Introduction

The evaluation of 3D spine deformity in scoliosis is challenging and optimally requires comprehension and use of 3D clinical parameters [1, 2]. The correct interpretation of spinal deformities is mandatory to define the optimal treatment strategy for the patients. Different methods for 3D evaluations have been used and evaluated [3], and reconstruction based on stereo-radiography is a commonly used method. Several studies have investigated the possibility of predicting progression of scoliosis based on 3D parameters [1, 2, 4,5,6], since predicting scoliosis progression at an early stage would be of paramount importance. Apical vertebra rotation (AVR), torsional index of the spine (TI), and intra-vertebral rotation (IAR) have been proven significant parameters in determining progression in mild scoliosis (Cobb angle < 25° [2, 4]). In the recent 20 years, a lot of effort has gone into defining the “gold standard” for 3D parameters, and to apply these for effective and easy-to-use tools in daily clinical life [3].

The repeated use of X-ray imaging needed for scoliotic patient follow-up has been of concern in recent years. Ionizing radiation has been associated with a potential risk of developing radiation-induced cancer in scoliotic patients [7,8,9,10]. Children have a long life expectancy and are thought to be especially sensitive to long-term stochastic effects from ionizing radiation. Thus, it is of great importance taking steps towards using methods reducing the radiation exposure to our patients. The best approach of course would be to define robust methods of early detection of progressive scoliosis and more efficient methods of treatment in order to limit the number of radiographic exams needed for follow-up. However, although promising results have been reported in the literature, such methods are still not validated or widespread [4, 6, 11]. The second-best approach is to reduce the ionizing radiation delivered by the radiological exam.

EOS® low-dose stereo-radiography (EOS Imaging) is an imaging system that allows for high-quality imaging at a radiation dose lower than most conventional systems [5, 8, 12], adhering to the ALARA dose-optimization principle of keeping dose as low as reasonably achievable [13]. 3D reconstruction from EOS imaging stereo-radiography has been described in several previous studies [4, 14,15,16,17]. Good reliability on 3D parameters has been reported for both standard-dose and micro-dose protocols [14, 15, 17]. Ilharreborde et al [15, 17] looked at both standard-dose and micro-dose protocols with regard to intra- and inter-observer reproducibility. Results were satisfactory for both modalities and a significant reduction of dose compared with the original standard-dose protocol was described. We hypothesized that the radiation dose delivered to the patient could be reduced even further without compromising reliability of 3D reconstructions. The aim of the present study was to investigate the possibility of reducing the dose of the established micro-dose protocol retaining the possibility of trustworthy 3D reconstructions from the EOS imaging stereo-radiography.

Materials and methods

Defining the reduced micro-dose protocol

The minimal dose judged to yield sufficient image quality for recognition of anatomical landmarks was defined by imaging a clinically validated ATOM dosimetry child phantom (CIRS, Computerized Imaging Reference System, Inc.) [18]. Figure 1 shows the phantom in posterior-anterior-lateral (PAL) positioning within an EOS scanner. Radiographic expositions were made with sequentially lower dose settings. Radiation dose exposure from the EOS micro-dose protocol was reduced by decreasing the current, milliamps (mA), and the scan speed. Both parameters are directly proportional to radiation dose: a 25% decrease of mA reduces exposure by 25%. A change of scan speed from speed 4 to speed 3 likewise results in a reduction of radiation dose by 25%. An experienced surgeon rated image quality with a semi-quantitative approach: phantom images were cut in regions of interest (lumbar, thoracic and full body, in frontal and lateral views) and anonymized, so the surgeon could blindly grade them, in a random order of region and quality. A score from 1 to 5 was assigned to each image by the surgeon (1 = optimal, 5 = unacceptable), and all images were scored twice. A cumulative score was calculated for each dose and plotted against dose. A sharp increase of image quality was noticed at 28 mGy.cm2 (50 mA and 60 kV for frontal imaging and 50 mA and 80 kV for lateral imaging, with a scan speed of 2): although the score increase was not statistically significant, this cutoff value was chosen. Preliminary in vivo measurement confirmed the readability of the X-rays with these settings.

Fig. 1
figure 1

The anthropomorphic phantom, representing a 5-year-old child, placed in posterior-anterior-lateral positioning within the EOS scanner

Full size image

Theoretic dose reductions were calculated from proportional differences of dose area product (DAP) values between the standard-dose, micro-dose, and reduced micro-dose protocols (Table 1).

Table 1 Scan protocols and resulting DAP values for the 5-year-old anthropomorphic phantom
Full size table

Inclusions

The local ethics review board approved of the study design and methods. A consecutive group of 18 children, 12 years of age or younger, planned for routine clinical and radiological investigation of scoliosis were offered micro-dose and reduced micro-dose images instead of one standard-dose image. An informed consent was obtained for each patient prior to imaging. Images with both protocols were obtained at the same radiological session, one after the other, no more than 2 min apart. This method allowed for direct comparison of 3D parameter reproducibility between the two modalities. Exclusion criteria were severe obesity, previous spine surgery with implants, and mal-positioning of the patients.

3D reconstructions

A validated method of 3D reconstruction of the spine from EOS 2D biplane images was used [14]. Patient data and acquisition settings were blinded and reconstructions took place in random order. Three operators, all trained within 3D reconstructions, did two reconstructions for each obtained image. One operator determined, for each patient, the levels of junctional and apical vertebrae for each scoliotic curve. Table 2 lists the 3D parameters investigated. Figure 2 illustrates 3D reconstruction images using the reduced micro-dose protocol.

Fig. 2
figure 2

Examples of 3D reconstruction from reduced micro-dose protocol, coronal and lateral views

Full size image
Table 2 The different 3D parameters investigated
Full size table

Statistics

Intra- and inter-operator reproducibility were determined according to the ISO 5725-2:1994 standard, in terms of standard deviation. Bland-Altman plots were used to observe measurement agreement. Results were compared with previously published data on 3D reconstruction based on stereo-radiography and micro-dose [14, 17]. Correlations were analyzed with Spearman’s rank coefficient; significance was set at 0.05.

Results

The reduced micro-dose protocol corresponds to a theoretical reduction of radiation exposure of approximately 58% and 93% compared with micro-dose and standard-dose protocols, respectively. Table 1 shows the three scan settings and DAP values for the child phantom.

Preliminary in vivo images with the new reduced micro-dose setting allowed sufficient quality for 3D reconstruction. Figure 3 illustrates an example of micro-dose and reduced micro-dose full-spine imaging.

Fig. 3
figure 3

a Coronal full-spine image in EOS scanner using micro-dose protocol. b Coronal full-spine image in EOS scanner using reduced micro-dose protocol

Full size image

A group of 18 consecutive children going for routine clinical investigation for scoliosis were then assessed with both micro-dose and reduced micro-dose imaging. Three children were excluded; two were carrying braces during imaging, one had an abnormal number of vertebrae (14 thoracic vertebrae). The remaining 15 children were included in the study. The mean age was 10.7 years (range 4–12), gender distribution amongst the included patients: four males and 11 females. Mean reconstruction time was 10 min (range 6–21 min) for the micro-dose and 9 min (range 5–16 min) for the reduced micro-dose. Reconstruction time was not correlated with Cobb angle (i.e., with scoliosis severity, p > 0.05).

Reproducibility

A total of 180 3D reconstructions were made (15 patients × 2 modalities × 3 operators × 2 occurrences). 3D reconstructions were possible for all patients, and key anatomical landmarks needed for 3D reconstructions were visible for patients in both protocols. However, for both protocols, mostly the reduced micro-dose group, spinous processes were in some cases difficult to visualize because of increased vertebral rotation. Other anatomical landmarks such as vertebral endplates and pedicles were not affected to the same degree. Tables 3 and 4 show results on 3D repeatability and reproducibility along with results from previously published papers. Both micro-dose and reduced micro-dose showed good reproducibility; however, 3D reconstruction from standard-dose as demonstrated by Humbert et al 2009 [14] remained superior. Reproducibility between micro-dose and reduced micro-dose within this study was better for the micro-dose protocol. The highest degree of variability was on AVR and kyphosis parameters. Table 4 shows that reduced micro-dose was better on all parameters except pelvic tilt (PT) and T4-T12 kyphosis, compared with Ilharreborde et al (2016) [17] “fast-spine” micro-dose reconstructions.

Table 3 Intra-operator repeatability of clinical parameters, in terms of standard deviation of uncertainty, obtained in the current study and compared with existing literature. All parameters are expressed in degrees
Full size table
Table 4 Inter-operator reproducibility of clinical parameters, in terms of standard deviation of uncertainty, obtained in the current study and compared with existing literature. All parameters are expressed in degrees
Full size table

Discussion

The purpose of this study was to investigate and validate reproducibility of 3D reconstruction of the spine from stereo-radiography with a reduced micro-dose protocol in scoliotic pediatric patients. For most 3D parameters in mild, reproducibility was comparable to previous studies [14, 15, 17]. As expected, the reduced micro-dose protocol was less reliable than standard-dose and micro-dose for some parameters. 3D transverse rotational parameter uncertainty, AVR and torsion, was higher in this study on both reduced micro-dose and micro-dose protocols than the reported values from Humbert et al (2009) [14] using standard-dose, as well as uncertainties on Cobb angle and T4-T12 kyphosis (5.4° and 6.0° in reduced micro-dose, respectively, versus 3.1° and 3.8° in the previous work). However, the reproducibility obtained using reduced micro-dose “full” 3D reconstruction was superior in all clinical parameters except for PT to the results obtained using “fast spine 5min process” (Ilharreborde 2016) using micro-dose in patients with scoliosis severity comparable to this study. Thus, the reduced micro-dose protocol offered acceptable 3D reconstruction reliability of the spine in patients with mild scoliosis. Depending on the objective of the exam, such reliability would be fine for initial screening and follow-up of scoliosis.

Limitations

The definition of minimal dose was inherently subjective, since only one surgeon was implicated in the semi-quantitative definition of the cutoff dose, but efforts were taken to make it as objectively as possible. Quantitative parameters to determine image quality were also tested (such as signal-to-noise ratio), but they tended to vary linearly with dose variations, showing no useful cutoff value. The semi-quantitative approach utilized, on the other hand, implicitly accounted for the visibility of the anatomical landmarks of interest for the interpretation of the radiographic information and it showed a cutoff value indicating that image interpretation below a certain radiation dose (28 mGy.cm2) would suffer significantly.

A reduction of radiation dose exposure to the patients of more than 50% could be beneficial to the patients reducing potential harmful side effects to ionizing radiation considering the ALARA principle. Still, the risk benefit balance needs to always be evaluated according to the needs of a given radiological assessment. Existing EOS standard-dose protocol already offers high-quality images suitable for 3D reconstruction of the spine at a low radiation dose, as shown by Humbert et al [14]. For instance, the reduced micro-dose protocol would not be accurate enough to calculate the severity index of scoliosis progression [4] or simulate or plan surgery. Moreover, the reliability might not be accurate enough for research, where the development of algorithms and decision trees needs higher accuracy. Images obtained with reduced micro-dose were as expected of lower quality than standard-dose and micro-dose, i.e., more noisy and with less contrast. In standard-dose and micro-dose, the spinous processes are often difficult to visualize, which was generally worse for reduced micro-dose. Spinous process location is, along with pedicles, an important landmarks used to evaluate the axial orientation of the vertebra. However, pedicles were sufficiently recognizable in most patients, except one patient with severe kyphosis. This was independent of the imaging dose as it is inherent to the patient’s spinal geometry; this type of patient would also have been challenging with other 2D modalities and does in fact put a restriction on usability of 3D reconstruction from stereo-radiography. In some cases, regular CT should be advocated for.

For both modalities, T1, which is one of the landmarks needed to initialize the 3D reconstruction, was not always visible in lateral projection due to overlapping upper extremities/shoulders, although correct validated patient positioning was adopted in this work and patient mal-positioning was the cause for exclusion. Nevertheless, sagittal inclination of T1 can usually be inferred by the orientations of the adjacent vertebrae. The same applies in the mid-thoracic region where there is low visibility in the lateral view because of the large body span traversed by the X-rays.

As shown above, operator time is still a limiting factor since the average 3D reconstruction time was 10 min, which is often not compatible with everyday clinical routine. A new and faster method is needed to benefit optimally from this 3D analysis method, potentially automated, to reduce user dependence [14, 15, 17]. We do not recommend this new protocol for children with implants or wearing braces as these cases were not yet investigated.

Conclusion

We propose a new reduced micro-dose protocol for 3D reconstructions based on stereo-radiography which offers reliable 3D reconstructions for preliminary screening and follow-up in children with mild scoliosis. However, standard-dose protocol remains the option of choice for most accurate assessment and 3D reconstruction. The reduced micro-dose protocol is applicable to existing EOS systems and can be taken into use for children being assessed for mild scoliosis right away and could replace micro-dose for these patients.