, respectively (detailed results are presented in Supplementary Fig. S1 and Supplementary Table S1). The biases of the radiomics feature values between the scan and re-scan were not significantly different from zero (all P > 0.05).
Reproducibility of Radiomic Features Between Dose Levels
The average percentage of features with ICC > 0.90 and CCC > 0.90 was 21.28% and 20.75% for AV-40 images, and 39.90% and 35.11% for AV-100 images, respectively. Detailed results are presented in Fig. 2 and Supplementary Figs. S2 and S3. The improvements for the AV-100 images were mainly identified in the texture features. The average percentage of features with ICC > 0.90 and CCC > 0.90 between images acquired at 10 and 20 mGy dose levels increased with increasing strength level of the DLIR algorithm from 15.43 to 45.22% and from 15.43 to 44.15%, respectively, which was supported by corresponding mean ICC and CCC values. Detailed results are presented in Table 2 and Supplementary Table S2.
Reproducibility of Radiomic Features in Reference to FBP Images
The average percentage of features with ICC > 0.90 and CCC > 0.90 in reference to FBP images was 26.06% and 25.80% for AV-40 images, and 18.88% and 18.62% for AV-100 images, respectively. Detailed results are presented in Fig. 3 and Supplementary Figs. S4 and S5. The average percentage of the feature with ICC > 0.90 and CCC > 0.90 in reference to FBP images decreased with increasing strength level of the DLIR algorithm, from 27.93 to 17.82% and from 27.66 to 17.29%, respectively, which was supported by corresponding mean ICC and CCC values (Table 3 and Supplementary Table S3). In both cases, the reproducibility decreased more obviously within the texture features due to the stronger image noise reduction with the increased reconstruction strengths.
Reproducibility of Radiomics Features Within Scan Mode
The overall reproducibility within scan mode was low (Fig. 4). Within SECT scans at different dose levels, DLIR-H images at 10 mGy and DLIR-M images at 20 mGy showed the highest percentage with ICC > 0.90 and CCC > 0.90 of 79.78% and 76.60%. Within the same reconstruction algorithm, DLIR-H images at 10 and 20 mGy showed the highest percentage of features with ICC > 0.90 and CCC > 0.90 of 50.00% and 47.87%. The percentage of features with ICC > 0.90 and CCC > 0.90 were 46.81% and 40.43% between AV-100 images at 10 and 20 mGy. Within DECT scans at different dose levels, DLIR-H images at 10 and 20 mGy presented the highest percentage of features with ICC > 0.90 and CCC > 0.90 of 40.42% and 40.42%. The percentage of features with ICC > 0.90 and CCC > 0.90 were 32.97% and 29.79% between AV-100 images at 10 and 20 mGy.
Discussion
In this study, we investigated the influence of ASIR-V and DLIR algorithms on the robustness of radiomics features in reference to the traditional FBP reconstruction algorithm and evaluated whether DLIR provided an opportunity for minimizing CT radiomics variability existing at different dose levels. Our study showed that the reproducibility in reference to FBP images was generally low and decreased with increasing strength level of ASIR-V and DLIR algorithms, indicating image reconstruction algorithms potentially altered radiomics features. However, the reproducibility between the standard and low dose levels increased with increasing strength level of ASIR-V and DLIR algorithms, showing the potential for minimizing radiomics variability with higher reconstruction strength for using acquisitions at different dose levels. Overall, the DLIR algorithm showed a higher possibility for reducing variability due to dose reduction than the ASIR-V algorithm. Within SECT scans, DLIR-H images at the dose levels of 10 and 20 mGy were considered the most reproducible when the same reconstruction algorithm was used. Within DECT scans, the DLIR-H algorithm showed the highest reproducibility between images at 10 and 20 mGy. These results provided insights for retrospective data collection and future protocol implementations.
The dose reduction is an important source of nonreproducible features caused by the higher image noise at reduced radiation dose levels [12, 29,30,31]. The higher level of ASIR-V and DLIR algorithms allow greater reduction of image noise [23,24,25,26,27,28] and are therefore expected to provide higher reproducibility of features between images acquired at different dose levels. Our results showed that the highest level of ASIR-V (AV-100) and DLIR (DLIR-H) images both had a better ability to reduce dose-induced radiomics variability within both SECT and DECT scans. We believe that ASIR-V and DLIR algorithms could at least partially harmonize the radiomics variability due to dose reduction protocols. In an era of pursuing lower radiation dose as possible, our findings may have important implications, because they provided insights into the possibility for generalizability of radiomics models derived from scan protocols of different dose levels.
However, the higher strength level of ASIR-V and DLIR algorithms may further alter radiomics features compared with the FBP algorithm. The reproducibility of AV-100 in reference to FBP images was low. The ASIR-V algorithm reduces and regulates image noise using nonlinear operations, which allows improvement of the reproducibility of features for different scans. It comes at the expense of impaired reproducibility in reference to other reconstruction algorithms such as FBP, because the ASIR-V algorithm with a higher strength level further alters the image texture from FBP when it reduces the noise [23]. The reproducibility of DLIR-H images in reference to FBP images was also low. Nevertheless, there is some doubt whether FBP images are informative enough because a significant part of the reproducible results in FBP images is considered due to repetitive noise [32]. The features which are not closely related to noise (e.g., mean) were stable among FBP, ASIR-V, and DLIR images, while those reflecting correlations between pixels (e. g. the majority of texture features) showed a decreasing trend of reproducibility with increasing strength level of ASIR-V and DLIR algorithms. DLIR algorithm uses a deep learning neural network to remove noise and is expected to maintain texture in the FBP images [23,24,25,26,27,28]. We suspected that the significant noise reduction could explain in part the low radiomics reproducibility between FBP and DLIR images. In theory, the DLIR algorithm has more possibility to preserve original informative features than the IR algorithm, because the high strength level of the DLIR algorithm did not significantly change image texture [32] and was more acceptable for clinical diagnosis than a high strength level of IR algorithm [28, 43,44,45,46,47,48]. Meanwhile, DLIR-H images did show higher reproducibility of features between images acquired at two dose levels than that of AV-100 images. Therefore, further study is recommended to investigate whether the altered radiomics features due to the high strength level of the DLIR algorithm have an impact on the discriminative power of these features.
Comparison of the DLR algorithm (Canon Medical System) with FBP and IR algorithms has shown the advantage of the DLR algorithm for improving the yield of stable and reliable radiomics features in SECT images [32]. However, DLR and DLIR trained their models with different gold standards: DLR uses model-based image reconstruction images, while DLIR uses the high-dose FBP images. Therefore, they have different behaviors in noise reduction [48]. To maximize the data usage, especially retrospectively, in clinical applications, it is of interest to explain the varying reproducibility of images acquired at different dose levels and reconstruction algorithms. Our study applied the DLIR algorithm (GE Healthcare) to confirm the potential of deep learning for reproducible CT radiomics in both SECT and DECT images and further demonstrated that the increasing strength level of the DLIR algorithm allowed higher reproducibility for CT scans of different dose levels. Our study revealed opportunities with the DLIR algorithm in retrospective data collection and future protocol implementations for radiomics [49]. The current work differed from previous studies that applied the deep learning method as an image conversion filter to improve CT radiomics reproducibility [50, 51], but underlined that image reconstruction with the deep learning method has a high potential to improve radiomics research.
Several limitations of our study should be acknowledged. First, our study was a phantom study. Therefore, the results of our study should be carefully interpreted as hypothesis generating. The generalizability of our results to tumors or diseases in clinical application may be limited, partially due to the homogeneity of our inserts [52]. However, we consider our findings to give an important insight into the performance of different reconstruction algorithms and whether the DLIR algorithm could reduce variability in radiomics features from clinical examinations. Second, we only assessed the reproducibility between standard dose and half-dose protocols. Our findings may not directly guide algorithm selection in clinical when the degree of radiation reduction varies, especially when ultra-low dose protocols are used, but we believe our study demonstrated the possibility for the DLIR algorithm to improve radiomics reproducibility even with a greater dose reduction. Third, we did not investigate the influence of reconstruction algorithms on the discriminative power of radiomics features. Because DLIR and IR algorithms may alter radiomics features in reference to FBP images, further studies are required to evaluate their impact on the reproducibility of radiomics features as diagnostic, prognostic, or predictive biomarkers [32]. Finally, our study was performed with the only available standard kernel in one CT system. Other manufacturers provide different deep learning-based algorithms with distinctly different reconstruction kernels for clinical use and may have different effects on radiomics reproducibility from those obtained in the current study [53].
To summarize, increasing the strength level of ASIR-V and DLIR algorithms improved the reproducibility of features between standard and low dose levels but decreased the reproducibility of features in reference to FBP images. DLIR algorithm may be applied for minimizing radiomics variability when combining images from protocols with different radiation doses is desired.
Availability of Data and Material
All data generated or analyzed during this study are included in this published article and its supplementary information files.
Abbreviations
- ASIR-V:
-
Adaptive statistical iterative reconstruction-V
- CCC:
-
Concordance correlation coefficient
- DECT:
-
Dual-energy computed tomography
- DLIR:
-
Deep learning image reconstruction
- DLR:
-
Deep learning reconstruction
- FBP:
-
Filtered back-projection
- ICC:
-
Intraclass correlation coefficient
- IR:
-
Iterative reconstruction
- ROI:
-
Region of interest
- SECT:
-
Single-energy computed tomography
- VMI:
-
Virtual monoenergetic image
References
Lambin P, Rios-Velazquez E, Leijenaar R et al (2012) Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer 48(4):441-446. https://doi.org/10.1016/j.ejca.2011.11.036
Gillies RJ, Kinahan PE, Hricak H (2016) Radiomics: images are more than pictures, they are data. Radiology 278(2):563-577. https://doi.org/10.1148/radiol.2015151169
O'Connor JP, Aboagye EO, Adams JE et al (2017) Imaging biomarker roadmap for cancer studies. Nat Rev Clin Oncol 14:169-186. https://doi.org/10.1038/nrclinonc.2016.162
Lambin P, Leijenaar RTH, Deist TM et al (2017) Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol 14(12):749-762. https://doi.org/10.1038/nrclinonc.2017.141
Huang EP, O'Connor JPB, McShane LM et al (2023) Criteria for the translation of radiomics into clinically useful tests. Nat Rev Clin Oncol. 2023 Feb;20(2):69-82. https://doi.org/10.1038/s41571-022-00707-0
van Timmeren JE, Cester D, Tanadini-Lang S, Alkadhi H, Baessler B (2020) Radiomics in medical imaging-“how-to” guide and critical reflection. Insights Imaging 11(1):91. https://doi.org/10.1186/s13244-020-00887-2
Park JE, Park SY, Kim HJ, Kim HS (2019) Reproducibility and generalizability in radiomics modeling: possible strategies in radiologic and statistical perspectives. Korean J Radiol 20(7):1124-1137. https://doi.org/10.3348/kjr.2018.0070
Zwanenburg A (2019) Radiomics in nuclear medicine: robustness, reproducibility, standardization, and how to avoid data analysis traps and replication crisis. Eur J Nucl Med Mol Imaging 46(13):2638-2655. https://doi.org/10.1007/s00259-019-04391-8
Cattell R, Chen S, Huang C (2019) Robustness of radiomic features in magnetic resonance imaging: review and a phantom study. Vis Comput Ind Biomed Art 2(1):19. https://doi.org/10.1186/s42492-019-0025-6
Pfaehler E, Zhovannik I, Wei L et al (2021) A systematic review and quality of reporting checklist for repeatability and reproducibility of radiomic features. Phys Imaging Radiat Oncol 20:69-75. https://doi.org/10.1016/j.phro.2021.10.007
Berenguer R, Pastor-Juan MDR, Canales-Vázquez J et al (2018) Radiomics of CT features may be nonreproducible and redundant: influence of CT acquisition parameters. Radiology 288(2):407–415. https://doi.org/10.1148/radiol.2018172361
Meyer M, Ronald J, Vernuccio F et al (2019) Reproducibility of CT radiomic features within the same patient: influence of radiation dose and CT reconstruction settings. Radiology 293(3):583-591. https://doi.org/10.1148/radiol.2019190928
Chen Y, Zhong J, Wang L et al (2022) Robustness of CT radiomics features: consistency within and between single-energy CT and dual-energy CT. Eur Radiol 32(8):5480-5490. https://doi.org/10.1007/s00330-022-08628-3
Zhong J, Xia Y, Chen Y et al (2023) Deep learning image reconstruction algorithm reduces image noise while alters radiomics features in dual-energy CT in comparison with conventional iterative reconstruction algorithms: a phantom study. Eur Radiol 33(2):812-824. https://doi.org/10.1007/s00330-022-09119-1
Zhong J, Pan Z, Chen Y et al (2023) Robustness of radiomics features of virtual unenhanced and virtual monoenergetic images in dual-energy CT among different imaging platforms and potential role of CT number variability. Insights Imaging 14(1):79. https://doi.org/10.1186/s13244-023-01426-5
Lennartz S, O'Shea A, Parakh A, Persigehl T, Baessler B, Kambadakone A (2022) Robustness of dual-energy CT-derived radiomic features across three different scanner types. Eur Radiol 32(3):1959-1970. https://doi.org/10.1007/s00330-021-08249-2
Peng X, Yang S, Zhou L et al (2022) Repeatability and reproducibility of computed tomography radiomics for pulmonary nodules: a multicenter phantom study. Invest Radiol 57(4):242-253. https://doi.org/10.1097/RLI.0000000000000834
Baliyan V, Kordbacheh H, Parameswaran B, Ganeshan B, Sahani D, Kambadakone A (2018) Virtual monoenergetic imaging in rapid kVp-switching dual-energy CT (DECT) of the abdomen: impact on CT texture analysis. Abdom Radiol (NY) 43(10):2693- 2701. https://doi.org/10.1007/s00261-018-1527-y
Euler A, Laqua FC, Cester D et al (2021) Virtual monoenergetic images of dual-energy ct-impact on repeatability, reproducibility, and classification in radiomics. Cancers (Basel) 13(18):4710. https://doi.org/10.3390/cancers13184710
Geyer LL, Schoepf UJ, Meinel FG et al (2015) State of the art: iterative CT reconstruction techniques. Radiology 276(2):339-357. https://doi.org/10.1148/radiol.2015132766
Willemink MJ, Noël PB (2019) The evolution of image reconstruction for CT-from filtered back projection to artificial intelligence. Eur Radiol 29(5):2185-2195. https://doi.org/10.1007/s00330-018-5810-7
Singh R, Wu W, Wang G, Kalra MK (2020) Artificial intelligence in image reconstruction: the change is here. Phys Med 79:113-125. https://doi.org/10.1016/j.ejmp.2020.11.012
Greffier J, Frandon J, Larbi A, Beregi JP, Pereira F (2020) CT iterative reconstruction algorithms: a task-based image quality assessment. Eur Radiol 30(1):487-500. https://doi.org/10.1007/s00330-019-06359-6
Greffier J, Hamard A, Pereira F et al (2020) Image quality and dose reduction opportunity of deep learning image reconstruction algorithm for CT: a phantom study. Eur Radiol 30(7):3951-3959. https://doi.org/10.1007/s00330-020-06724-w
Greffier J, Si-Mohamed S, Guiu B et al (2022) Comparison of virtual monoenergetic imaging between a rapid kilovoltage switching dual-energy computed tomography with deep-learning and four dual-energy CTs with iterative reconstruction. Quant Imaging Med Surg 12(2):1149–1162. https://doi.org/10.21037/qims-21-708
Masuda S, Yamada Y, Minamishima K, Owaki Y, Yamazaki A, Jinzaki M (2022) Impact of noise reduction on radiation dose reduction potential of virtual monochromatic spectral images: comparison of phantom images with conventional 120 kVp images using deep learning image reconstruction and hybrid iterative reconstruction. Eur J Radiol 149:110198. https://doi.org/10.1016/j.ejrad.2022.110198
Zhong J, Shen H, Chen Y et al (2023) Evaluation of image quality and detectability of deep learning image reconstruction (DLIR) algorithm in single- and dual-energy CT. J Digit Imaging 36:1390-1407. https://doi.org/10.1007/s10278-023-00806-z
Zhong J, Wang L, Shen H et al (2023) Improving lesion conspicuity in abdominal dual-energy CT with deep learning image reconstruction: a prospective study with five readers. Eur Radiol 33(8):5331-5343. https://doi.org/10.1007/s00330-023-09556-6
Varghese BA, Hwang D, Cen SY et al (2019) Reliability of CT-based texture features: Phantom study. J Appl Clin Med Phys 20(8):155-163. https://doi.org/10.1002/acm2.12666
Kolossváry M, Szilveszter B, Karády J, Drobni ZD, Merkely B, Maurovich-Horvat P (2019) Effect of image reconstruction algorithms on volumetric and radiomic parameters of coronary plaques. J Cardiovasc Comput Tomogr 13(6):325-330. https://doi.org/10.1016/j.jcct.2018.11.004
Ye K, Chen M, Zhu Q, Lu Y, Yuan H (2021) Effect of adaptive statistical iterative reconstruction-V (ASIR-V) levels on ultra-low-dose CT radiomics quantification in pulmonary nodules. Quant Imaging Med Surg 11(6):2344–2353. https://doi.org/10.21037/qims-20-932
Michallek F, Genske U, Niehues SM, Hamm B, Jahnke P (2022) Deep learning reconstruction improves radiomics feature stability and discriminative power in abdominal CT imaging: a phantom study. Eur Radiol 32(7):4587-4595. https://doi.org/10.1007/s00330-022-08592-y
The National Health Commission of People’s Republic of China (2018) Diagnostic reference levels for adults in X-ray computed tomography. Accessed via http://www.nhc.gov.cn/wjw/pcrb/201810/d3bb2f7acef248f0a1347a2da93cb41f.shtml on Dec 2022
Matsumoto K, Jinzaki M, Tanami Y, Ueno A, Yamada M, Kuribayashi S (2011) Virtual monochromatic spectral imaging with fast kilovoltage switching: improved image quality as compared with that obtained with conventional 120-kVp CT. Radiology 259(1):257-262. https://doi.org/10.1148/radiol.11100978
Bland JM, Altman DG (1999) Measuring agreement in method comparison studies. Stat Methods Med Res 8(2):135–160. https://doi.org/10.1177/096228029900800204
Koo TK, Li MY (2016) A guideline of selecting and reporting intraclass correlation coefficients for reliability research. J Chiropr Med 15(2):155–163. https://doi.org/10.1016/j.jcm.2016.02.012
Lin LI (1989) A concordance correlation coefficient to evaluate reproducibility. Biometrics 45(1):255–268.
Lin LI (2000) A note on the concordance correlation coefficient. Biometrics 56(1):324–325.
Eck B, Chirra PV, Muchhala A et al (2021) Prospective evaluation of repeatability and robustness of radiomic descriptors in healthy brain tissue regions in vivo across systematic variations in T2-weighted magnetic resonance imaging acquisition parameters. J Magn Reson Imaging 54(3):1009-1021. https://doi.org/10.1002/jmri.27635
Mitchell-Hay RN, Ahearn TS, Murray AD, Waiter GD (2022) Investigation of the inter- and intrascanner reproducibility and repeatability of radiomics features in T1-weighted brain MRI. J Magn Reson Imaging 56(5):1559–1568. https://doi.org/10.1002/jmri.28191
Adelsmayr G, Janisch M, Kaufmann-Bühler AK et al (2023) CT texture analysis reliability in pulmonary lesions: the influence of 3D vs. 2D lesion segmentation and volume definition by a Hounsfield-unit threshold. Eur Radiol 33(5):3064–3071. https://doi.org/10.1007/s00330-023-09500-8
Mangiafico SS (2016) Summary and analysis of extension program evaluation in R, version 1.19.10, revised 2016. Accessed via http://rcompanion.org/handbook/ on Dec 2022
Park J, Shin J, Min IK et al (2022) Image quality and lesion detectability of lower-dose abdominopelvic CT obtained using deep learning image reconstruction. Korean J Radiol 23(4):402-412. https://doi.org/10.3348/kjr.2021.0683
Noda Y, Kawai N, Nagata S et al (2022) Deep learning image reconstruction algorithm for pancreatic protocol dual-energy computed tomography: image quality and quantification of iodine concentration. Eur Radiol 32(1):384-394. https://doi.org/10.1007/s00330-021-08121-3
Park HJ, Choi SY, Lee JE et al (2022) Deep learning image reconstruction algorithm for abdominal multidetector CT at different tube voltages: assessment of image quality and radiation dose in a phantom study. Eur Radiol 32(6):3974-3984. https://doi.org/10.1007/s00330-021-08459-8
Xu JJ, Lönn L, Budtz-Jørgensen E, Hansen KL, Ulriksen PS (2022) Quantitative and qualitative assessments of deep learning image reconstruction in low-keV virtual monoenergetic dual-energy CT. Eur Radiol 32(10):7098-7107. https://doi.org/10.1007/s00330-022-09018-5
Sato M, Ichikawa Y, Domae K et al (2022) Deep learning image reconstruction for improving image quality of contrast-enhanced dual-energy CT in abdomen. Eur Radiol 32(8):5499-5507. https://doi.org/10.1007/s00330-022-08647-0
Xu JJ, Lönn L, Budtz-Jørgensen E, Jawad S, Ulriksen PS, Hansen KL (2023) Evaluation of thin-slice abdominal DECT using deep-learning image reconstruction in 74 keV virtual monoenergetic images: an image quality comparison. Abdom Radiol (NY) 48(4):1536-1544. https://doi.org/10.1007/s00261-023-03845-w
Espinasse M, Pitre-Champagnat S, Charmettant B et al (2020) CT Texture analysis challenges: influence of acquisition and reconstruction parameters: a comprehensive review. Diagnostics (Basel) 10(5):258. https://doi.org/10.3390/diagnostics10050258
Lee SB, Cho YJ, Hong Y et al (2022) Deep learning-based image conversion improves the reproducibility of computed tomography radiomics features: a phantom study. Invest Radiol 57(5):308-317. https://doi.org/10.1097/RLI.0000000000000839
Choe J, Lee SM, Do KH et al (2019) Deep Learning-based image conversion of CT reconstruction kernels improves radiomics reproducibility for pulmonary nodules or masses. Radiology 292:365– 373. https://doi.org/10.1148/radiol.2019181960
Li Y, Reyhan M, Zhang Y et al (2022) The impact of phantom design and material-dependence on repeatability and reproducibility of CT-based radiomics features. Med Phys 49(3):1648-1659. https://doi.org/10.1002/mp.15491
Kawashima H, Ichikawa K, Takata T, Seto I (2022) Comparative assessment of noise properties for two deep learning CT image reconstruction techniques and filtered back projection. Med Phys 49(10):6359-6367. https://doi.org/10.1002/mp.15918
Acknowledgements
The authors would like to express their gratitude to Dr. Guangyao Wu for his suggestions on data interpretation, Prof. Baisong Wang for his suggestions on statistical analysis, and Dr. Shiqi Mao for his comments on data visualization.
Funding
This study has received funding from the National Natural Science Foundation of China (82302183, 82271934), the Yangfan Project of Science and Technology Commission of Shanghai Municipality (22YF1442400), the Medicine and Engineering Combination Project of Shanghai Jiao Tong University (YG2021QN08), the Research Fund of Tongren Hospital, Shanghai Jiao Tong University School of Medicine (TRKYRC-XX202204), and the Guangci Innovative Technology Launch Plan of Ruijin Hospital, Shanghai Jiao Tong University School of Medicine (2022–13). They played no role in the study design, data collection or analysis, decision to publish, or manuscript preparation.
Author information
Authors and Affiliations
Contributions
All the authors contributed to the study concepts and design. JYZ, ZYW, and LYW contributed to the literature research. JYZ, ZYW, LYW, YC, YHX, LW, and JXF contributed to the experimental studies and data analysis. JYZ, ZYW, and LYW contributed to the statistical analysis. JYZ prepared the original version of the manuscript preparation. All authors read and approved the final version of the manuscript. WWY is the guarantor of the integrity of the entire study.
Corresponding authors
Ethics declarations
Ethics Approval
Institutional Review Board approval was not required because of the nature of our study, which was a phantom study.
Consent to Participate
Written informed consent was not required for this study because of the nature of our study, which was a phantom study.
Consent for Publication
Consent to publish was not required because of the nature of our study, which was a phantom study.
Competing Interests
Mr. Wei Lu and Dr. Jianying Li are employees of GE Healthcare. However, they neither had access nor control over the phantom data acquisition and analysis. All other authors of this manuscript have no relevant financial or non-financial interests to disclose.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Below is the link to the electronic supplementary material.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Zhong, J., Wu, Z., Wang, L. et al. Impacts of Adaptive Statistical Iterative Reconstruction-V and Deep Learning Image Reconstruction Algorithms on Robustness of CT Radiomics Features: Opportunity for Minimizing Radiomics Variability Among Scans of Different Dose Levels. J Digit Imaging. Inform. med. 37, 123–133 (2024). https://doi.org/10.1007/s10278-023-00901-1
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10278-023-00901-1