Accuracy of 3D printed spine models for pre-surgical planning of complex adolescent idiopathic scoliosis (AIS) in spinal surgeries: a case series

Highlights • Surgical planning for Adolescent idiopathic scoliosis (AIS) is very complex.• 3D printed model provides an excellent opportunity for surgical planning.• Accuracy of 3D printed model for complex for AIS planning is not reported before.• Case series, evaluates accuracy of 3D printed models for varying spinal deformities indicated by Cobb's angle (as the 40° to 95°)• Accuracy of surgical model is dependent on Imaging, post processing parameters & 3D printing technology.


Introduction
Scoliosis is a three-dimensional (3D) structural deformity of spine with symptoms including pain, psychological morbidity and cardiorespiratory dysfunction [1]. The most common indication for surgical intervention in spinal deformity is adolescent idiopathic scoliosis (AIS) [2]. Although 80% of all scoliosis cases are classified as idiopathic, the current view is that AIS is likely a multifactorial disorder with genetic predisposition [3]. The prevalence of AIS is estimated to be between 1-3% affecting children between 10 -16 years of age (adolescent). AIS is defined by a spinal curvature of 10 degrees in the coronal plane usually affecting females [4,5]. Enlarged Cobb angles present a lower frequency with curves greater than 40 degrees constituting 0.1% of the AIS population [2]. The principles of scoliosis management have remained unchanged. A Cobb angle of 45-50 degrees indicates the need for surgery, hence a high risk of progressive deterioration into adulthood [6,7].
Surgical intervention aims to reduce pain associated with progression of the deformity and thereby stabilising the curve to reduce the impact on internal organs [8,9]. Pedicle screws are commonly used to achieve a three-column fixation of the spine, despite the wide variation in the pedicle anatomy [10]. The risks and complications associated with pedicle screws (poly axial or mono axial) include perforation of the bony cortex from incorrect screw fixation with the reported incidence of misplaced screws ranging from 3% to as high as 25% [11−13].
Accurate placement of pedicle screws is crucial to avoid complications in spinal procedures. Thus, the surgeon may have to use intraoperative fluoroscopy, which is an imaging technique using the Xrays to obtain real-time images of the reconstructive metallic device several times during a surgery to aid precise pedicle screw placement. This incurs an increased radiation exposure to the patient and increases the intra-operative time [14,15]. Surgical complications in spinal procedures can be potentially devastating due to the proximity of vital structures. Most common complications include neurological injury or wound infection [13−16]. Vigorous planning in the preoperative stages can aid the surgical team to anticipate and minimise risks in such cases.
The advancement in the quality and spatial resolution of medical imaging has made it possible to create virtual 3D reconstruction of complex anatomy from computed tomography (CT) and magnetic resonance imaging (MRI), using various image processing techniques [17−21]. Clinical decision making can be done based on 3D reconstruction created after segmentation of CT image data. Therefore, accuracy of segmentation is a determinant factor for the success of such cases. Selection of optimal procedure and armamentarium including screws and implants by the surgeon can be done using patient's 3D virtual construct. This preoperative planning can help to mitigate the risks associated with surgery [22].
Development in imaging and sophistication in 3D printing methods has helped in the evolution of preoperative surgical planning. Today anatomical models can be fabricated with much ease using 3D Printing/Additive manufacturing/rapid prototyping. [13,14,23,24]. In the present scenario, spinal surgeons rely heavily on pre-operative CT-scans, which provide an ideal opportunity to manufacture 3D printed models, resulting in improved outcomes, less repetitive scanning, hence decreased radiation exposure.
Application of 3D printing technology plays an important role in identification and measurement of 'Cobb angle' for complex vertebral deformation [23,25,26]. 3D Printed models can be a suitable alternative towards reduction of the risk associated with surgeries based on visual inspection [26]. According to current literature, there is no report or long-term randomised control studies to support the role of 3d printing in reducing the surgical risk for complex AIS cases. A repeatable method to fabricate highly accurate model, could enable widespread use in simulation and pre-operative planning minimising clinical risks. This could also enhance patient education, decrease operative time and prevent unanticipated problems. 3D printed models are expected to be the exact replica of targeted anatomy. However, achieving precision and accuracy with image processing along with suitable fabrication process also incur potential errors. Therefore, acceptable clinical tolerance in the fabricated models demand rigorous examination [27].
The present study is a case series evaluating image data from patients with varying degrees of scoliosis (Cobb angle varying from 40 to 92 degrees). The entire process from CT scan to 3D model is optimised. Further, the 3D printed spine models have been rescanned using CT and their accuracy has been tested. This study has the potential of improving presurgical planning for complex spinal surgery by accurate pedicle screw placement along with reduction in operative time and radiation exposure in spinal surgeries especially for young people. This challenge is still unaddressed in current literature.

Method and materials
A brief overview of the process from image acquisition to analysis has been documented in following sections.

3D model development
Anonymised CT-scan images (slice thickness: 1-2 mm; image voxel: 512£512£434) of five recruited patients, suffering from varying degrees of scoliosis were obtained from Royal National Orthopaedic Hospital, Stanmore, United Kingdom (ethical permission was obtained prior to the study; ethical permission reference no. 19/LO/ 1466). Informed consent was also obtained from the patients. The five CT-scan datasets used in this study consisted of two female and three male subjects (between 9-16 years age), with mean age of 13 years suffering from idiopathic scoliosis. These were chronologically ordered as, K1, K4, K7, K10 and K11, with Cobb angles of 39.78°, 95°, 49°, 51°, and 92°degrees, respectively (Table 1). There were multiple scoliotic curves in few of our cohort chosen for the present study, however, only one deformation curve was considered.
The 3D solid models of the patient's scoliotic spines were obtained using an image analysis software package from Simpleware ScanIP (Synopsys, Inc., UK). The process comprised of the following steps: segmentation of the CT-scan images based on Hounsfield Units (HU) of cortical bone (HU 100-2000), mask development [28], smoothing of the contours of each slice using manual segmentation, and preparing a solid model of the spine from the masks ( Table 2). The segmentation and surface mesh quality were checked for irregularities, holes and overlapping edges. The segmentation of each spine data formed the ground-truth (models developed from the patients CT scan data) comparison for the corresponding 3D printed model segmentation, which was used to evaluate the accuracy of the 3D printing workflow. The methodology from data acquisition to generation of 3D printed model is described in Fig. 1. The created 3D surface model from patient CT-Scan was then exported for 3D printing in STL file format for 3D printing of the spinal models.

3D printing of the anatomical models
The STL file was pre-processed for 3D printing (Dreamer, Zhejiang Flashforge 3D technology Co., Ltd., China). Vertebral bodies of the models (Fig. 2) were printed using PolyWood (PolyMaker, China), where layer height was maintained at 0.18 mm, infill pattern was gyroid, print speed 50 mm/s, the extruder nozzle width was 0.4 mm, nozzle temperature was 220°C and the platform temperature was 55°C.
CT-scans were obtained for 3D Printed models. The same methodology (Section 2.1) was followed to generate virtual 3D masks of each printed model. Manual selection was required as the threshold ranges for segmentation were adjusted to accommodate for the difference in greyscale values (HU) as the 3D models were made of thermoplastics. The CT data was exported in DICOM file format and processed in Simpleware ScanIP (Synopsys Inc., UK). Once satisfied that the vertebral bone had been optimally isolated, a smoothing filter was applied only once using the recursive Gaussian tool with a 1pixel radius to smooth the surface and minimise the effect on geometrical accuracy. The 3D printed models were visually inspected after printing for any obvious printing errors.

Surface deviation analysis of the 3D printed models
Simpleware ScanIP (Synopsis Inc., UK) software was used for superimposition and surface deviation analysis of the virtual segmentations of the patient and model. The original patient segmentation mask was isolated, and a model preview was generated for the highest quality. This was further converted from 'Mask to Surface' using the surface tools, and a new surface was generated. These steps were repeated for the segmentation mask of the 3D printed model.
The two datasets (from the patient CT scans, and the scans from the 3D printed models of the scoliotics spines) were registered; the patient's surface was chosen as the 'fixed reference dataset' and the corresponding printed model surface for the corresponding registration dataset. For consistency, six corresponding landmarks were then manually selected on each surface, three in the sagittal plane and three in the coronal plane, in the upper, middle, and lower zones of the patient surface and corresponding model surface (Figure 3). This  resulted in the root mean square error (RMSE) within the two consecutive anatomical models. The cobb angle of the scoliotic spines was measured by the surgical team and analysed further to identify any correlation to the RMSE errors from surface deviation analysis of the severely affected areas.

Result and discussion
This study evaluated accuracy of 3D printed spinal models from CT scan image of 5 patients with scoliosis and cobb angle varying between 40-95°. Accuracy of spinal model is paramount if it must be used by surgical team for pre/intra operative planning or surgical guide designs in complex scoliosis cases. As there was lack of evidence for accuracy of 3D printed spinal models, validation study was planned by performing a rescan of the 3D printed model using standard CT protocol and further reconfirming its accuracy, so that any variation during 3D printing process, use of materials etc. can be rigorously evaluated and optimised.

CT reconstruction of complex spine
Image acquisition and segmentation form the backbone in determining the accuracy of printed models. Image acquisition needs to be performed by software for medical purposes. This study used commercially available Simpleware ScanIP (Synopsys Inc., UK). Auto segmentation is helpful for most bone segmentation [29,30], but complex structures demand manual segmentation. Knowledge of human anatomy helps to understand anatomical structures to perform complex segmentation such as in AIS [17]. Auto segmentation was checked manually for any inaccuracies and corrected, to ensure accuracy of segmentation process for complex spinal anatomies.

CT scanning of 3D printed models and their accuracy
Fused filament fabrication (FFF) printing method has been commonly cited for production of anatomical models using PolyWood plastic filament due to its relatively low cost and marked accuracy for medical modelling. [20,21,31]. FFF as a printing method provides print accuracy of 0.4 mm, thus can be accurately scanned using current CT protocol (using slices ranged from 0.5 to 1 mm) [32]. This does substantiate the use of (FFF) printing with PolyWood materials for fabrication of complex scoliosis models.

Accuracy of 3D printed model and correlation with cobb angle
Both the 3D models (Fig. 4 a-e), developed from ground-truth CTscans of the scoliotic patients and the ones developed from re-scans of the 3D printed models (Fig. 4 a1-e1)were found to be accurate on visual inspection (Fig. 4). To establish a quantitative difference between ground-truth CT-scans and 3D printed spine models, reconstructed CT data from both were superimposed to calculate values of deviation and represented as the root mean square error (RMSE) values for individual patient spine. RMSE is a reliable measure for consideration of accuracy between 3D scans [18,33−36]. These data were quantitatively represented as colour map as shown in (Fig. 4  a2-e2).
To establish correlation between complexity of spine and accuracy of corresponding 3D printed model, Cobb angle with increasing deformity for 5 patients was plotted against RMSE values as shown in , which support surgical accuracy established by our models. The effect of scanning with different models of CT scanner using different parameters did not appear to affect or diminish the accuracy of the models produced. Moreover, the CT scan machines are supposed to be calibrated using phantoms to identify accurate HU values for the living tissues. The good agreement of the actual contours of the pathological spines, also served as an indirect measure of the optimal calibration, hence justifying the reliability of the machine.
In this study, the focus was scoliosis affecting the thoracolumbar spine. The present methodology could be implemented to evaluate other spinal pathologies. The unique, complex anatomy of the cervical vertebrae along with its interface with the skull is a challenging area to undertake surgical instrumentation. Prospective controlled studies with larger samples would help assess the validity and reliability of this method. Accurate and repeatable methods would help increase the potential uptake of 3D printing technology in the surgical community.

Conclusion
This work critically examined the capabilities of patient specific 3D printed spinal models for complex scoliosis surgery with Cobb angle varying from 40 to 95°. The work demonstrated that FFF based 3D printing workflow could be adapted for presurgical planning of complex adolescent scoliotic patients. This provides clinically acceptable level of accuracy for surgical planning and screw placements practice within 0.5 mm accuracy, shows promise for this technology adoption for safer surgical planning for complex AIS. The benefits and drawbacks for both patients and staff and the long-term clinical efficacy and safety of using 3D printed models need to be further evaluated if we are to see more widespread uptake. This would require larger patient cohorts and long-term studies to investigate this expanding clinical field.

Research ethics
The study was conducted following ethical permission reference no. 19/LO/1466. Consent was also obtained for the use of the CT data in accordance with the study protocol.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.