This retrospective study was approved by the local institutional review board (registration number: 62/18S) and was conducted in accordance with the Declaration of Helsinki. Overall, 35 patients were included (80.0% females, mean age: 70.6 ± 14.2 years, age range: 26.2–93.1 years), with 23 patients (65.7%) showing at least one vertebral fracture (fracture group) and 12 patients (34.3%) showing no vertebral fracture (control group). Eligible patient cases were identified in the institutional digital Picture Archiving and Communication System (PACS) between November 2016 and November 2017.
Inclusion criteria were (1) availability of routine MDCT of the spine (irrespective of the initial suspected diagnosis or distinct clinical symptoms leading to MDCT), (2) additional spinal magnetic resonance imaging (MRI) performed prior or subsequent to MDCT (including short-tau inversion recovery sequences; only for the fracture group), and (3) diagnosis of at least one vertebral fracture according to MDCT (only for the fracture group). Exclusion criteria for both the fracture and control group were (1) age below 18 years, (2) movement artifacts in imaging data, (3) malignant bone lesions (e.g., bone metastases), (4) any history of metabolic bone disorders aside from osteoporosis, and (5) any implants captured by the field of view (FOV).
Multi-detector computed tomography
All scans were performed with a 64-channel MDCT scanner (Brilliance 64; Philips Healthcare). Parameters of the scanning protocol are shown in Table 1. X-ray tube current was modulated implicitly by the scanner during the helical scan according to the examined body part and estimated body size as derived from the scout scan. All examinations were performed without administration of a contrast agent.
Virtual tube current reduction and sparse sampling
Based on raw projection data, we used a simulation algorithm to generate lower tube currents for MDCT scans [16, 21,22,23]. System parameters of the scanner were considered and electronic noise was calibrated for each pixel at the detector. Simulations were generated as if the scans were made at 50% (D50 P100), 25% (D25 P100), and 10% (D10 P100) of the original x-ray tube current and used for image evaluation, in addition to the original imaging data defined as D100 P100. Furthermore, sparse sampling was applied at levels of 50% (D100 P50), 25% (D100 P25), and 10% (D100 P10) of the original projection data, which was achieved by reading every second, fourth, and tenth projection angle and deleting the remaining projections in the sinogram [16, 23, 24]. While the projection number per full rotation was lowered, other parameters, including patient location and projection geometry, were not changed. All images were reconstructed using FBP and a standard Ram-Lak filter [25, 26]. Table 1 provides an overview of image reconstruction parameters.
Two board-certified radiologists (6 and 8 years of experience in radiology) evaluated all imaging data (35 patients × 7 imaging datasets per patient = 245 datasets for evaluation for reader 1 [R1] and reader 2 [R2], respectively), which were uploaded and stored in IntelliSpace Portal (version 9.0; Philips Healthcare) for visualization and evaluation.
First, both readers evaluated the original images with 100% tube current and projections (D100 P100) as the clinical standard in consensus with the MRI available to confirm the diagnosis of an acute or old vertebral fracture. Then, both readers independently evaluated the remaining datasets in random order (D50 P100, D25 P100, D10 P100, D100 P50, D100 P25, and D100 P10), assessing images derived from the same tube current or number of projections within 1 day for all patients and sticking to an interval of at least 7 days before continuing with images of another tube current or number of projections. The order of patient cases was also randomized for each tube current and number of projections, with the readers being blinded to all clinical patient data, the evaluations of each other, the assignments of patients to the fracture or control group, and all previous evaluations performed.
During evaluations, the number of vertebral fractures per patient had to be determined first, with the vertebrae included in the FOV being provided for each case to allow assignment of single fractures to specific vertebrae. Then, the items and scores presented in Table 2 were considered.
For statistical analyses and generation of graphs, SPSS (version 20.0; IBM SPSS Statistics for Windows, IBM Corp.) was used. A p value < 0.05 was considered statistically significant.
Descriptive statistics were calculated for patient demographics and all items of evaluation. Wilcoxon signed-rank tests were performed to compare overall image quality, overall artifacts, contrast of vertebrae, and diagnostic confidence between MDCT with virtually lowered tube current and sparse-sampled images, i.e., D50 P100 vs. D100 P50, D25 P100 vs. D100 P25, and D10 P100 vs. D100 P10, respectively. Furthermore, overall image quality, overall artifacts, and contrast of vertebrae of D100 P100 as the gold standard were compared with MDCT with virtually lowered tube current and sparse-sampled images using Wilcoxon signed-rank tests, which were achieved separately for each reader.
Interreader intraclass correlation coefficients (ICCs) were calculated for overall image quality, overall artifacts, contrast of vertebrae, and diagnostic confidence in MDCT with virtually lowered tube current and sparse sampling, respectively [27, 28]. As a measure of agreement between imaging with virtual lowering of tube current and sparse sampling, Cohen’s kappa coefficients were determined for age of fracture. Moreover, weighted Cohen's kappa coefficients were determined between the results of both readers for age of fracture [29,30,31].