A reliability and validity study for Scolioscan: a radiation-free scoliosis assessment system using 3D ultrasound imaging
The results of our study showed very good reliability of Scolioscan for scanning conducted by the same and different operators as well as for the angle measurement performed by the same and different raters on the formed coronal spine images, with a mean ICC value of 0.94?±?0.04 (ranging from 0.88 to 0.97) between the two operators and among the three raters. The high intra- and inter-rater reliability for the angle measurement showed in this study was consistent with that reported earlier in a feasibility study of VPI method [32]. The RMS difference was smaller than 3.7° between the angles obtained from different measurements or different scans of the same patient. The sophisticated supporting frame, boards, supporters, and eye guiding spot on the patient screen within Scolioscan’s design may have partially contributed to the very good intra- and inter-operator reliability for scanning (with re-positioning for each scan), with the RMS difference smaller than 3.3°. The two Scolioscan operators/raters had been using the system for several months before this study, and the additional rater had been trained for a few days for the measurement. They were all graduates from Biomedical Engineering programmes with knowledge of ultrasound imaging. It will be worthwhile to understand the learning curve of new operators with different backgrounds in future studies.
Once a patient is confirmed with scoliosis, they have traditionally had to be exposed to radiography many times for monitoring, treatment planning, and treatment outcome measurement [9]. With the use of radiation-free Scolioscan, many of the radiation exposures may be avoidable, such as those used for progression monitoring, which may reduce the risk of inducing cancers [11–15]. As a consequence of this radiation hazard, it is conventionally not possible to use frequent radiography for monitoring scoliotic angle progression, thus there is no reference yet about the optimal frequency of taking image for scoliosis patients using the radiation-free Scolioscan system. It may be worthwhile to conduct investigations along this direction, with the consideration of the angle progression rate, risk factors, and cost effectiveness for different categories of patients.
Moderate to strong linear correlations were demonstrated between the Scolioscan angles and X-ray Cobb angles for the thoracic and lumbar regions and thoracic-lumbar data combined with coefficients of determination R2 larger than 0.72. Similar results were reported earlier using different laboratory prototypes of 3D ultrasound imaging system for scoliosis assessment [26, 30, 31, 34, 36]. It was found that the Scolioscan angle slightly underestimated the spinal deformity in comparison with Cobb angle for both the thoracic and lumbar regions. For the patients tested in the present study, the relationship between the Scolioscan angle (x) to Cobb angle (y) could be expressed by the equation y?=?1.1797x (R2?=?0.76). This finding is consistent with that reported previously using 3D ultrasound imaging for scoliotic angle measurement. The main reason for underestimation is that ultrasound images are taken posteriorly and provide anatomical features of vertebra posterior elements [32, 36, 43] rather than the vertebral bodies used in Cobb angle measurement from radiographs, as the processes are more identified than other spinal landmarks because of its sharp delineation in the ultrasound images. In this study, the profile formed by spinous processes in the VPI image was used for the deformity angle measurement. The RMS square difference between Scolioscan and Cobb angles obtained in this study (totally 73 angles) was 6.2°, and it became 4.7° when an adjustment for the Scolioscan angle was adopted using the obtained regression equation.
It has been well documented that the angle measured based on the profile of spinous processes in radiographs would underestimate the spinal deformity with reference to Cobb angle, showing that the spinous process angle was less angulated compared to Cobb angles [44]. It was reported that the magnitude of vertebral axial rotation correlated with the lateral deviation of vertebrae from the spinal axis [45, 46]. In fact, the spinous process deviations caused by vertebral rotation might result in the inaccuracy of interpretation on the vertebral body alignment on the radiographs of spine [47, 48]. A number of studies investigated how to transfer the spinous process angle to Cobb angle. An equation of y?=?1.3367x?+?1.3907 (R2?=?0.90) was proposed [44], with x representing the spinous angle and y the Cobb angle and both measured using radiographs. In the present study, the corresponding equations were y?=?1.1069x?+?1.7227 (R2?=?0.76) and y?=?1.1797x (R2?=?0.76) for the regression with and without intersection (Fig. 9). In the earlier feasibility study about angle measurement using VPI images, the results from 3D ultrasound was closer to Cobb angles, where the involved patients had a much smaller mean Cobb angle of 10.7?±?7.1° [32], in comparison with the mean Cobb angle of 22.6?±?9.5° in the present study. Further studies for Scolioscan with larger patient numbers and a wider range of Cobb angle would be necessary to investigate whether different regression equations should be used for scoliosis subjects with different Cobb angles and different types of spinal deformity.
Future studies can also be followed up to understand whether considering the vertebral rotation and other spinal deformities can further improve the agreement [49]. AIS is a three-dimensional spine deformity problem in coronal and sagittal planes and vertebral rotation [1], and deformity parameters in different planes may be dependent on each other [50–52]. Therefore, it is necessary to quantify spinal curvatures in sagittal or vertebral rotation in addition to coronal deformity, which will be useful for planning surgery, predicting prognosis and monitoring curve progression [30, 53, 54]. A recent study showed that the correlation between the spinous angle and the Cobb angle measured on radiographs could be improved with the consideration of vertebral rotation [49]. However, standing radiograph as the current gold standard for scoliosis investigation is difficult to directly acquire vertebral rotation, since these radiographs do not demonstrate the exact magnitude of the 3-dimensional spinal deformity present in patients with scoliosis [55]. Using the information obtained from the coronal and sagittal radiographs with reduced dose, EOS system can reconstruct 3D view of spine [17, 18]. However, it may still take some time to make the system more popularly used because of its high cost, low accessibility, and radiation (though with dose reduced), and long time required for building 3D spine model. Furthermore, its 3D presentation of the spine achieved using two orthogonal projection images requires further research to validate for different cases. Scolioscan used in the present study provided VPI images of spine for spinal deformity measurement in the coronal plane. During the scanning, Scolioscan actually acquires volumetric images of spine. It has been demonstrated in earlier studies that it is feasible to extract bony landmarks from the volumetric image data set to form virtual 3D spine model for the assessment of scoliotic deformity [25, 30, 31]. Further studies are going on to integrate this function into the system so that the evaluation of scoliotic deformity in 3D can be achieved using Scolioscan, including the measurement of spinal axial rotation and the deformity in sagittal plane. In this study, patients with Cobb angle larger than 50° were excluded due to the concern of the effect from rotation. Perhaps if the spinal rotation can be measured, future studies can include patients with larger Cobb angles.
While the reliability of using the VPI images generated by Scolioscan for the scoliosis assessment has been clearly demonstrated in this study, there are a number of areas to improve so as to achieve a more user-friendly clinical tool. First, patients with AIS are often observed to have winged scapula, and the protruded scapula obstructed the probe from scanning upwards even when the patients were told to cross arm. Hence the quality of VPI image was affected, making it difficult for accurate measurement of angle. In this study, the patients with severe winged scapula that affected scanning were excluded, which counted for approximately 10 % of the patients. In future studies, ultrasound probes with different widths and shapes may be used to find optimal configurations for different situations. In addition, patients with BMI larger than 25.0 kg/m2 were excluded from this study, counting approximately 10 % of patients. The current Scolioscan system used an ultrasound probe with frequency of 4–10 MHz, and bony features in images of the lumbar region of subjects were affected by the thick tissue layer in high BMI patients due to its attenuation to the ultrasound signals. One potential solution is to use a probe with relatively lower ultrasound frequency for obese patients, with the trade-off of reduced image resolution. Further study is necessary to investigate the optimized ultrasound frequencies for patients with different BMI with the consideration of tissue penetration and image resolution simultaneously.
Second, the VPI images provided by Scolioscan show many more features than the profile of spinous processes used in the present study, but they have not been used for the analysis of spinal deformity. As shown in the images in Fig. 5, transverse processes and ribs can be observed in most of the VPI images. The feasibility of using transverse processes for the spinal deformity measurement has been demonstrated earlier based on VPI images [32]. Further studies would be worthwhile to follow up on how to utilize more features in VPI images to provide more parameters related to spinal deformity, including vertebral rotation. Since ultrasound images also recorded information of paraspinal muscle architecture, it will also be valuable to extract muscle related parameters for scoliosis assessment. In addition, the coronal images formed by the current Scolioscan only covered lumbar and thoracic regions, and not the whole spine structure. Therefore, the overall spinal alignment as well as the thoracic shift cannot be assessed yet. Further developments and studies are required to enable Scolioscan to provide more information of the whole spine structure, which will further widen its application.
Third, it was found that the VPI image formation took between one and two minutes dependent on the height of the patient. While this is acceptable, it would be helpful if the VPI image can be provided immediately after the scanning so that the image quality can be confirmed and the patient can be discharged immediately after scanning. Related developments are underway and it has been demonstrated that real-time image formation is feasible using the system.
Fourth, the discrepancy between Cobb’s angle and Scolioscan angle could arise from the different time and day used for conducting both images. The inclusion criteria used in this study was shorter than 3 months between the two, with most of them within two weeks of each other. There may be some changes of angle during the period, and thus room for improvement exists.
Fifth, while the manual measurement of angle using VPI images appears very repeatable as demonstrated by the three raters in the present study, it reamins a subjective method. Although the three raters were mutually blinded for the measurement, they were in the same research team. Thus there was some common understanding about how to draw the lines on VPI images among the three raters. This may not be the case when Scolioscan is used in different clinical units, and different users may have different methods for drawing lines to measure angles. This may make the results difficult to compare among different clinical or research groups. This issue is not unique for obtaining Scolioscan angles in VPI images and can likely be overcome with clear operation and measurement guidelines. Radiographic Cobb angle measurement has been facing the same challenge, but it will greatly facilitate the measurement of spinal deformity angle based on VPI images if an automatic method can be developed. Related development work is ongoing.