Dose optimisation for intraoperative cone-beam flat-detector CT in paediatric spinal surgery

The phantom model allowed a systematic evaluation of different dose settings from the highest to the lowest possible. It also allowed different bone inserts containing pedicle screws to be evaluated with regard to image quality. This is especially important as the cornerstone of image guidance during pedicle screw insertion is the ability to clearly identify the cortex relative to the pedicle screw. Streak artefacts from metallic implants are a major concern, and any realistic model needs to incorporate this. Our phantom model could only be used to evaluate 90-degree perforations of the cortex due to the small size of the bone samples. However, to compensate for this limitation, for the sample with the perforating screw, the perforation was made as small as possible.

The CTDI₁₀₀ does not measure the actual absorbed dose by the individual patient but should be considered an index for comparisons. The homogeneous PMMA does not simulate the different tissue types and heterogeneities in vivo. The CTDI₁₀₀ will underestimate the radiation dose because the 100-mm pencil ion chamber only partly covers the collimated beam width. For X-ray beams wider than 40 mm, a pencil chamber longer than 100 mm is required. The CTDI_w will underestimate the ideal CTDI_w with approximately 20% at a collimated beam width of 20 mm [15]. Dixon [16] has described an alternative method to directly measure the dose at the central scan plane using a small ion chamber.

The patient dose of the cone-beam system relative to CT has been described by Zhang et al. [7], who showed that with identical techniques (kVp, mAs, etc.) and with the same scan length, the cone-beam system in 3-D mode delivered approximately half the radiation dose of a 64-slice CT scanner. Their use of a Farmer ion chamber gave a more correct estimation of patient dose compared to the 100-mm ion chamber used in our study for dose comparison as it could be used to directly measure the point dose at the centre of the scan length. Differences between cone-beam systems and CT are to be expected due to differences in source-to-isocentre distance (SID). The SID of the cone-beam system was 64.7 cm while the SID of the 64-slice CT scanner in question was 57.1 cm. This accounts for approximately 29% of the dose difference [7]. Second, there is a difference in fan beam angle. The cone-beam system has an angle of approximately 20 degrees, and the CT scanner approximately 45 degrees. This leads to less peripheral radiation dose measurements for the cone-beam system because the ion chamber is outside the beam during parts of the scan [7]. The differences in over beaming between the cone-beam system and the CT scanner will account for a slight dose reduction in favour of the cone-beam system [7].

Would a vertebra have been a better bone sample for the phantom part of the study? This would without doubt have better represented the normal anatomy. The cortical thickness of the femoral head is greater than the cortical thickness of the vertebra, and this would have influenced the accuracy of the imaging representation [17]. A more accurate model would probably improve the surgeons’ ability to discern whether a screw was placed correctly. The bone mineral density of children and adolescents changes significantly with age [18, 19]. The bone samples used here were acquired from the local bone bank. The femoral necks used in this study were obtained solely from patients undergoing hip arthroplasty, and these patients are in general not osteoporotic [18, 21]. Mean BMD for a hip arthroplasty patient is approximately 1 g/cm² with slight variation in several recent studies [20, 21]. Lumbar spine BMD in children varies from approximately 0.6 g/cm² (age 5 years) to 1.0 g/cm² (age 17 years) with much greater variation [18, 19]. A single bone sample would not model the bone density of any childhood vertebra since BMD changes with age. In this series (no BMD measurements), the probable difference between BMD in the femoral neck of elderly patients with hip arthrosis and children’s spines would tend to underestimate the radiation dose needed to satisfactorily visualise the correct placement of pedicle screws in young children because the BMD is considerably lower in children 1–9 years of age. However, this did not cause any problems for the implementation of the optimised protocol. For most children who undergo corrective spinal surgery (10–17 years of age), BMD of the lumbar spine is probably not very different from that of the bone samples used in our study.

A substantial number of pedicle screws are misplaced when navigation is not used. The rate of pedicle perforation varies between 10% and 40% [22‐24]. Our own data (unpublished) for the first 30 children where the screws were placed under fluoroscopic guidance and then position-controlled with volumetric imaging indicate that approximately 8% of a total of 424 screws were misplaced (according to the Ohlin classification [24]) and 1% of screws had to be repositioned or removed. Since we started using the cone-beam flat-panel detector scanner in 2008, we have not had any nerve or spinal cord damage caused by misplaced screws. This has been corroborated by several other studies [25‐27]. The cost-benefit of navigation is well-documented [5]. It is also well-documented that 3-D techniques are superior to other techniques used for navigation [22, 23]. Some surgeons use freehand techniques relying only on anatomical landmarks. The anatomy of the scoliotic spine in children is highly variable, the pedicles often very small and the deformity often great. Freehand techniques are only feasible for a few very talented and experienced spinal deformity surgeons. Most surgeons would prefer to use some kind of intraoperative imaging guidance. The number of misplaced screws in degenerative adult lumbar spines using freehand techniques was significantly greater than the number of misplaced screws placed following a 3-D scan with subsequent navigation [27]. According to the manufacturers, the radiation dose of one 3-D scan equals that of 35 s of fluoroscopy [7]. Fluoroscopy time has been calculated to be approximately 7–20 s per screw [28‐31]. With an average ten screws per child in our case study, this equals 68–193 s of fluoroscopy time per operation, or the equivalent of two to six volumetric scans, representing a higher radiation dose compared to our case study. One also needs to take into account that the pedicle screw misplacement rate is significantly reduced when navigation is based on 3-D imaging [22‐24]. The misplacement rate is even higher in children, which further strengthens the case for using volumetric guidance [27].

There are currently few alternatives to the system used in our study [32, 33]. Other available systems are slower and only two to three spinal levels can be imaged simultaneously, resulting in an imaging time of 8–9 min (as compared to 1 min of discontinuation time for the described system). Other low-dose systems like EOS [34, 35] or the system described by Abul-Kassim et al. [24] cannot be used intraoperatively. The EOS system in particular can only be used for standing or sitting patients [34, 35].

Abul-Kasim et al. [36] have also tried to optimise the radiation exposure and image quality of the same system used in our study. Using a completely different model (anthropomorphic adult chest model and porcine spine with pedicle screws), they arrive at almost the same dose settings as we recommend. The pedicle perforation rate in the porcine spine model was higher than the perforation rate simulated in our model, which may explain some difference in observer agreement at low dose.

The cylindrical-equivalent diameter of a body is defined as the diameter of the cylinder that the body would form if laterally compressed into a cylinder of equal cross-sectional area. This definition should not be confused with the definition of patient-equivalent cylinder based on the patient’s weight and height as suggested by The Danish National Board of Health [37]. The equivalent diameter may be calculated as d = 2√ab where a and b are the minor and major diameters of the ellipse. Note that the circumference of the patient-equivalent circle and the elliptic cross-section of the patient are not equal, and can therefore only estimate the diameter of the patient equivalent cylinder with an error of 10–15% depending on the eccentricity of the patient. Given that the difference in the diameter between the patient equivalent phantoms is about 50%, it might be sufficient to use the circumference to calculate the diameter of the patient to select the appropriate protocol.

The phantoms in our study simulated the lumbar anatomy. The results may therefore not adequately reflect the dose reduction achievable in the thorax. Based on the level of scattered radiation in the thorax compared with the lumbar level, we would anticipate the dose in the thorax to be even lower.

The lower-contrast images of the cone-beam scanner are not comparable with a standard CT [4], but in the case of spinal surgery, spatial resolution is of greater concern. Dealing with very high-density material (bone and metal) relative to water, we showed that the needed spatial resolution can be obtained at a fraction of the preset exposure values (Table 2).

The parameters of the optimised protocols were to some extent dictated by limitations of the generator. Even at the lowest possible tube current (10 mA), all phantoms at all voltages above 90 kVp showed acceptable image quality. If the mA were lowered further, a lower patient dose might have been achievable since the same dose at a higher kVp would result in a lower absorbed dose to the patient. Dose to the patient may have been reduced further by using a bow-tie filter typically employed in standard CT scanners [38]. A preliminary study with such a filter has shown that dose may be additionally reduced by a factor of 2. Automated current modulation as proposed by Kalender et al. [39] might also further reduce the dose as would adaptive statistical iterative reconstruction combined with conventional filtered back projection (a further estimated reduction of 30–40% [40]).