Study design and patient inclusion
This retrospective study was approved by the local institutional review board and was conducted in accordance with the Declaration of Helsinki. Patients who underwent non-contrast MDCT imaging of the spine at our department according to clinical indication (suspected degenerative spine disease or follow-up in degenerative spine disease) during a period of 1 month (June/July 2019) were identified in our hospital’s picture archiving and communication system (PACS). Exclusion criteria were (1) age below 18 years, (2) motion artifacts in imaging data, (3) previous surgery with instrumentation at the spine, (4) presence of any implants in the field of view (FOV), and (5) vertebral fractures, malignant bone lesions, or spondylodiscitic lesions captured by the FOV. Overall, 26 patients were eligible and included in this study.
Imaging by multi-detector computed tomography
Image acquisition was performed in supine position using a 128-slice MDCT scanner (Ingenuity Core 128, Philips Healthcare). An initial scout scan was used for planning of the FOV, and subsequent helical scanning was acquired with implicit tube current modulation. Table 1 shows scanning details for MDCT imaging.
Simulations and image reconstruction
Tube current reduction
Initial preprocessing of imaging data used a total-variation method for the projection data to reduce image noise (λ = 0.01, n = 50) [24, 25]. By the use of a simulation algorithm based on raw imaging data, we generated MDCT scans with virtually lowered tube currents in a stepwise fashion [26,27,28,29,30,31]. The approach for simulations of LD MDCT has been validated previously . Hence, simulations were generated as if MDCT was conducted with 50% (D50P100), 10% (D10P100), 5% (D5P100), and 3% (D3P100) of the original X-ray tube current. The original imaging data was defined as D100P100.
Sparse sampling was simulated by reading only a reduced amount of projection angles and by deleting the remaining projections in the sinogram [27,28,29, 32]. The original imaging was defined as D100P100, and virtual sparse-sampled imaging was generated as if MDCT was performed with only 50% (D100P50), 10% (D100P10), 5% (D100P5), and 3% (D100P3) of the original projection data.
Statistical iterative reconstruction
For image reconstruction of simulated MDCT with lowered tube current or sparse sampling, we used the same in-house developed SIR algorithm that was based on ordered-subset separable paraboloidal surrogate combining a momentum accelerating approach [33, 34]. A Gaussian noise model was applied and the likelihood term for SIR was computed with log-converted projection data. To enhance convergence and to further depress image noise while achieving adequate bone/soft tissue contrast, a regularization term based on a Huber penalty was applied. The distinct strength of the regularization term was selected in consensus with three board-certified radiologists. Linear attenuation coefficients of resulting imaging data were translated to Hounsfield units by using air and water information from the MDCT calibration data.
Qualitative image analysis
Qualitative image evaluation was performed using the PACS viewer (IDS7, Sectra AB). Two radiologists (reader 1 [R1] and reader 2 [R2], 7 years of experience in radiology each) systematically assessed all reconstructed imaging data in all patients (D100P100, D50P100, D10P100, D5P100, D3P100, D100P50, D100P10, D100P5, and D100P3). Evaluations were performed after patient pseudonymization, and the readers had no access to the clinical reports for original imaging and were unaware of the distinct clinical indication that resulted in MDCT imaging. The readers evaluated the SD scan (D100P100) in consensus reading. All other imaging data were assessed separately, with the readers being strictly blinded to the ratings of each other. In detail, the readers performed evaluations of LD scans in the context of eight reading rounds, with an interval of at least 1 week between each round. Within each round, one reconstructed dataset of each patient was evaluated, with the distinct dataset shown per round being subject to randomization. Furthermore, the order of patient cases was also randomized per reading round.
Overall image quality, overall artifacts, and image contrast were evaluated first based on 5-point Likert scales considering the entire FOV (Table 2). Additionally, the readers performed segment-wise evaluation of degenerative changes to determine the presence or absence of such changes (dichotomous evaluation). Furthermore, in case of detected degenerative changes, the readers had to specify them per segment considering spondylosis, pseudospondylolisthesis, spondylolisthesis (with spondylolysis), non-calcified disc herniation, and disc herniation with calcification. In case of presence of more than one of the mentioned degenerative changes, the readers were requested to provide all segment-specific degenerative changes.
Statistical data analysis
GraphPad Prism (version 6.0; GraphPad Software Inc.) and SPSS (version 25.0; IBM SPSS Statistics for Windows, IBM Corp.) were used for statistical data analyses. The level of statistical significance was set at p < 0.05.
For patient details, scanning parameters and dose characteristics, and scores assigned by the readers, descriptive statistics were calculated. Furthermore, the number of segments with reported degenerative changes and the absolute frequency of each specific degenerative change was counted. The number of any missed degenerative changes when compared with consensus reading of the SD scans was noted. Analyses were performed separately for the evaluations of R1 and R2 and for all reconstructed image data, respectively.
To compare overall image quality, overall artifacts, and image contrast of MDCT with virtually lowered tube current or sparse sampling against SD scanning, Wilcoxon matched-pairs signed-rank tests were performed (D100P100 vs. D50P100/D10P100/D5P100/D3P100 and D100P50/D100P10/D100P5/D100P3 for R1 and R2, respectively). Moreover, Wilcoxon matched-pairs signed-rank tests were also conducted between MDCT with virtually lowered tube current or sparse sampling at each level of reduction (D50P100 vs. D100P50, D10P100 vs. D100P10, D5P100 vs. D100P5, and D3P100 vs. D100P3 for R1 and R2, respectively). Intraclass correlation coefficients (ICCs) were computed to assess inter-reader agreement (two-way mixed model).