Measurement of distances and locations of thoracic and lumbar vertebral bodies from CT scans in cases of spinal deformation

According to the World Health Organization (WHO), idiopathic and neurogenic scoliosis are among the largest contributors to the need for rehabilitation services in both children and adults [1]. Scoliosis is a condition where the vertebral column is laterally curved by at least 10\(^\circ\) in coronal view in combination with rotation and torsion around the vertical axis [2]. Depending on the cause, scoliosis can occur at all ages and may develop in different degrees of severity. Its prevalence increases from less than 1 % in newborns to 1-2 % in adolescents [3]. In elderly people of the age of 60 to 90 the prevalence ranges up to 68 %, including secondary scoliosis [4, 5]. The high rates in the elderly may be either a continuation of adolescent idiopathic scoliosis, or de novo due to degenerative processes and other causes [6]. Depending on the cause and severity of scoliosis and the age of the affected person, different forms of treatment are required, ranging from conservative physiotherapy to surgical intervention [7]. Due to the potentially rapid progression of scoliosis, especially during growth, an early diagnosis and regular follow-ups are important [8]. In addition to the clinical examination, X-rays are the gold standard for initial diagnosis, followed by CT and MRI, if necessary [9]. The lateral curvature of the spine is quantified by measuring the Cobb angle from the X-ray image in coronal view [10]. The state and progression of scoliosis are then usually monitored by regular X-rays. As a result, young patients who show signs of scoliosis are often exposed to significant ionizing radiation from regular X-rays during follow-up examinations, which can be associated with an increasing risk of radiation-related health problems later in life [11, 12]. Therefore, young individuals in particular could benefit from alternative scoliosis assessment methods that are free of ionizing radiation and help to reduce the number of X-rays in follow-up examinations. This could lower the potential risk of long-term radiation related health issues.

Towards an alternative for scoliosis assessment during follow-ups that is free of the use of ionizing radiation, among others [13], methods based on the analysis of 3D body scanner images of the torso have been developed [14‐16]. This body scanner uses infrared scanning and video technologies to reconstruct a 3D image of the scanned person’s outer body contour and thus is completely non-invasive and free of ionizing radiation. Moreover, the entire scanning and image reconstruction process is fairly fast and comfortable for the patient [14]. For the analysis of the reconstructed 3D image of the patient’s outer body contour and to further derive the spinal curvature from it, finite-element method (FEM) simulations have been performed on a biomechanical model of the spine and ribcage [14‐16]. The geometries of the individual model components were designed separately for females and males, based on diverse anatomical data available in the literature, such as the anatomical dimensions of the vertebral bodies and intervertebral disks per level, as well as the different geometries of the human ribs for female and male [17, 18]. Hence, there are gender-specific biomechanical models for females and males. After the FEM simulations, the patient-specific deformed biomechanical models were then fitted into the 3D body scan image of the patient. To do this in the best possible way, additional information about typical distances and the locations of characteristic reference points is required. This includes, for example, vertebral body to skin distances at different levels and the positions of the vertebral bodies in relation to the transverse body contours. The best fitting biomechanical model in the current 3D body scan image of the patient can then be used to derive the course of the vertebral column (Fig. 1), further described also in the Discussion section. In follow-up examinations, the patient-specific deformed biomechanical model and the spinal course derived from it can be compared with the previous simulation results.

Here, the individual distances from the centre of the vertebral body to the end tip of the spinous process and further along this axis to the skin have been measured from transverse CT slices at thoracic and lumbar vertebral levels. These measurements are in complement to other measures, where the distance from the centre of the vertebral body to the skin has been measured in the sagittal plane [14]. In the case of spinal curvature and vertebral rotation, however, the axis measured here from the centre of the vertebral body through the spinous process end tip to the skin is generally not in the sagittal plane. Further, the locations of the vertebral bodies have been analysed in relation to the patient’s individual transverse body contours and shown together with those of other patients. Although the method presented here was developed based on available CT images from adult patients, these measurements were taken in order to find and evaluate potential trends and characteristics that might be useful for further development of the scoliosis assessment methods that are based on the combination of 3D images from the body scanner and FEM simulations on a self-developed custom biomechanical models [15, 16]. These methods have the potential to become an alternative that does not require the use of ionizing radiation in follow-up examinations and could thus reduce the number of X-rays.

Figure 4 shows the transverse euclidean distances from the centre of the vertebral body to the end tip of the spinous process, \(L_{12}\) (light colour), and from the vertebral body to the skin, \(L_{13}\) (dark colour), respectively, for all thoracic and lumbar vertebral levels. The distances are given in absolute units. Further, data were grouped according to the severity of the patient’s spinal curvature from mild to strong, and by gender (left to right). If no clear end tip position of the spinous process could be identified, only the distance from the centre of the vertebral body to the skin is given (dark colour). Figure 5 shows the corresponding arithmetic mean values and their standard deviations of the lengths \(L_{12}\) and \(L_{13}\). Table 1 lists the corresponding arithmetic mean values and their standard deviations per level (top) and region (bottom), with no separation by gender.

Accordingly, Table 2 shows the arithmetic mean values and their standard deviations per region, after grouping by the range of spinal deformation and with no separation by gender.

Irrespective of the severity of the spinal curvature, the transverse euclidean distances from the centres of the vertebral bodies to the end tips of the spinous processes, \(L_{12}\) (cf. light coloured horizontal bars), follow a physiological trend. That is, they are maximal in the lumbar region (L2...L5) and decrease towards their minima around T7 level before they slightly increase again towards the upper thoracic/lower cervical region (T1...T4). Accordingly, the distances from the centres of the vertebral bodies to the skin, \(L_{13}\) (cf. dark coloured horizontal bars) follow a similar trend, although their increase towards the upper thoracic/lower cervical region appears slightly more pronounced: \(\frac{\overline{L_{12}}\text {(T1...T4)}}{\overline{L_{12}}\text {(T5...T9)}} < \frac{\overline{L_{13}}\text {(T1...T4)}}{\overline{L_{13}}\text {(T5...T9)}}\) for mild, medium and strong, where \(\overline{L_{ij}}\) denotes the regional arithmetic mean. When grouped by gender, the arithmetic mean values are often smaller for females than for males (cf. Fig. 5).

Further, the variations among different patients of the vertebral body to skin distances are larger and much increase with the severity of the spinal curvature in comparison to the corresponding vertebral body to spinous processes distances. That is, \(\text {SD}(\overline{L_{13}}) > \text {SD}(\overline{L_{12}})\) for mild medium and strong, and generally \(\text {SD}^{\text {Mild}}(\overline{L_{13}})< \text {SD}^{\text {Medium}}(\overline{L_{13}}) < \text {SD}^{\text {Strong}}(\overline{L_{13}})\), but for the lumbar region in case of strong, with \(\text {SD}(\overline{L_{1i}})\) being the corresponding standard deviation (SD) of \(\overline{L_{1i}}\) (cf. Table 1 bottom). Similarly, the standard deviation of the arithmetic mean of the vertebral body to skin distances appears to be maximal in the region that corresponds to the range of spinal curvature. That is, in case of thoracic curvature \(\text {SD}^{\text {T2...T11}}(\overline{L_{13}})> \text {SD}^{\text {T12/L1}}(\overline{L_{13}}) > \text {SD}^{\text {L2...L5}}(\overline{L_{13}})\) and accordingly \(\text {SD}^{\text {T12/L1}}(\overline{L_{13}}) > \text {SD}^{\text {T2...T11}}(\overline{L_{13}})\) & \(\text {SD}^{\text {T12/L1}}(\overline{L_{13}}) > \text {SD}^{\text {L2...L5}}(\overline{L_{13}})\) in case of thoraco-lumbar, and \(\text {SD}^{\text {L2...L5}}(\overline{L_{13}}) > \text {SD}^{\text {T12/L1}}(\overline{L_{13}})\) & \(\text {SD}^{\text {L2...L5}}(\overline{L_{13}}) > \text {SD}^{\text {T2...T11}}(\overline{L_{13}})\) (cf. Table 2).

Figure 6 shows the distribution of the centres of the vertebral bodies from all patients at all thoracic and lumbar vertebral levels. The three different symbols – circle, triangle, and square – are associated with the three classes of severity of the spinal deformation. Filled and open symbols refer to female and male patients, while different colours per symbol correspond to different patients within the same class of severity.

The location of the cluster of the centres of the vertebral bodies varies with vertebral level. While at the upper thoracic and lower lumbar levels the vertebral body positions accumulate near the centre of the ellipse, the cluster shifts towards the middle of the lower/back half section at levels T7...T10. Strong spinal curvatures further show a clear lateral spread. With increasing spread their positions also seem to shift towards the centre line between back and front. The data do not show any clear trend or difference between females and males.

In the present study, software tools and methods are presented to measure the distances between the vertebral body and the end tip of the spinous process, and between the vertebral body and the skin from transverse CT images. Furthermore, normalization with respect to the area enclosed by the individual transverse body contour allowed to display the location of the vertebral bodies in relation to a generalized transverse body contour shape along with those of other patients, regardless of their body shape and size.

To our knowledge, this is the first study analysing the vertebral body to skin distances in the transverse plane along the axis through the spinous process end tip and showing the centre position of the vertebral bodies normalized to the area encompassed by the transverse body contour from transverse CT images. The present measurements along the axis from the vertebral body through the spinous process end tip are thus different from other measurements [14], where the distances have been measured along the intersection of the transverse and sagittal plane. In case of spinal deformation, however, the axis from the vertebral body through the spinous process end tip is not collinear to the sagittal plane. Several anatomical data are available in the literature for the vertebral body structures [17, 18]. The present study aims to investigate the localization of vertebral bodies within body contours in scoliosis using CT images. The final aim is to develop a non-invasive scoliosis assessment method in which a biomechanical thoracic spine model [15] based on anatomical data will be optimally integrated into body contours. The present findings might help to optimize the positioning of the model into the transverse cross sections of the torso.

The relatively small variations in the distances between the vertebral bodies and the end tips of the spinous processes in different patients, represented by the standard deviation (SD) of the corresponding \(L_{12}\) mean (cf. Fig. 5, Table 1), indicate that this measured distance seems to be similar in adult patients with different ages and within the same gender group. However, no clear evidence that this distance is independent of age could be given, as the patients were randomly selected and the number of patients available for this study does not allow for further statistics after grouping by age. Furthermore, as the images were anonymized, no further information about the individual body shapes is available than can be seen in the CT images themselves. Since these distances were measured in transverse view (cf. Fig. 2), they do not necessarily correspond at all vertebral levels to the distances between the centre of the vertebral bodies and the spinous processes end tips of the same vertebral bodies. The variation of the distances between the vertebral bodies and the skin is much larger than the variation of the distances between the vertebral bodies and spinous processes end tips, i.e. generally \(\text {SD}(L_{13}) > \text {SD}(L_{12})\), since the vertebral body to skin measures include also the individual thicknesses of the skin tissue, fat and musculature. Moreover, the variation of this distance appears to increase with the severity of the spinal deformation (cf. Table 1), since in some cases, rotation of the vertebral bodies about the vertical axis may additionally be involved. Accordingly, the arithmetic mean of the variations is maximal at its corresponding range of scoliosis (cf. Table 2). Methodically, the measured distance between the vertebral body and the skin is therefore also affected by the rotation of the vertebral body around the vertical body axis. Since the length of the straight from the vertebral body through the end tip of the spinous process was measured, this distance increases with increasing rotation of the vertebral body. These distances seem not to be correlated with the range of scoliosis, though, but on average increase from top – thoracic – to bottom – lumbar – region in either case of grouping (cf. \(\overline{L_{13}}\) in Tables 1 & 2). However, a direct comparison with other results is complicated by the fact that, to our knowledge, these particular distances have not yet been measured anywhere else.

Further, CT images are taken in the supine position, while the patient is standing upright for the body scan. This is one of the limitations of the present study because the spine bears more load in the standing position while the spine is unloaded in the supine position. However, previous studies showed that Cobb angles measured in supine positions were linearly correlated with the Cobb angles measured in standing positions  [26]. Therefore, the present measured distances could have been underestimated but might be correlated to scan data from standing upright position with an acceptable degree of accuracy for the application presented here [27‐29]. However, the correlation has not been investigated here, since CT images and body scans were not available from the same person. Furthermore, the shape of individual transverse body contours may vary in the supine and standing positions due to the influence of muscles and fat tissues.

The method of presenting the positions of the vertebral bodies as a function of the normalized transverse body area quadrants enables the plotting of the locations with those from other patients, regardless of their individual body shape and contour, and to identify potential clusters. The distinct lateral spread of some of the locations of the vertebral bodies at certain vertebral levels is associated with a pronounced spinal deformation. In these cases, the accompanying shift of the location of the vertebral bodies towards the mid-line between back and front is due to the distinct rotation of the vertebral bodies around the vertical body axis and a clear asymmetry in the transverse body contour, due to the natural rib hump with posterior elevation at the convex side.

For female patients, the position of the thoracic vertebra is systematically biased by the proportion of the transverse area of the female breast. Hence, the positions of the thoracic vertebrae are slightly shifted towards the back at the levels of the female breast. This systematic effect, however, is expected to be rather small, as the fraction of the female breast of the transverse area is minor. An analysis of the location of the vertebral bodies did not show any clear trend upon gender grouping (cf. Fig. 6) and thus is in agreement with the prior assumption. For the same reason, a large abdomen also slightly affects the normalized positions of the vertebrae, irrespective of gender.

Intra- and inter-observer variability would be expected to be rather small, but mostly could have affected the positioning of the markers and the selection of the transverse slices closest to the vertical centre of the vertebral bodies. However, symmetry detection is an efficient, salient human visual property [30] and thus the precision of the human positioning of the markers can be expected to be high. For a similar reason, the observer variability in the selection of transverse slices is also expected to be rather small at the present step size of 5 mm along the CT scan direction. With potentially smaller step sizes, the differences among slices are getting smaller, which in turn reduces the potential error in visual selection of the centre slice.

Measurements of distances and angles of anatomical structures are common in orthopaedic and radiological practice but can be difficult on three-dimensional (3D) stacked images, such as CT and MRI, because different parts of the structure are seen on different images, contrary to plain X-ray. Various tools thus have been developed, for instance by Maizlin and Vos, to measure the Cobb’s angles and others on MRI and CT [31]. Recently also machine learning gained importance in 3D image analysis [32] and in getting anatomical information from CT and MRI images [33]. Further, machine learning is used in this field also for localization, segmentation [34], and others.

The results of the present study, however, could be of interest to health professionals and medical scientists. Detailed measurements of the anatomy are important in various fields. These measurements may help to further understand the pathomechanisms of spinal deformations [35] or to identify reference points in X-ray examination that could be used for reconstruction of a 3D spinal model [36, 37]. Or in another example, they could be used in combination with other morphological data and anatomical information to improve the prediction of 3D spinal alignment from external shape [38], or the fitting of individually distorted biomechanical models of the vertebral column to body scanner images in the course of developing scoliosis assessment methods that do not require the use of ionizing radiation during follow ups [16]. Although young patients who show signs of scoliosis and are particularly affected by regular X-rays usually have idiopathic or neurogenic scoliosis, the methods presented here can also be transferred to other types of scoliosis, such as muscular, degenerative, and congenital scoliosis at all ages.

A schematic representation of the potential use of the current results in combination with data from other studies [15‐17] was indicated in the Background section and depicted in Fig. 1. There, one of the challenges appears to be an automated fitting of an individually distorted biomechanical model (Fig. 1b.) inside the patient’s 3D body scan (Fig. 1a./d.). Data from the present study, like the positions of the vertebral bodies and typical distances from the skin could help to confine the number of degrees of freedom in the fitting algorithm at certain vertebral levels (Fig. 1c.). Furthermore, also health professionals and in particular orthopaedic specialists could profit from this internally developed, user friendly software that displays the anatomical structures in high detail from different perspectives – coronal, sagittal, and transversal – and provides different measurement tools that could be easily further developed according to the specialists desires. Despite major achievements in identifying, segmenting, and localizing anatomical structures by artificial intelligence and machine learning, visual inspection of the individual CT images and diverse measuring tools are important for medical professionals to provide proper individual treatment.

In the future, the present principle study will be extended to more cases and the marker positions will additionally be used to analyse the rotation of the vertebral bodies. Also, the positioning of the markers could potentially be automated, using (semi-) automatic localization tools to find characteristic points in 3D volumes. Further, the measured distances could be corrected for different body sizes, using an approach similar to that used to correct the location of the centre of the vertebral within an ellipse.

Distances between the centres of the vertebral bodies and the spinous processes end tips, and from the vertebral bodies to the skin have been measured from transverse CT images at all thoracic and lumbar levels from patients with various severe spinal deformation. The distances between the vertebral bodies and the spinous process end tips appeared to be similar at corresponding vertebral levels after grouping by gender. However, trends could be found as a function of the vertebral level. Further, the vertebral body centers positions could be displayed at thoracic and lumbar vertebral levels across different patients by transverse body contour area normalization. Again, the first trends as a function of vertebral level and upon grouping into mild, medium, and strong spinal deformation were found within the limited number of analysed CT scanes, which could further be used in context with biomechanical modeling of patient’s individual spinal deformation in scoliosis assessment.