The Accuracy and Effectiveness of Automatic Pedicle Screw Trajectory Planning Based on CT Values: An in Vitro Osteoporosis Model Study.

Pedicle screw placement in patients with osteoporosis is a serious clinical challenge. The bone mineral density (BMD) of the screw trajectory has been positively correlated with the screw pull-out force, while the CT value has been linearly correlated with the BMD. The purpose of this study was to establish an in vitro osteoporosis model and verify the accuracy and effectiveness of automated pedicle screw planning software based on CT values in this model.


Background
Osteoporosis is a metabolic bone disease characterized by low bone mass. It is a common disease in elderly people, especially in postmenopausal women, and can cause changes in the biomechanical properties of the vertebrae 1 . In diseases requiring pedicle screw placement, the intensity of pedicle screw xation may be reduced, loosened, and pulled out, which eventually leads to failure of the operation 2,3 . Many studies have been conducted to increase the local tissue density of the vertebrae by increasing the screw length, diameter, and thread structure or by injecting bone cement to increase the pull-out force of screws 4 . However, these methods all have shortcomings. Cement-augmented screws carry the risk of cement leakage which the most serious complications are spinal cord compression and pulmonary embolism 5 . The increase in screw diameter may be accompanied by the occurrence of pedicle microfractures, leading to a decrease in the screw pull-out force 6 .
The placement position of the pedicle screw is a crucial factor affecting the pull-out force of the screw, and its accuracy is related not only to malposition but also to the biomechanical function of the pedicle screw. Previous studies have demonstrated a linear relationship between the CT value and BMD and an exponential relationship between BMD and Young's modulus of bone tissue [7][8][9][10] . Therefore, placing the pedicle screw in a position where the CT value of the screw trajectory is as large as possible may be another method of pedicle screw placement for osteoporosis.
The application of navigation-guidance technology based on CT images can improve the accuracy of pedicle screw placement, avoid pedicle perforation, and effectively reduce the occurrence of nerve and vascular complications caused by pedicles 11 . However, these assistive techniques require preoperative or intraoperative analysis of the patient's CT images and manual planning of the entry point and trajectory of pedicle screws by the surgeon, which is a complicated and time-consuming process. Moreover, CT images alone are not enough for the surgeon to consider whether the bone mass at the screw trajectory provides adequate xation of the pedicle screw, especially in patients with osteoporosis. With the development of computer technology, more powerful computers and better algorithms have been developed, and automated planning using preoperative or intraoperative CT can nd a trajectory that meets the highest CT value and does not violate the placement standard. Several autoplanning methods have been designed [12][13][14][15] , but the studies were not planned in a way that speci cally targeted osteoporosis patients, and most studies have not conducted in vitro experiments to verify their accuracy and screw pull-out force.
To address these problems, this study used an automatic planning software to analyze CT data on an osteoporosis vertebra model to search trajectories of the highest CT value. Trajectories with nonperforating pedicles and non-piercing vertebrae were found to obtain a better effect of screw xation. The in vitro osteoporosis model was used to verify its accuracy and effectiveness.

Establishment of osteoporosis model in vitro
Ten lumbar vertebrae were taken from mature pigs (L1-L5, weighing 120-130 kg). All pigs were healthy, and none had been exposed to any factors that might affect BMD. All vertebrae were stripped of surrounding muscle tissue, ligaments, and periosteum. Part of the spinous process and transverse process were excised and embalmed with 10% paraformaldehyde for 48 hours. Micro-CT (μCT) was used for detection. Five of the 10 vertebrae were used as a decalci cation group, and the other 5 were used as a control group. Five vertebrae were randomly selected from all specimens using Ethylenediaminetetraacetic acid (EDTA) decalci ed uid for decalci cation. A 3 mm *10 mm hole was drilled in the upper and lower endplates of the vertebral body. During the decalci cation process, a stepper motor was connected to these holes through a silicone tube, and EDTA was pumped at a rate of 1 ml/min. This allowed the cancellous bone and cortical bone of the vertebrae to be decalci ed at the same time. The EDTA solution was replaced weekly, and the vertebrae were scanned weekly with μCT until the osteoporosis criteria were met.
The BMD level was expressed by T value, T value = (measured value -average normal BMD)/standard deviation of normal BMD. Due to the linear relationship between the CT value and BMD, the CT value of normal vertebrae could be used as a substitute value for normal BMD. The formula was rewritten as T' value = (measured CT value -average value of normal vertebral CT value)/standard deviation of normal vertebral CT value. Osteoporosis was de ned when the T' value ≤ -2.5.

Trajectory planning of pedicle screw placement
Software automatic planning and manual planning were used for pedicle screw trajectory planning respectively. Software automatic planning was used for the left pedicle, and manual planning was used for the right pedicle. All trajectories were required not to pierce the pedicle and vertebral anterior margin. In order to ensure the uniformity of biomechanical testing, the speci cations of all planned tracks were uni ed as 6mm*30mm.

Automatic planning
Automatic planning software based on CT values was provided by Chongqing Bosscom Technology Co., Ltd. (Chongqing, China). The surgeon only needs to manually adjust the area of interest to mark the target vertebral body and the narrowest part of the pedicle and specify the screw diameter and length according to the recommended screw diameter and length calculated by the software. Then, the software can automatically calculate and plan the screw trajectory with the largest CT value 16 . The software used an optimization algorithm to address the optimization problem of maximizing the mean CT value of the pedicle screw trajectory under a set of relevant constraints. The software was in the developing and testing stage, and this study was designed to study its accuracy and effectiveness. After its function was veri ed, the software will be added as a functional module in the optical and magnetic integrated surgical navigation system of Chongqing Bosscom Technology Co., Ltd.

Conventional manual planning
The arti cially planned trajectory passed through the midpoint of the minimum section of the pedicle, and the direction was parallel to the endplate on the vertebral body with an angle position the screw on the central axis of the pedicle. All trajectories were manually planned by the same senior spine surgeon. Mimics software (Version 21, Materialise, Belgium) was used to analyze the μCT data of the osteoporosis model, the trajectory of the automatic planning and manual planning were analyzed, and the CT values of the trajectory obtained by different planning methods were measured.

Design and manufacture of 3D printed guide plate
In this study, 3D printed guide plates were used to realize planned trajectories. Mimics and 3-Matic (Version 13, Materialise, Belgium) software were used to import CT data, and 3D image reconstruction was conducted. The 3D printing guide plates were designed according to the screw entry points and stop point coordinates obtained by automatic planning and manual planning. The automatic planning trajectories were on the left side, and manual planning trajectories were on the right side. The designed 3D printing guide plates were imported into the 3D printer software, the printing parameters were adjusted, and the plates were printed.

Pedicle screw insertion
Pedicle screw placement was assisted by 3D-printed guide plates. A 2.5 mm Kirschner wire was used to turn the hole through the guide head of the 3D-printed guide plate to obtain the screw trajectory, and a pedicle screw was inserted along the trajectory (6 mm*30 mm, Sanyou, Shanghai, China).

Measurement of pedicle screw placement error
After placement, X-rays and CT were used to verify the position of the pedicle screw to ensure that the pedicle screw was completely located in the pedicle without penetrating the pedicle and the vertebral cortex. The screw placement error was measured by preoperative and postoperative CT images. Similar to 2.3, d1 and θ1 represent the actual screw position as measured by postoperative CT. The displacement error is (d1-d); the angle error is (θ1-θ).

Screw pull-out test
Each vertebra was xed so that the long axis of the pedicle screw was coaxial with the direction of tension (ElectroPuls E10000 All-Electric Dynamic Test Instrument, Instron Industrial Products, Grove City, USA). The axial pull-out test was performed at a speed of 5 mm/min, and the force applied to the screw was recorded in displacement increments of 0.05 mm until the screw was completely removed. The maximum load encountered during the test was de ned as the peak load (pull-out force) at the time of failure. The test sequence of the left and right pedicles was alternated with each specimen to eliminate the effect of the test sequence.

Statistical analysis
SPSS (version 23.0, SPSS Inc., Chicago, IL, USA) was used for statistical analysis, and the data were expressed as the mean ± standard deviation. The normality of the two sets of continuous variables was assessed by the Shapiro-Wilk test (W test), and the F-test was used to determine whether the overall variance of the two sets of continuous variables was homogeneous. If the data were normally distributed and the variance was homogeneous, A paired sample T-test was used for pairwise comparisons of decalci cation group data, and an independent sample T-test was used for intergroup comparisons. P < 0.05 indicated a signi cant difference.

Results for the establishment of the osteoporosis model
The CT values of all vertebral bodies were 2767.87 ± 26.15 Hu. After 2 weeks of EDTA decalci cation, the CT value of the decalci cation group decreased from 2767.15 ± 27.08 Hu to 2688.8 ± 7.67 Hu (P <0.05), and the T' value decreased from -0.03 ± 1.03 to -3.03 ± 0.29 (P <0.05). All the vertebral bodies in the decalci cation group met the standard of osteoporosis (T' <-2.5) (Tab.1). 2. The position and CT value of the pedicle screw planning trajectory were measured.
The automatically planned screw trajectory was imported into the reconstructed CT image, and whether the screw perforated the lateral wall of the pedicle and the anterior edge of the vertebral body was evaluated from the axial view. None of the screws, whether automatic planning or manual planning, resulted in pedicle perforation or vertebral anterior edge perforation. The axial angle, coronal distance and coronal angle of the screw entry point in automatic planning were greater than those in manual planning (p<0.05), while the axial distance was less than that in manual planning (p<0.05) (Tab.2). That is, the automatic planned screw entry point is inward and upward, and the trajectory has a larger tail inclination and a smaller abduction angle.

Accuracy of pedicle screw placement
After pedicle screw placement assisted by 3D-printed guide plates, CT scans and Xray examinations were performed to verify the accuracy of pedicle screw placement ( Fig. 2 and Fig.3

Comparison of pull-out force
In the decalci ed osteoporotic vertebral body, the screw pull force of the manual planning group was 48.6% lower than that of the control group (P <0.05). The pull-out strength of the automatically planned trajectory was 44.7% higher than that of the manually planned trajectory (P <0.05) and reached 97.4% of the manually planned trajectory in the control group. There was no signi cant difference between the two groups (P >0.05) (Tab.3).

Discussion
Pedicle screw xation is a serious challenge in spinal surgery when confronted with osteoporosis. The in vitro osteoporosis experiment is the most direct and effective method to study the biomechanical properties of osteoporotic vertebrae. In this paper, the research was conducted by constructing in vitro models of osteoporotic pig vertebrae. In these models, the CT value and screw pull-out force of automatic trajectory planning based on the CT value and manual trajectory planning were evaluated and compared. The results showed that automatic trajectory planning was signi cantly better than manual trajectory planning in terms of screw pull-out force, which may suggest that automatic pedicle screw trajectory planning based on CT values is a valuable method for pedicle screw placement in osteoporosis.
The best models for studying osteoporosis in humans are from human cadavers, which are di cult to study because of their limited availability. One possible approach is to study vertebral specimens from large animals [19][20][21] . In this study, we used EDTA immersion and continuous pumping to demineralize pig vertebrae, which ensured that the cortical bone and cancellous bone of the vertebrae could be demineralized at the same time, which greatly shortened the time needed to reach osteoporosis. Moreover, this method does not seriously damage the structure of bone trabeculae in the vertebral body and retains the biomechanical properties of the vertebral body. Lee, C. Y. et al. 22 used EDTA to decalcify the pig vertebrae and, by measuring its microstructure and mechanical properties, concluded that EDTA decalci cation can help to produce a vertebral model with biomechanical properties consistent with human osteoporosis.
Pedicle screw placement in patients with osteoporosis has always been a di cult problem in spine surgery. Existing solutions include cement augmentation screw placement, cortical bone track screw placement (CBT), placement of varied materials in the screw path to increase local tissue density, increasing the screw diameter, and changing the screw thread structure. However, these methods all have their limitations. Bone cement leakage is a common complication of bone cement augmented screws, with a leakage rate of 11.6-82.4%, which can cause serious complications such as nerve injury, vascular injury, pulmonary cement embolism, cardiac embolism and anaphylactic shock 23 . In CBT, screws are inserted through the isthmus, and the screws can only be inserted into the posterior 1/3 of the vertebral body. The pull-out force is limited to the midposterior column, and the potential for correction of spinal malformation is poor. Filling the screw path with different materials, such as autologous bone, can increase the density around the screw in a certain time, but with time, the implanted material may gradually absorb or decompose, and the tissue density around the screw will gradually decrease, eventually leading to xation failure 24 . Increasing the screw diameter and changing the screw thread structure can increase the pull-out force, but blindly increasing the screw diameter may cause pedicle fracture and decrease the pull-out force 6, 25 . It is, therefore, necessary to seek other solutions.
The development and validation of automatic pedicle screw planning systems based on preoperative or intraoperative CT are in line with the current development trend of digital and intelligent spinal surgery. Computer-aided preoperative planning pedicle screw placement is a relatively fast process that can be performed without the automatic participation of a spine surgeon 26 . The optimization of the pedicle screw placement trajectory combined with the CT value is expected to improve the xation effect of pedicle screws. The automated planning system reduces the time required for preoperative or intraoperative planning, optimizes the procedure of spinal navigation surgery, avoids the problem of different optimal trajectories planned by different doctors, and optimizes the xation strength of pedicle screws. In this paper, we report software that can plan the trajectory of the maximum pull-out force screw according to the CT value of the vertebrae and verify the safety and effectiveness of software automatic trajectory planning. Since the widely used preoperative CT and intraoperative O-arm images can only display the CT value of the tissue, this study adopted the CT value as the evaluation index to simulate the actual application scenario. Since previous studies have proven that there is a linear relationship between the CT value and BMD, this method is feasible. In contrast to the previously reported automated planning trajectory, the automated planning trajectory used in this study did not require pedicle screws to pass through the midpoint of the pedicle, as the midpoint of the pedicle tends not to be the area with the densest bone tissue 27 . To achieve a higher extractable force of the screw, the screw needs to be as close to the cortical bone as possible without penetrating or causing the cortical bone fracture. Compared with Vijayan, R, et al. 28 , this planning method does not adopt the method of establishing a model atlas for pedicle screw planning. Although building a database can greatly reduce the time required for planning, it also loses the individualized characteristics of patients. The planning method in this study was to conduct screw planning according to the actual CT reconstructed images of patients, which can avoid screw misplacement caused by failure to match the atlas due to vertebral variation.
In this study, the manual planning group adopted parallel placement technology which inserted screw parallel to the upper endplate of the vertebral body, and the sagittal angle was 3.23 ± 1.64°, while the automatic planning was 19.19 ± 1.87°, which was signi cantly different (P <0.05), but both were within the allowable range 29 . The automatically planned trajectory is closer to the upper wall of the pedicle and eventually points to the upper endplate of the vertebral body to maximize the CT value of the whole trajectory to obtain the maximum screw pull-out force. Through measurement comparison, we found that in the osteoporosis model, the trajectory CT value of the software automatic planning group was signi cantly higher than that of the manual planning group by 25.7% (P <0.01). It should be noted that the CT values of screw tracks automatically planned in the decalci cation group were 21.4% higher than those manually planned in the control group (P <0.01). The use of 3D printed guide plates to place pedicle screws was exactly accurate, and the displacement error and angle error of the actual screw position and the planned screw position showed no signi cant difference (P >0.05). The screws were completely located in the cortical bone, there was no perforation of the pedicle or vertebral anterior margin, and no local microfracture occurred. In vitro tests, 3D-printed guide plates are a viable alternative in scenarios where computer navigation is not available. However, in clinical application, it is impossible for surgeons to completely remove the soft tissue on the surface of the vertebrae, which makes it di cult to accurately x the 3D-printed guide plates in the corresponding positions, resulting in the actual positions of the inserted pedicle screws signi cantly deviated from the planned trajectories. Therefore, the combination of automatic planning system and surgical navigation system may be the best way to make the automatic planning system used in clinical practice.
Biomechanical research also proves that the automatic planning method is effective. The screw trajectory planned according to the CT value obtained a larger screw pull-out force than the traditional manually planned trajectory (P <0.05). It should be noted that although the CT value of the automatically planned trajectory in the decalci cation group was 21.4% higher than that of the manually planned trajectory in the control group, the screw pull-out force was only 97.4% of that in the control group. This may suggest that the BMD of the bone tissue around the screw trajectory also in uences the pull-out force of the screw, which requires further nite element analysis to explain the reason. Despite this, the automatic plan increased the pull force by 44.7% compared with the manual plan in the decalci cation group, which was close to the proportion of improvement in bone cement-reinforced screws (47%) 18 . These results suggest that automatic planning based on CT values is an effective method for pedicle screw placement in osteoporotic vertebrae. This method does not require the placement of additional ller materials and avoids the serious complications caused by leakage of ller materials.
The software can export trajectory coordinates and match them with reconstructed CT images, which provides a basis for future association and matching with surgical navigation systems. In the future, combined with navigation systems or surgical robot technology, optimal trajectory planning and screw placement of the osteoporotic vertebral body can be achieved increasingly quickly, providing a new option for pedicle screw placement of the osteoporotic vertebral body.
In this study, an in vitro demineralized osteoporosis model of a porcine lumbar spine was used, and the evaluation index was only the CT value, without considering microscopic parameters such as bone trabecular structure. The structure of the pig lumbar vertebra is different from that of the human vertebra and EDTA decalci cation simulates not true osteoporosis but more osteomalacia. The applicability of the pedicle screw should be veri ed in the patient's CT and cadaver osteoporosis models. Specimens treated with preservatives inescapability caused changes in the material properties of specimens. Therefore, using fresh specimens to further improve the accuracy of the experimental study is necessary. In the pedicle screw pull-out experiment, we simulated direct violent pull-out, which had limited agreement with the actual pull-out situation in the human body. The pull-out force could be measured again after the cyclic fatigue test. The sample size of the test was small, and subsequent studies with larger sample sizes are needed to con rm its accuracy and effectiveness. Different designs and populations demographics to further investigate the effect of this automatic planning software are also needed.

Conclusion
In the osteoporosis model, automatic planning of the pedicle screw trajectory based on CT value software is safe and effective without causing pedicle screw puncture. Automatic trajectory planning can obtain a higher screw pull-out force, which is 44.7% higher than that of manual planning and even reaches a size similar to that of normal manual planning. This method may be a valuable new method of pedicle screw placement in osteoporotic vertebral bodies.

Availability of data and materials
The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests
The authors declare that they have no competing interests.

Funding
Not applicable.

Authors' contributions
JL constructed the osteoporosis model, measured and analyzed the data, and was a major contributor in writing the manuscript. RZ designed and manufactured 3D-printed guide plates and implanted pedicle screws. WZ made a substantial revision to the work. CL interpreted the data. CZ designed the research. YZ conceived this research. All authors read and approved the nal manuscript.