Multi-detector CT imaging: impact of virtual tube current reduction and sparse sampling on detection of vertebral fractures

Vertebral fractures are frequent in clinical routine and are primarily observed in the context of injuries or as major manifestations of osteoporosis, even in the absence of any obvious trauma [1‐3]. Spine radiography is commonly applied for the detection of suspected vertebral fractures; however, it has been shown that computed tomography (CT) is superior by reducing the risk of missing a fracture, thus resulting in a higher sensitivity and specificity with fracture detection rates of 97 to 100% at the spine [4‐6].

The increased use of CT instead of radiography for the purpose of improved diagnostics comes at the cost of elevated radiation exposure for the patient: one-time scanning with a modern CT scanner applies an estimated effective dose of 5.6 mSv and 10.0 mSv for the lumbar and whole dorsal spine, respectively [7, 8]. The use of CT entails an estimated cancer risk ratio that is multifold higher than in radiography, and it can further increase due to cumulative effects when additional imaging is performed [7‐9]. Thus, CT with reduced radiation exposure, but without simultaneous constraints for image quality or diagnostic accuracy seems crucial.

Despite its clinical relevance, previous research on dose reductions in CT at the spine is generally scarce. In vivo, radiation exposure reductions have been achieved by lowered tube current or voltage at the level of the cervical spine, resulting in largely preserved image quality except for the lower cervical spine [10, 11]. Recently, iterative reconstruction (IR) algorithms have been applied together with low-dose CT, but led to worse image quality for soft tissue and cervical vertebrae when compared to standard-dose CT using filtered back projection (FBP) [12]. To date, only few studies investigated CT with reduced doses specifically for diagnostics of vertebral fractures, showing that low-dose CT with IR may maintain a high diagnostic performance compared to standard-dose CT with IR in trauma patients [13, 14].

In addition to lowering tube current or voltage to reduce x-ray exposure during CT, the number of acquired projections can be decreased with sparse-sampled acquisition schemes. Reducing projection views is a promising strategy since lowering the number of projections can clearly reduce the radiation dose, with previous research indicating a high potential of this approach resulting in reasonable image quality [15‐20]. However, sparse sampling has not been applied at the spine for fracture diagnostics yet.

Against this background, the aim of this study is to evaluate the effects of virtual tube current reduction and sparse sampling on image quality and vertebral fracture diagnostics in multi-detector CT (MDCT). Our hypothesis is that MDCT with sparse sampling would provide better image and diagnostic quality when compared to MDCT with virtual lowering of tube current and, thus, might allow for more drastic reductions in radiation exposure.

Virtual lowering of tube current and sparse sampling were successfully achieved in all patients (Figs. 1 and 2). A median of eight vertebrae (range 4–19 vertebrae) was captured by the FOV of MDCT scans, which covered the cervical spine in 20.0%, the cervico-thoracic spine in 8.6%, the thoracic spine in 5.7%, the thoraco-lumbar spine in 28.6%, and the lumbar spine in 37.1%. The average volumetric CT dose index recorded in the dose reports was 11.7 ± 5.7 mGy for original MDCT scans (Table 1), and was amounted 5.9 mGy, 2.9 mGy, and 1.2 mGy for MDCT with virtually lowered tube current or sparse sampling at 50%, 25%, and 10% of original data, respectively.

Both readers correctly identified all patients of the control group (34.3% of included patients) without any assignments of vertebral fractures to controls in MDCT with virtually lowered tube current or sparse sampling. Among patients of the fracture group (65.7% of included patients), a total of 48 vertebral fractures was observed in the original MDCT with 100% tube current and projections (D100 P100). Patients of the fracture group showed a median of two vertebral fractures (range 1–6 vertebral fractures). These fractures affected the cervical spine in 10.0%, the thoracic spine in 40.0%, and the lumbar spine in 50.0%. Based on original MDCT and MRI scanning, vertebral fractures were diagnosed as acute in 58.3% and old in 41.7%.

This study investigated the effects of virtual tube current reduction and sparse sampling on image quality and diagnostic accuracy of vertebral fractures in MDCT. When comparing virtual tube current reductions to sparse sampling, superior results for image quality and fracture diagnostics were evident for sparse-sampled MDCT. Specifically, no missed vertebral fractures occurred for MDCT with a reduction of 50% in projection numbers, and determination of fracture age was still reliably possible in MDCT with a reduction in projection numbers of 75%.

CT is increasingly applied for first-line diagnostics of vertebral fractures due to its high sensitivity and specificity and excellent fracture detection rates [4‐6]. However, clearly higher estimated effective doses of 5.6 mSv for the lumbar and 10.0 mSv for the whole dorsal spine in CT compared with radiography result in a considerably increased risk of developing cancer later in life [7‐9]. Importantly, cancer risks are summative, and radiography or CT performed for initial diagnostics are not the only sources of radiation exposure, with a patient suffering from an acute traumatic vertebral fracture being exposed to a cumulative effective dose of about 38 mSv only during inpatient stay and without taking into account later follow-up imaging [32]. Consequently, reduction of CT-related radiation exposure is necessary, but should ideally be achieved without loss of image quality or diagnostic accuracy. Despite evident clinical relevance, only a limited body of literature distinctly evaluated approaches for radiation exposure reductions in CT of the spine.

Concerning CT with reduced doses, a small increase in image noise and no difference in subjective image quality evaluation was reported for cervical structures, allowing dose reductions of 61–71% [11]. Low-kV CT with reduced radiation doses by approximately 34% demonstrated good image quality for structures of the neck, but compromised image quality for the lower cervical spine [10]. In patients with lumbar disc herniation, simulated low-dose CT with reductions in tube charge settings to 65% of the standard dose were considered adequate for diagnostic purposes previously [33]. Furthermore, IR algorithms have been introduced for CT with reduced doses, leading to better image quality for intervertebral discs, neural foramina, and ligaments, but worse image quality for vertebrae when compared with standard-dose CT using FBP [12]. Ultra-low-dose CT may still provide an acceptable image quality and exhibited a diagnostic accuracy similar to that of low-dose CT in patients with chronic lumbar back pain [34].

To the authors’ knowledge, only few recent studies investigated CT with reduced doses specifically for diagnostics of vertebral fractures. The diagnostic performance of lumbar low-dose CT (47–69% radiation dose reduction) combined with IR was comparable to that of standard-dose CT with IR [13]. Higher levels of IR for low-dose CT (50% radiation dose reduction) still provided high image quality and diagnostic confidence [14]. In contrast to these studies, we simulated tube current reduction, which allows for intra-subject comparisons between standard- and low-dose MDCT with virtually lowered tube currents down to even 10% of original imaging. Thus, low-dose MDCT was simulated with relative dose reduction steps in comparison to original MDCT, which enables systematic virtual tube current reductions as a fraction of the initially performed, optimal scanning protocol according to the scanner’s automatic tube current modulation.

We further applied sparse sampling, which is novel for diagnostics of vertebral fractures. So far, assessments of bone mineral density and microstructure at the spine derived from MDCT with sparse sampling have been performed, with sparse-sampled imaging appearing more robust in comparison to MDCT with virtually lowered tube currents [16, 17]. In the present study, sparse sampling was superior in terms of overall image quality, overall artifacts, and contrast of vertebrae when compared with MDCT with virtually lowered tube current (Table 3 and Fig. 3). These results were obtained with good to excellent correlations between two experienced readers for MDCT with virtually lowered tube currents and sparse sampling, respectively (Table 3 and Fig. 3). Furthermore, sparse sampling led to superior results in detecting vertebral fractures, with no missed fractures for D100 P50 in contrast to D50 P100 (Table 4). Diagnostic confidence and correct determination of fracture age was better for sparse sampling, with clinically acceptable determination of fracture age in D100 P25 compared with D25 P100 (Table 4 and Fig. 4).

There are limitations to this study. First, it is not yet possible to apply sparse sampling at commercial MDCT scanners, thus restricting direct clinical applicability. However, first results from a prototype were recently reported, indicating that sparse sampling for MDCT could become broadly available in future generations of MDCT scanners [19, 20]. Second, we used FBP instead of IR, but IR has the potential to provide increased image quality particularly for imaging with reduced doses [35‐37]. The use of algorithms taking advantage of artificial intelligence for image reconstructions might further regularly improve image quality in the near future [37]. Third, we solely enrolled patients with vertebral fractures and without implants, such as spinal instrumentation. Thus, upcoming studies may evaluate sparse sampling in cohorts with spinal implants to distinctly evaluate whether sparse sampling is also beneficial and even superior to tube current restrictions when implant-related metal artifacts are present in MDCT. Fourth, the retrospective design and the comparatively small patient cohort have to be acknowledged as a limitation. Prospective approaches including more patients are needed to confirm the results of the present study.

In conclusion, our results demonstrate the feasibility of using sparse sampling for fracture detection at the spine, with clear superiority as compared to MDCT with virtual reduction of tube current. Therefore, sparse sampling represents a promising option that might allow for even more drastic radiation dose reductions while revealing better image quality and diagnostic characteristics than MDCT with tube current reduction does.