No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Computational Modeling, Augmented Reality, and Artificial Intelligence in Spine Surgery
Abstract
Over the past decade, advancements in computational modeling, augmented reality, and artificial intelligence (AI) have been driving innovations in spine surgery. Much of the research conducted in these fields is from the past 5 years. In 2021, the market value for augmented reality and virtual reality reached around $22.6 billion, highlighting the rise in demand for these technologies in the medical industry and beyond. Currently, these modalities have a wide variety of potential uses, from preoperative planning of pedicle screw placement and assessment of surgical instrumentation to predictions for postoperative outcomes and development of educational tools. In this chapter, we provide an overview of the applications of these technologies in spine surgery. Furthermore, we discuss several avenues for further development, including integrations between these modalities and areas of improvement for more immersive, informative surgical experiences.
ResearchGate has not been able to resolve any citations for this publication.
From real–time tumor classification to operative outcome prediction, applications of machine learning to neurosurgery are powerful. However, the translation of many of these applications are restricted by the lack of “big data” in neurosurgery. Important restrictions in patient privacy and sharing of imaging data reduce the diversity of the datasets used to train resulting models and therefore limit generalizability. Synthetic learning is a recent development in machine learning that generates synthetic data from real data and uses the synthetic data to train downstream models while preserving patient privacy. Such an approach has yet to be successfully demonstrated in the spine surgery domain. Spine radiographs were collected from the VinDR–SpineXR dataset, with 1470 labeled as abnormal and 2303 labeled as normal. A conditional generative adversarial network (GAN) was trained on the radiographs to generate a spine radiograph and normal/abnormal label. A modified conditional GAN (SpineGAN) was trained on the same task. A convolutional neural network (CNN) was trained using the real data to label abnormal radiographs. A CNN was trained to label abnormal radiographs using synthetic images from the GAN and in a separate experiment from SpineGAN. Using the real radiographs, an AUC of 0.856 was achieved in abnormality classification. Training on synthetic data generated by the standard GAN (AUC of 0.814) and synthetic data generated by our SpineGAN (AUC of 0.830) resulted in similar classifier performance. SpineGAN generated images with higher FID and lower precision scores, but with higher recall and increased performance when used for synthetic learning. The successful application of synthetic learning was demonstrated in the spine surgery domain for the classification of spine radiographs as abnormal or normal. A modified domain–relevant GAN is introduced for the generation of spine images, evidencing the importance of domain–relevant generation techniques in synthetic learning. Synthetic learning can allow neurosurgery to use larger and more diverse patient imaging sets to train more generalizable algorithms with greater patient privacy.
BACKGROUND
Synthetic computed tomography (sCT) can be created from magnetic resonance imaging (MRI) utilizing newer software. sCT is yet to be explored as a possible alternative to routine CT (rCT). In this study, rCT scans and MRI-derived sCT scans were obtained on a cadaver. Morphometric analysis was performed comparing the 2 scans. The ExcelsiusGPS robot was used to place lumbosacral screws with both rCT and sCT images.
OBSERVATIONS
In total, 14 screws were placed. All screws were grade A on the Gertzbein-Robbins scale. The mean surface distance difference between rCT and sCT on a reconstructed software model was –0.02 ± 0.05 mm, the mean absolute surface distance was 0.24 ± 0.05 mm, and the mean absolute error of radiodensity was 92.88 ± 10.53 HU. The overall mean tip distance for the sCT versus rCT was 1.74 ± 1.1 versus 2.36 ± 1.6 mm (p = 0.24); mean tail distance for the sCT versus rCT was 1.93 ± 0.88 versus 2.81 ± 1.03 mm (p = 0.07); and mean angular deviation for the sCT versus rCT was 3.2° ± 2.05° versus 4.04°± 2.71° (p = 0.53).
LESSONS
MRI-based sCT yielded results comparable to those of rCT in both morphometric analysis and robot-assisted lumbosacral screw placement in a cadaver study.
The complexity of patients with spine pathology and high rates of complications has driven extensive research directed toward optimizing outcomes and reducing complications. Traditional statistical analysis has been limited both in validity and in the number of predictor variables considered. Over the past decade, artificial intelligence and machine learning have taken center stage as the possible solution to creating more accurate and applicable patient-centered predictive models in spine surgery. This review discusses the current published machine learning applications on preoperative optimization, risk stratification, and predictive modeling for the cervical, lumbar, and adult spinal deformity populations.
Background:
Measuring intraoperative forces in real time can provide feedback mechanisms to improve patient safety and surgical training. Previous force monitoring has been achieved through the development of specialized and adapted instruments or use designs that are incompatible with neurosurgical workflow.
Objective:
To design a universal sensorised surgical glove to detect intraoperative forces, applicable to any surgical procedure, and any surgical instrument in either hand.
Methods:
We created a sensorised surgical glove that was calibrated across 0 to 10 N. A laboratory experiment demonstrated that the sensorised glove was able to determine instrument-tissue forces. Six expert and 6 novice neurosurgeons completed a validated grape dissection task 20 times consecutively wearing the sensorised glove. The primary outcome was median and maximum force (N).
Results:
The sensorised glove was able to determine instrument-tissue forces reliably. The average force applied by experts (2.14 N) was significantly lower than the average force exerted by novices (7.15 N) (P = .002). The maximum force applied by experts (6.32 N) was also significantly lower than the maximum force exerted by novices (9.80 N) (P = .004). The sensorised surgical glove's introduction to operative workflow was feasible and did not impede on task performance.
Conclusion:
We demonstrate a novel and scalable technique to detect forces during neurosurgery. Force analysis can provide real-time data to optimize intraoperative tissue forces, reduce the risk of tissue injury, and provide objective metrics for training and assessment.
Purpose
Pedicle screw loosening is a frequent complication in lumbar spine fixation, most commonly among patients with poor bone quality. Determining patients at high risk for insufficient implant stability would allow clinicians to adapt the treatment accordingly. The aim of this study was to develop a computational model for quantitative and reliable assessment of the risk of screw loosening.
Methods
A cohort of patient vertebrae with diagnosed screw loosening was juxtaposed to a control group with stable fusion. Imaging data from the two cohorts were used to generate patient-specific biomechanical models of lumbar instrumented vertebral bodies. Single-level finite element models loading the screw in axial or caudo-cranial direction were generated. Further, multi-level models incorporating individualized joint loading were created.
Results
The simulation results indicate that there is no association between screw pull-out strength and the manifestation of implant loosening (p = 0.8). For patient models incorporating multiple instrumented vertebrae, CT-values and stress in the bone were significantly different between loose screws and non-loose screws (p = 0.017 and p = 0.029, for CT-values and stress, respectively). However, very high distinction (p = 0.001) and predictability (R²Pseudo = 0.358, AUC = 0.85) were achieved when considering the relationship between local bone strength and the predicted stress (loading factor). Screws surrounded by bone with a loading factor higher than 25% were likely to be loose, while the chances of screw loosening were close to 0 with a loading factor below 15%.
Conclusion
The use of a biomechanics-based score for risk assessment of implant fixation failure might represent a paradigm shift in addressing screw loosening after spondylodesis surgery.
Objective
Intradural spinal tumors are uncommon and while associations between clinical characteristics and surgical outcomes have been explored, there remains a paucity of literature unifying diverse predictors into an integrated risk model. To predict postresection outcomes for patients with spinal tumors.
Methods
IBM MarketScan Claims Database was queried for adult patients receiving surgery for intradural tumors between 2007 and 2016. Primary outcomes-of-interest were nonhome discharge and 90-day postdischarge readmissions. Secondary outcomes included hospitalization duration and postoperative complications. Risk modeling was developed using a regularized logistic regression framework (LASSO, least absolute shrinkage and selection operator) and validated in a withheld subset.
Results
A total of 5,060 adult patients were included. Most surgeries utilized a posterior approach (n=5,023, 99.3%) and tumors were most commonly found in the thoracic region (n=1,941, 38.4%), followed by the lumbar (n=1,781, 35.2%) and cervical (n=1,294, 25.6%) regions. Compared to models using only tumor-specific or patient-specific features, our integrated models demonstrated better discrimination (area under the curve [AUC] [nonhome discharge] = 0.786; AUC [90-day readmissions] = 0.693) and accuracy (Brier score [nonhome discharge] = 0.155; Brier score [90-day readmissions] = 0.093). Compared to those predicted to be lowest risk, patients predicted to be highest-risk for nonhome discharge required continued care 16.3 times more frequently (64.5% vs. 3.9%). Similarly, patients predicted to be at highest risk for postdischarge readmissions were readmitted 7.3 times as often as those predicted to be at lowest risk (32.6% vs. 4.4%).
Conclusion
Using a diverse set of clinical characteristics spanning tumor-, patient-, and hospitalization-derived data, we developed and validated risk models integrating diverse clinical data for predicting nonhome discharge and postdischarge readmissions.
Study design:
Collectively, seven cadavers were instrumented with 124 thoracolumbar pedicle screws using VisAR augmented reality/guidance. Sixty-five screws were inserted into four donors using open dissection spine surgery. Fifty-nine screws were positioned in 3 donors with a minimally invasive spine surgery (MISS) procedure. For both open and MISS, VisAR was used exclusively for pedicle screw navigation.
Objective:
The objective of this study was to determine the accuracy of pedicle screw placement using VisAR for open spine and MISS procedures.
Summary of background data:
Pedicle screw placement can be challenging depending on anatomical location and a surgeon's experience. AR may minimize fluoroscopy use and speed screw insertion.
Methods:
Prior to CT a series of 4-5 image visible AprilTag optical fiducials were attached to the backs' of the donors. Resulting images were used pre-operatively for planned virtual pedicle screw pathways including entry point, trajectory, and depth. The study was encrypted on a QR code, printed, and viewed in the OR by the surgeon using VisAR (HoloLens 2 headset). Viewing the code wirelessly uploads and launches the study, converting the DICOM data to holographic images which register to the fiducials on the donor's back. The annotated pathways for each pedicle were called up by voice command and the surgeon positioned each screw by aligning with the virtual guidance hologram.
Results:
Overall, 124 pedicle screws were inserted with VisAR navigation with 96% accuracy (Gertzbein-Robbins grades A and B). The combined angle of error was 2.4°and the distance error was 1.9 mm.
Conclusions:
Augmented reality is a highly accurate, emerging technology for navigating both open and minimally invasive spine surgery techniques with off-the-shelf headset hardware.
Simplified loading conditions such as pure moments are frequently used to compare different instrumentation techniques to treat spine disorders. The purpose of this study was to determine if the use of realistic loading conditions such as muscle forces can alter the stresses in the implants with respect to pure moment loading. A musculoskeletal model and a finite element model sharing the same anatomy were built and validated against in vitro data, and coupled in order to drive the finite element model with muscle forces calculated by the musculoskeletal one for a prescribed motion. Intact conditions as well as a L1-L5 posterior fixation with pedicle screws and rods were simulated in flexion-extension and lateral bending. The hardware stresses calculated with the finite element model with instrumentation under simplified and realistic loading conditions were compared. The ROM under simplified loading conditions showed good agreement with in vitro data. As expected, the ROMs between the two types of loading conditions showed relatively small differences. Realistic loading conditions increased the stresses in the pedicle screws and in the posterior rods with respect to simplified loading conditions; an increase of hardware stresses up to 40 MPa in extension for the posterior rods and 57 MPa in flexion for the pedicle screws were observed with respect to simplified loading conditions. This conclusion can be critical for the literature since it means that previous models which used pure moments may have underestimated the stresses in the implants in flexion-extension and in lateral bending.
An optimal surgical strategy for adolescent idiopathic scoliosis (AIS) is to provide maximal deformity correction while preserving spinal mobile segments as much as possible and obtaining a balanced posture. From a spatiotemporal deformity correction standpoint, we recently showed that anatomical four-dimensional (4D) spinal correction could be accomplished by curving the rod. In the surgical procedure, two rods are bent identically to confirm spinal anatomical alignment without referring to the intraoperative alignment of the deformity. Therefore, anatomically designed rods have been developed as notch-free, pre-bent rods for easier anatomical reconstruction. In addition to providing the best spinal instrumentation configurations as pre-bent rods, prediction of surgical outcome along with its biomechanical impact can be obtained by simulation of the surgical procedures with computer modeling. However, an objective model that can simulate the surgical outcome in patients with AIS has not been completely elucidated. The present study aimed to compare simulated deformity corrections based on our newly developed spatiotemporal morphological 4D planning simulation system incorporating pre-bent rods and actual deformity corrections in patients with AIS. A consecutive series of 47 patients who underwent anatomical posterior correction for AIS curves were prospectively evaluated. After multilevel facetectomy, except for the lowest instrumented segment, 11 types of pre-bent rods were used. Patient demographic data, radiographic measurements, and sagittal rod angles were analyzed within 1 week of surgery. Our simulation system incorporating pre-bent rods showed a significant correlation with the actual postoperative spinal alignment. The present study demonstrated the feasibility of our simulation system and the ability to simulate the surgical procedure using the pre-bent rods. The simulation system can be used to minimize the differences between the optimal and possible outcomes related to the instrumentation levels and rod shapes. Preoperative assumption of rod shape and length can contribute to a reduction in operative time which decreases blood loss and risk of infection. The results of the finite element analysis in the simulation system measured for each individual patient would also provide a more realistic representation of the surgical procedures.
Background: Anterior cervical discectomy and fusion (ACDF) sacrifices segmental mobility, which can lead to the acceleration of adjacent segment degeneration. The challenge has promoted cervical artificial disc replacement (CADR) as a substitute for ACDF. However, CADR has revealed a series of new issues that are not found in ACDF, such as hypermobility, subsidence, and wear phenomenon. This study designed a cervical subtotal discectomy prosthesis (CSDP) consisting of a cervical disc prosthesis structure (CDP structure), cervical vertebra fixation structure (CVF structure), link structure, and locking screw, aiming to facilitate motion control and reduce subsidence. The aim of this study was to assess the biomechanics of the CSDP using finite element (FE) analysis, friction-wear test, and non-human primates implantation study.
Study Design: For the FE analysis, based on an intact FE C2-C7 spinal model, a CSDP was implanted at C5-C6 to establish the CSDP FE model and compare it with the Prestige LP prosthesis (Medtronic Sofamor Danek, Minneapolis, MN, United States). The range of motion (ROM), bone-implant interface stress, and facet joint force were calculated under flexion extension, lateral bending, and axial rotation. In addition, CSDP was elevated 1 mm to mimic an improper implantation technique to analyze the biomechanics of CSDP errors in the FE model. Moreover, the friction-wear test was conducted in vitro to research CSDP durability and observe surface wear morphology and total wear volume. Finally, the CSDP was implanted into non-human primates, and its properties were evaluated and verified by radiology.
Results: In the FE analysis, the ROM of the CSDP FE model was close to that of the intact FE model in the operative and adjacent segments. In the operative segment, the CSDP error FE model increased ROM in flexion extension, lateral bending, and axial rotation. The maximum stress in the CSDP FE model was similar to that of the intact FE model and was located in the peripheral cortical bone region. The facet joint force changes were minimal in extension, lateral bending, and axial rotation loads in CSDP. In the friction-wear test, after the 150-W movement simulation, both the CVF-link-junction and the CDP-link-junction had slight wear. In the CSDP non-human primate implantation study, no subsidence, dislocation, or loosening was observed.
Conclusion: In the FE analysis, the biomechanical parameters of the CSDP FE model were relatively close to those of the intact FE model when compared with the Prestige LP FE model. In terms of CSDP error FE models, we demonstrated that the implantation position influences CSDP performance, such as ROM, bone-implant interface stress, and facet joint force. In addition, we performed a friction-wear test on the CSDP to prove its durability. Finally, CSDP studies with non-human primates have shown that the CSDP is effective.
To establish a predictive model for pain response following radiotherapy using a combination of radiomic and clinical features of spinal metastasis. This retrospective study enrolled patients with painful spine metastases who received palliative radiation therapy from 2018 to 2019. Pain response was defined using the International Consensus Criteria. The clinical and radiomic features were extracted from medical records and pre-treatment CT images. Feature selection was performed and a random forests ensemble learning method was used to build a predictive model. Area under the curve (AUC) was used as a predictive performance metric. 69 patients were enrolled with 48 patients showing a response. Random forest models built on the radiomic, clinical, and ‘combined’ features achieved an AUC of 0.824, 0.702, 0.848, respectively. The sensitivity and specificity of the combined features model were 85.4% and 76.2%, at the best diagnostic decision point. We built a pain response model in patients with spinal metastases using a combination of clinical and radiomic features. To the best of our knowledge, we are the first to examine pain response using pre-treatment CT radiomic features. Our model showed the potential to predict patients who respond to radiation therapy.
Scoliosis is a deformity of the spine that in severe cases requires surgical treatment. There is still disagreement among clinicians as to what the aim of such treatment is as well as the optimal surgical technique. Numerical models can aid clinical decision making by estimating the outcome of a given surgical intervention. This paper provided some background information on the modelling of the healthy spine and a review of the literature on scoliotic spine models, their validation, and their application. An overview of the methods and techniques used to construct scoliotic finite element and multibody models was given as well as the boundary conditions used in the simulations. The current limitations of the models were discussed as well as how such limitations are addressed in non-scoliotic spine models. Finally, future directions for the numerical modelling of scoliosis were addressed.
This article is protected by copyright. All rights reserved.
PurposeSurgical treatment of herniated lumbar intervertebral disks is a common procedure worldwide. However, recurrent herniated nucleus pulposus (re-HNP) may develop, complicating outcomes and patient management. The purpose of this study was to utilize machine-learning (ML) analytics to predict lumbar re-HNP, whereby a personalized risk prediction can be developed as a clinical tool.MethodsA retrospective, single center study was conducted of 2630 consecutive patients that underwent lumbar microdiscectomy (mean follow-up: 22-months). Various preoperative patient pain/disability/functional profiles, imaging parameters, and anthropomorphic/demographic metrics were noted. An Extreme Gradient Boost (XGBoost) classifier was implemented to develop a predictive model identifying patients at risk for re-HNP. The model was exported to a web application software for clinical utility.ResultsThere were 1608 males and 1022 females, 114 of whom experienced re-HNP. Primary herniations were central (65.8%), paracentral (17.6%), and far lateral (17.1%). The XGBoost algorithm identified multiple re-HNP predictors and was incorporated into an open-access web application software, identifying patients at low or high risk for re-HNP. Preoperative VAS leg, disability, alignment parameters, elevated body mass index, symptom duration, and age were the strongest predictors.Conclusions
Our predictive modeling via an ML approach of our large-scale cohort is the first study, to our knowledge, that has identified significant risk factors for the development of re-HNP after initial lumbar decompression. We developed the re-herniation after decompression (RAD) profile index that has been translated into an online screening tool to identify low–high risk patients for re-HNP. Additional validation is needed for potential global implementation.
Study Design
Retrospective review.
Objective
To use predictive modeling and machine learning to identify patients at risk for venous thromboembolism (VTE) following posterior lumbar fusion (PLF) for degenerative spinal pathology.
Methods
Patients undergoing single-level PLF in the inpatient setting were identified in the National Surgical Quality Improvement Program database. Our outcome measure of VTE included all patients who experienced a pulmonary embolism and/or deep venous thrombosis within 30-days of surgery. Two different methodologies were used to identify VTE risk: 1) a novel predictive model derived from multivariable logistic regression of significant risk factors, and 2) a tree-based extreme gradient boosting (XGBoost) algorithm using preoperative variables. The methods were compared against legacy risk-stratification measures: ASA and Charlson Comorbidity Index (CCI) using area-under-the-curve (AUC) statistic.
Results
13, 500 patients who underwent single-level PLF met the study criteria. Of these, 0.95% had a VTE within 30-days of surgery. The 5 clinical variables found to be significant in the multivariable predictive model were: age > 65, obesity grade II or above, coronary artery disease, functional status, and prolonged operative time. The predictive model exhibited an AUC of 0.716, which was significantly higher than the AUCs of ASA and CCI (all, P < 0.001), and comparable to that of the XGBoost algorithm ( P > 0.05).
Conclusion
Predictive analytics and machine learning can be leveraged to aid in identification of patients at risk of VTE following PLF. Surgeons and perioperative teams may find these tools useful to augment clinical decision making risk stratification tool.
Background: Spinopelvic fractures and approaches of operative stabilization have been a source of controversial discussion. Biomechanical data support the benefit of a spinopelvic stabilization and minimally invasive procedures help to reduce the dissatisfying complication rate. The role of a cross connector within spinopelvic devices remains inconclusive. We aimed to analyze the effect of a cross connector in a finite element model (FE model).
Study Design: A FE model of the L1-L5 spine segment with pelvis and a spinopelvic stabilization was reconstructed from patient-specific CT images. The biomechanical relevance of a cross connector in a Denis zone I (AO: 61-B2) sacrum fracture was assessed in the FE model by applying bending and twisting forces with and without a cross connector. Biomechanical outcomes from the numerical model were investigated also considering uncertainties in material properties and levels of osseointegration.
Results: The designed FE model showed comparable values in range-of-motion (ROM) and stresses with reference to the literature. The superiority of the spinopelvic stabilization (L5/Os ilium) ± cross connector compared to a non-operative procedure was confirmed in all analyzed loading conditions by reduced ROM and principal stresses in the disk L5/S1, vertebral body L5 and the fracture area. By considering the combination of all loading cases, the presence of a cross connector reduced the maximum stresses in the fracture area of around 10%. This difference has been statistically validated (p < 0.0001).
Conclusion: The implementation of a spinopelvic stabilization (L5/Os ilium) in sacrum fractures sustained the fracture and led to enhanced biomechanical properties compared to a non-reductive procedure. While the additional cross connector did not alter the resulting ROM in L4/L5 or L5/sacrum, the reduction of the maximum stresses in the fracture area was significant.
While spinal fusion using rigid rods remains the gold standard treatment modality for various lumbar degenerative conditions, its adverse effects, including accelerated adjacent segment disease (ASD), are well known. In order to better understand the performance of semirigid constructs using polyetheretherketone (PEEK) in fixation surgeries, the objective of this study was to analyze the biomechanical performance of PEEK versus Ti rods using a geometrically patient-specific poroelastic finite element (FE) analyses. Ten subject-specific preoperative models were developed, and the validity of the models was evaluated with previous studies. Furthermore, FE models of those lumbar spines were regenerated based on postoperation images for posterolateral fixation at the L4–L5 level. Biomechanical responses for instrumented and adjacent intervertebral discs (IVDs) were analyzed and compared subjected to static and cyclic loading. The preoperative model results were well comparable with previous FE studies. The PEEK construct demonstrated a slightly increased range of motion (ROM) at the instrumented level, but decreased ROM at adjacent levels, as compared with the Ti. However, no significant changes were detected during axial rotation. During cyclic loading, disc height loss, fluid loss, axial stress, and collagen fiber strain in the adjacent IVDs were higher for the Ti construct when compared with the intact and PEEK models. Increased ROM, experienced stress in AF, and fiber strain at adjacent levels were observed for the Ti rod group compared with the intact and PEEK rod group, which can indicate the risk of ASD for rigid fixation. Similar to the aforementioned pattern, disc height loss and fluid loss were significantly higher at adjacent levels in the Ti rod group after cycling loading which alter the fluid–solid interaction of the adjacent IVDs. This phenomenon debilitates the damping quality, which results in disc disability in absorbing stress. Such finding may suggest the advantage of using a semirigid fixation system to decrease the chance of ASD.
Objective
To determine whether machine learning based on conventional magnetic resonance imaging (MRI) sequences have the potential for the differential diagnosis of multiple myeloma (MM), and different tumor metastasis lesions of the lumbar vertebra.
Methods
We retrospectively enrolled 107 patients newly diagnosed with MM and different metastasis of the lumbar vertebra. In total 60 MM lesions and 118 metastasis lesions were selected for training classifiers (70%) and subsequent validation (30%). Following segmentation, 282 texture features were extracted from both T1WI and T2WI images. Following regression analysis using the least absolute shrinkage and selection operator (LASSO) algorithm, the following machine learning models were selected: Support‐Vector Machine (SVM), K-Nearest Neighbor (KNN), Random Forest (RF), Artificial Neural Networks (ANN), and Naïve Bayes (NB) using 10-fold cross validation, and the performances were evaluated using a confusion matrix. Matthews correlation coefficient (MCC), sensitivity, specificity, and accuracy of the models were also calculated.
Results
To differentiate MM and metastasis, 13 features in the T1WI images and 9 features in the T2WI images were obtained. Among the 10 classifiers, the ANN classifier from the T2WI images achieved the best performance (MCC = 0.605) with accuracy, sensitivity, and specificity of 0.815, 0.879, and 0.790, respectively, in the validation cohort. To differentiate MM and metastasis subtypes, eight features in the T1WI images and seven features in the T2WI images were obtained. Among the 10 classifiers, the ANN classifier from the T2WI images achieved the best performance (MCC = 0.560, 0.412, 0.449), respectively, with accuracy = 0.648; sensitivity 0.714, 0.821, 0.897 and specificity 0.775, 0.600, 0.640 for the MM, lung, and other metastases, respectively, in the validation cohort.
Conclusions
Machine learning–based classifiers showed a satisfactory performance in differentiating MM lesions from those of tumor metastasis. While their value for distinguishing myeloma from different metastasis subtypes was moderate.
Study design:
Biomechanical study.
Purpose:
To quantitatively investigate the effect of screw size on screw fixation in osteoporotic vertebrae with finite element analysis (FEA).
Overview of literature:
Osteoporosis poses a challenge in spinal instrumentation; however, the selection of screw size is directly related to fixation and is closely dependent on each surgeon's experience and preference.
Methods:
Total 1,200 nonlinear FEA with various screw diameters (4.5-7.5 mm) and lengths (30-50 mm) were performed on 25 patients (seven men and 18 women; mean age, 75.2±10.8 years) with osteoporosis. The axial pullout strength, and the vertebral fixation strength of a paired-screw construct against flexion, extension, lateral bending, and axial rotation were examined. Thereafter, we calculated the equivalent stress of the bone-screw interface during nondestructive loading. Then, using diameter parameters (screw diameter or screw fitness in the pedicle [%fill]), and length parameters (screw length or screw depth in the vertebral body [%length]), multiple regression analyses were performed in order to evaluate the factors affecting various fixations.
Results:
Larger diameter and longer screws significantly increased the pullout strength and vertebral fixation strength; further, they decreased the equivalent stress around the screws. Multiple regression analyses showed that the actual screw diameter and %length were factors that had a stronger effect on the fixation strength than %fill and the actual screw length. Screw diameter had a greater effect on the resistance to screw pullout and flexion and extension loading (β =0.38-0.43, p <0.01); while the %length had a greater effect on resistance to lateral bending and axial rotation loading (β =0.25-0.36, p <0.01) as well as mechanical stress of the bone-screw interface (β =-0.42, p <0.01).
Conclusions:
The screw size should be determined based on the biomechanical behavior of the screws, type of mechanical force applied on the corresponding vertebra, and anatomical limitations.
Study design:
Experimental in-vivo animal study.
Objective:
The aim of this study was to evaluate an Artificial Intelligence (AI)-enabled ultrasound imaging system's ability to detect, segment, classify, and display neural and other structures during trans-psoas spine surgery.
Summary of background data:
Current methodologies for intraoperatively localizing and visualizing neural structures within the psoas are limited and can impact the safety of lateral lumbar interbody fusion (LLIF). Ultrasound technology, enhanced with AI-derived neural detection algorithms, could prove useful for this task.
Methods:
The study was conducted using an in vivo porcine model (50 subjects). Image processing and machine learning algorithms were developed to detect neural and other anatomic structures within and adjacent to the psoas muscle while using an ultrasound imaging system during lateral lumbar spine surgery (SonoVision,™ Tissue Differentiation Intelligence, USA). The imaging system's ability to detect and classify the anatomic structures was assessed with subsequent tissue dissection. Dice coefficients were calculated to quantify the performance of the image segmentation.
Results:
The AI-trained ultrasound system detected, segmented, classified, and displayed nerve, psoas muscle, and vertebral body surface with high sensitivity and specificity. The mean Dice coefficient score for each tissue type was >80%, indicating that the detected region and ground truth were >80% similar to each other. The mean specificity of nerve detection was 92%; for bone and muscle, it was >95%. The accuracy of nerve detection was >95%.
Conclusion:
This study demonstrates that a combination of AI-derived image processing and machine learning algorithms can be developed to enable real-time ultrasonic detection, segmentation, classification, and display of critical anatomic structures, including neural tissue, during spine surgery. AI-enhanced ultrasound imaging can provide a visual map of important anatomy in and adjacent to the psoas, thereby providing the surgeon with critical information intended to increase the safety of LLIF surgery.Level of Evidence: N/A.
This review aims to identify the role of augmented, virtual or mixed reality (AR, VR or MR) technologies in setting of spinal surgery. The authors address the challenges surrounding the implementation of this technology in the operating room. A technical standpoint addresses the efficacy of these imaging modalities based on the current literature in the field. Ultimately, these technologies must be cost-effective to ensure widespread adoption. This may be achieved through reduced surgical times and decreased incidence of post-operative complications and revisions while maintaining equivalent safety profile to alternative surgical approaches. While current studies focus mainly on the successful placement of pedicle screws via AR-guided instrumentation, a wider scope of procedures may be assisted using AR, VR or MR technology once efficacy and safety have been validated. These emerging technologies offer a significant advantage in the guidance of complex procedures that require high precision and accuracy using minimally invasive interventions.
Background:
To develop and validate an MRI-based radiomics nomogram for differentiation of cervical spine ORN from metastasis after radiotherapy (RT) in nasopharyngeal carcinoma (NPC).
Methods:
A radiomics nomogram was developed in a training set that comprised 46 NPC patients after RT with 95 cervical spine lesions (ORN, n = 51; metastasis, n = 44), and data were gathered from January 2008 to December 2012. 279 radiomics features were extracted from the axial contrast-enhanced T1-weighted image (CE-T1WI). A radiomics signature was created by using the least absolute shrinkage and selection operator (LASSO) algorithm. A nomogram model was developed based on the radiomics scores. The performance of the nomogram was determined in terms of its discrimination, calibration, and clinical utility. An independent validation set contained 25 consecutive patients with 47 lesions (ORN, n = 25; metastasis, n = 22) from January 2013 to December 2015.
Results:
The radiomics signature that comprised eight selected features was significantly associated with the differentiation of cervical spine ORN and metastasis. The nomogram model demonstrated good calibration and discrimination in the training set [AUC, 0.725; 95% confidence interval (CI), 0.622-0.828] and the validation set (AUC, 0.720; 95% CI, 0.573-0.867). The decision curve analysis indicated that the radiomics nomogram was clinically useful.
Conclusions:
MRI-based radiomics nomogram shows potential value to differentiate cervical spine ORN from metastasis after RT in NPC.
Study Design
A prospective, case-based, observational study.
Objectives
To investigate how microscope-based augmented reality (AR) support can be utilized in various types of spine surgery.
Methods
In 42 spinal procedures (12 intra- and 8 extradural tumors, 7 other intradural lesions, 11 degenerative cases, 2 infections, and 2 deformities) AR was implemented using operating microscope head-up displays (HUDs). Intraoperative low-dose computed tomography was used for automatic registration. Nonlinear image registration was applied to integrate multimodality preoperative images. Target and risk structures displayed by AR were defined in preoperative images by automatic anatomical mapping and additional manual segmentation.
Results
AR could be successfully applied in all 42 cases. Low-dose protocols ensured a low radiation exposure for registration scanning (effective dose cervical 0.29 ± 0.17 mSv, thoracic 3.40 ± 2.38 mSv, lumbar 3.05 ± 0.89 mSv). A low registration error (0.87 ± 0.28 mm) resulted in a reliable AR representation with a close matching of visualized objects and reality, distinctly supporting anatomical orientation in the surgical field. Flexible AR visualization applying either the microscope HUD or video superimposition, including the ability to selectively activate objects of interest, as well as different display modes allowed a smooth integration in the surgical workflow, without disturbing the actual procedure. On average, 7.1 ± 4.6 objects were displayed visualizing target and risk structures reliably.
Conclusions
Microscope-based AR can be applied successfully to various kinds of spinal procedures. AR improves anatomical orientation in the surgical field supporting the surgeon, as well as it offers a potential tool for education.
Instrumented spine procedures have been performed for decades to treat a wide variety of spinal disorders. New technologies have been employed to obtain a high degree of precision, to minimize risks of damage to neurovascular structures and to diminish harmful exposure of patients and the operative team to ionizing radiations. Robotic spine surgery comprehends 3 major categories: telesurgical robotic systems, robotic-assisted navigation (RAN) and virtual augmented reality (AR) systems, including AR and virtual reality. Telesurgical systems encompass devices that can be operated from a remote command station, allowing to perform surgery via instruments being manipulated by the robot. On the other hand, RAN technologies are characterized by the robotic guidance of surgeon-operated instruments based on real-time imaging. Virtual AR systems are able to show images directly on special visors and screens allowing the surgeon to visualize information about the patient and the procedure (i.e., anatomical landmarks, screw direction and inclination, distance from neurological and vascular structures etc.). The aim of this review is to focus on the current state of the art of robotics and AR in spine surgery and perspectives of these emerging technologies that hold promises for future applications.
Background: With the emergence of competency-based training, the current evaluation scheme of surgical skills is evolving to include newer methods of assessment and training. Artificial intelligence through machine learning algorithms can utilize extensive data sets to analyze operator performance. This study aimed to address 3 questions: (1) Can artificial intelligence uncover novel metrics of surgical performance? (2) Can support vector machine algorithms be trained to differentiate “senior” and “junior” participants who are executing a virtual reality hemilaminectomy? (3) Can other algorithms achieve a good classification performance?
Methods: Participants from 4 Canadian universities were divided into 2 groups according to their training level (senior and junior) and were asked to perform a virtual reality hemilaminectomy. The position, angle, and force application of the simulated burr and suction instruments, along with tissue volumes that were removed, were recorded at 20-ms intervals. Raw data were manipulated to create metrics to train machine learning algorithms. Five algorithms, including a support vector machine, were trained to predict whether the task was performed by a senior or junior participant. The accuracy of each algorithm was assessed through leave-one-out cross-validation.
Results: Forty-one individuals were enrolled (22 senior and 19 junior participants). Twelve metrics related to safety of the procedure, efficiency, motion of the tools, and coordination were selected. Following cross-validation, the support vector machine achieved a 97.6% accuracy. The other algorithms achieved accuracy of 92.7%, 87.8%, 70.7%, and 65.9%, respectively.
Conclusions: Artificial intelligence defined novel metrics of surgical performance and outlined training levels in a virtual reality spinal simulation procedure.
Clinical Relevance: The significance of these results lies in the potential of artificial intelligence to complement current educational paradigms and better prepare residents for surgical procedures.
Background:
To compare the biomechanical characteristics of different posterior intermediate screw fixation techniques (ISFTs) with hybrid monoaxial pedicle screws (Mps) and polyaxial pedicle screws (Pps) used in thoracolumbar burst fractures.
Methods:
Fixation techniques are compared with regard to the von Mises stress (VMS) of the instrumentations and intradiscal pressures (IDPs) of the adjacent segments by finite element method (FEM).
Results:
The redistributed ROM of the fixation models with Pps fixed at the lowest segment was twice of the other fixation models in flexion and extension. The largest value of maximal VMS of a pedicle screw was located at the lowest pedicle screws when Mps are fixed at the lowest segment. The largest value of maximal VMS of the rods was decreased when more Pps are fixed at the models. Maximal IDPs of the upper adjacent segments were all larger than those of the lower adjacent segments. The maximal IDPs of the fixation model with MPs fixed at the lowest segment were larger than the other fixation models in flexion and extension.
Conclusions:
Polyaxial pedicle screws could be placed at the upper or the median segment for the facilitated efficient application of the connecting rod. We should focus on the adjacent segmental degeneration especially the upper adjacent segment in the fixation model with Mps fixed at the lowest segment.
Degenerative cervical myelopathy (DCM) is a spinal cord condition that results in progressive non-traumatic compression of the cervical spinal cord. Spine surgeons must consider a large quantity of information relating to disease presentation, imaging features, and patient characteristics to determine if a patient will benefit from surgery for DCM. We applied a supervised machine learning approach to develop a classification model to predict individual patient outcome after surgery for DCM. Patients undergoing surgery for DCM as a part of the AOSpine CSM-NA or CSM-I prospective, multi-centre studies were included in the analysis. Out of 757 patients 605, 583, and 539 patients had complete follow-up information at 6, 12, and 24 months respectively and were included in the analysis. The primary outcome was improvement in the SF-6D quality of life indicator score by the minimum clinically important difference (MCID). The secondary outcome was improvement in the modified Japanese Orthopedic Association (mJOA) score by the MCID. Predictor variables reflected information about pre-operative disease severity, disease presentation, patient demographics, and comorbidities. A machine learning approach of feature engineering, data pre-processing, and model optimization was used to create the most accurate predictive model of outcome after surgery for DCM. Following data pre-processing 48, 108, and 101 features were chosen for model training at 6, 12, and 24 months respectively. The best performing predictive model used a random forest structure and had an average area under the curve (AUC) of 0.70, classification accuracy of 77%, and sensitivity of 78% when evaluated on a testing cohort that was not used for model training. Worse pre-operative disease severity, longer duration of DCM symptoms, older age, higher body weight, and current smoking status were associated with worse surgical outcomes. We developed a model that predicted positive surgical outcome for DCM with good accuracy at the individual patient level on an independent testing cohort. Our analysis demonstrates the applicability of machine-learning to predictive modeling in spine surgery.
Study design:
Prospective observational study.
Objective:
The aim of this study was to evaluate the accuracy of pedicle screw placement using augmented reality surgical navigation (ARSN) in a clinical trial.
Summary of background data:
Recent cadaveric studies have shown improved accuracy for pedicle screw placement in the thoracic spine using ARSN with intraoperative 3D imaging, without the need for periprocedural x-ray. In this clinical study, we used the same system to place pedicle screws in the thoracic and lumbosacral spine of 20 patients.
Methods:
The study was performed in a hybrid operating room with an integrated ARSN system encompassing a surgical table, a motorized flat detector C-arm with intraoperative 2D/3D capabilities, integrated optical cameras for augmented reality navigation, and noninvasive patient motion tracking. Three independent reviewers assessed screw placement accuracy using the Gertzbein grading on 3D scans obtained before wound closure. In addition, the navigation time per screw placement was measured.
Results:
One orthopedic spinal surgeon placed 253 lumbosacral and thoracic pedicle screws on 20 consenting patients scheduled for spinal fixation surgery. An overall accuracy of 94.1% of primarily thoracic pedicle screws was achieved. No screws were deemed severely misplaced (Gertzbein grade 3). Fifteen (5.9%) screws had 2 to 4 mm breach (Gertzbein grade 2), occurring in scoliosis patients only. Thirteen of those 15 screws were larger than the pedicle in which they were placed. Two medial breaches were observed and 13 were lateral. Thirteen of the grade 2 breaches were in the thoracic spine. The average screw placement time was 5.2 ± 4.1 minutes. During the study, no device-related adverse event occurred.
Conclusion:
ARSN can be clinically used to place thoracic and lumbosacral pedicle screws with high accuracy and with acceptable navigation time. Consequently, the risk for revision surgery and complications could be minimized.
Level of evidence:
3.
Objective:
The aim of this review was to summarize major topics in artificial intelligence (AI), including their applications and limitations in surgery. This paper reviews the key capabilities of AI to help surgeons understand and critically evaluate new AI applications and to contribute to new developments.
Summary background data:
AI is composed of various subfields that each provide potential solutions to clinical problems. Each of the core subfields of AI reviewed in this piece has also been used in other industries such as the autonomous car, social networks, and deep learning computers.
Methods:
A review of AI papers across computer science, statistics, and medical sources was conducted to identify key concepts and techniques within AI that are driving innovation across industries, including surgery. Limitations and challenges of working with AI were also reviewed.
Results:
Four main subfields of AI were defined: (1) machine learning, (2) artificial neural networks, (3) natural language processing, and (4) computer vision. Their current and future applications to surgical practice were introduced, including big data analytics and clinical decision support systems. The implications of AI for surgeons and the role of surgeons in advancing the technology to optimize clinical effectiveness were discussed.
Conclusions:
Surgeons are well positioned to help integrate AI into modern practice. Surgeons should partner with data scientists to capture data across phases of care and to provide clinical context, for AI has the potential to revolutionize the way surgery is taught and practiced with the promise of a future optimized for the highest quality patient care.
Study design:
Cross-sectional database study.
Objective:
To train and validate machine learning models to identify risk factors for complications following posterior lumbar spine fusion.
Summary of background data:
Machine learning models such as artificial neural networks (ANNs) are valuable tools for analyzing and interpreting large and complex datasets. ANNs have yet to be used for risk factor analysis in orthopedic surgery.
Methods:
The American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database was queried for patients who underwent posterior lumbar spine fusion. This query returned 22,629 patients, 70% of whom were used to train our models, and 30% were used to evaluate the models. The predictive variables used included sex, age, ethnicity, diabetes, smoking, steroid use, coagulopathy, functional status, American society for anesthesiology (ASA) class ≥ 3, body mass index (BMI), pulmonary comorbidities, and cardiac comorbidities. The models were used to predict cardiac complications, wound complications, venous thromboembolism (VTE), and mortality. Using ASA class as a benchmark for prediction, area under receiver operating curves (AUC) was used to determine the accuracy of our machine learning models.
Results:
Based on AUC values, ANN and LR both outperformed ASA class for predicting all four types of complications. ANN was the most accurate for predicting cardiac complications, and LR was most accurate for predicting wound complications, VTE, and mortality, though ANN and LR had comparable AUC values for predicting all types of complications. ANN had greater sensitivity than LR for detecting wound complications and mortality.
Conclusions:
Machine learning in the form of logistic regression and artificial neural networks were more accurate than benchmark ASA scores for identifying risk factors of developing complications following posterior lumbar spine fusion, suggesting they are potentially great tools for risk factor analysis in spine surgery.
Level of evidence:
3.
Purpose:
We present a novel augmented reality (AR) surgical navigation system based on ultrasound-assisted registration for pedicle screw placement. This system provides the clinically desired targeting accuracy and reduces radiation exposure.
Methods:
Ultrasound (US) is used to perform registration between preoperative computed tomography (CT) images and patient, and the registration is performed by least-squares fitting of these two three-dimensional (3D) point sets of anatomical landmarks taken from US and CT images. An integral videography overlay device is calibrated to accurately display naked-eye 3D images for surgical navigation. We use a 3.0-mm Kirschner wire (K-wire) instead of a pedicle screw in this study, and the K-wire is calibrated to obtain its orientation and tip location. Based on the above registration and calibration, naked-eye 3D images of the planning path and the spine are superimposed onto patient in situ using our AR navigation system. Simultaneously, a 3D image of the K-wire is overlaid accurately on the real one to guide the insertion procedure. The targeting accuracy is evaluated postoperatively by performing a CT scan.
Results:
An agar phantom experiment was performed. Eight K-wires were inserted successfully after US-assisted registration, and the mean targeting error and angle error were 3.35 mm and [Formula: see text], respectively. Furthermore, an additional sheep cadaver experiment was performed. Four K-wires were inserted successfully. The mean targeting error was 3.79 mm and the mean angle error was [Formula: see text], and US-assisted registration yielded better targeting results than skin markers-based registration (targeting errors: 2.41 vs. 5.18 mm, angle errors: [Formula: see text] vs. [Formula: see text].
Conclusion:
Experimental outcomes demonstrate that the proposed navigation system has acceptable targeting accuracy. In particular, the proposed navigation method reduces repeated radiation exposure to the patient and surgeons. Therefore, it has promising prospects for clinical use.
Study Design
A cadaveric laboratory study.
Objective
The aim of this study was to assess the feasibility and accuracy of thoracic pedicle screw placement using augmented reality surgical navigation (ARSN).
Summary of Background Data
Recent advances in spinal navigation have shown improved accuracy in lumbosacral pedicle screw placement but limited benefits in the thoracic spine. 3D intraoperative imaging and instrument navigation may allow improved accuracy in pedicle screw placement, without the use of x-ray fluoroscopy, and thus opens the route to image-guided minimally invasive therapy in the thoracic spine.
Methods
ARSN encompasses a surgical table, a motorized flat detector C-arm with intraoperative 2D/3D capabilities, integrated optical cameras for augmented reality navigation, and noninvasive patient motion tracking. Two neurosurgeons placed 94 pedicle screws in the thoracic spine of four cadavers using ARSN on one side of the spine (47 screws) and free-hand technique on the contralateral side. X-ray fluoroscopy was not used for either technique. Four independent reviewers assessed the postoperative scans, using the Gertzbein grading. Morphometric measurements of the pedicles axial and sagittal widths and angles, as well as the vertebrae axial and sagittal rotations were performed to identify risk factors for breaches.
Results
ARSN was feasible and superior to free-hand technique with respect to overall accuracy (85% vs. 64%, P < 0.05), specifically significant increases of perfectly placed screws (51% vs. 30%, P < 0.05) and reductions in breaches beyond 4 mm (2% vs. 25%, P < 0.05). All morphometric dimensions, except for vertebral body axial rotation, were risk factors for larger breaches when performed with the free-hand method.
Conclusion
ARSN without fluoroscopy was feasible and demonstrated higher accuracy than free-hand technique for thoracic pedicle screw placement.
Level of Evidence: N/A
Augmented reality (AR) has emerged as a potential surgical adjunct in spine surgery. We review key developments in surgical AR and detail the technological milestones foundational for AR applications in spine surgery. We evaluate the studies that have analyzed AR systems in spine surgery and discuss the quality of these investigations. Finally, we outline a path forward for better integration of AR into the spine surgery operating room, imagine applications of AR to improve spine surgical education, and hypothesize future applications of AR in spine surgery.
Background and objectives:
Robot-assisted pedicle screw placement is associated with greater accuracy, reduced radiation, less blood loss, shorter hospital stays, and fewer complications than freehand screw placement. However, it can be associated with longer operative times and an extended training period. We report the initial experience of a surgeon using a robot system at an academic medical center.
Methods:
We retrospectively reviewed all patients undergoing robot-assisted pedicle screw placement at a single tertiary care institution by 1 surgeon from 10/2017 to 05/2022. Linear regression, analysis of variance, and cumulative sum analysis were used to evaluate operative time learning curves. Operative time subanalyses for surgery indication, number of levels, and experience level were performed.
Results:
In total, 234 cases were analyzed. A significant 0.19-minute decrease in operative time per case was observed (r = 0.14, P = .03). After 234 operations, this translates to a reduction in 44.5 minutes from the first to last case. A linear relationship was observed between case number and operative time in patients with spondylolisthesis (-0.63 minutes/case, r = 0.41, P < .001), 2-level involvement (-0.35 minutes/case, r = 0.19, P = .05), and 4-or-more-level involvement (-1.29 minutes/case, r = 0.24, P = .05). This resulted in reductions in operative time ranging from 39 minutes to 1.5 hours. Continued reductions in operative time were observed across the learning, experienced, and expert phases, which had mean operative times of 214, 197, and 146 minutes, respectively (P < .001). General proficiency in robot-assisted surgery was observed after the 20th case. However, 67 cases were required to reach mastery, defined as the inflection point of the cumulative sum curve.
Conclusion:
This study documents the long-term learning curve of a fellowship-trained spine neurosurgeon. Operative time significantly decreased with more experience. Although gaining comfort with robotic systems may be challenging or require additional training, it can benefit surgeons and patients alike with continued reductions in operative time.
OBJECTIVE
Damage to the thoracolumbar spine can confer significant morbidity and mortality. The Thoracolumbar Injury Classification and Severity Score (TLICS) is used to categorize injuries and determine patients at risk of spinal instability for whom surgical intervention is warranted. However, calculating this score can constitute a bottleneck in triaging and treating patients, as it relies on multiple imaging studies and a neurological examination. Therefore, the authors sought to develop and validate a deep learning model that can automatically categorize vertebral morphology and determine posterior ligamentous complex (PLC) integrity, two critical features of TLICS, using only CT scans.
METHODS
All patients who underwent neurosurgical consultation for traumatic spine injury or degenerative pathology resulting in spine injury at a single tertiary center from January 2018 to December 2019 were retrospectively evaluated for inclusion. The morphology of injury and integrity of the PLC were categorized on CT scans. A state-of-the-art object detection region-based convolutional neural network (R-CNN), Faster R-CNN, was leveraged to predict both vertebral locations and the corresponding TLICS. The network was trained with patient CT scans, manually labeled vertebral bounding boxes, TLICS morphology, and PLC annotations, thus allowing the model to output the location of vertebrae, categorize their morphology, and determine the status of PLC integrity.
RESULTS
A total of 111 patients were included (mean ± SD age 62 ± 20 years) with a total of 129 separate injury classifications. Vertebral localization and PLC integrity classification achieved Dice scores of 0.92 and 0.88, respectively. Binary classification between noninjured and injured morphological scores demonstrated 95.1% accuracy. TLICS morphology accuracy, the true positive rate, and positive injury mismatch classification rate were 86.3%, 76.2%, and 22.7%, respectively. Classification accuracy between no injury and suspected PLC injury was 86.8%, while true positive, false negative, and false positive rates were 90.0%, 10.0%, and 21.8%, respectively.
CONCLUSIONS
In this study, the authors demonstrate a novel deep learning method to automatically predict injury morphology and PLC disruption with high accuracy. This model may streamline and improve diagnostic decision support for patients with thoracolumbar spinal trauma.
Objective: Augmented reality (AR) is a novel technology which, when applied to spine surgery, offers the potential for efficient, safe, and accurate placement of spinal instrumentation. The authors report the accuracy of the first 205 pedicle screws consecutively placed at their institution by using AR assistance with a unique head-mounted display (HMD) navigation system.
Methods: A retrospective review was performed of the first 28 consecutive patients who underwent AR-assisted pedicle screw placement in the thoracic, lumbar, and/or sacral spine at the authors' institution. Clinical accuracy for each pedicle screw was graded using the Gertzbein-Robbins scale by an independent neuroradiologist working in a blinded fashion.
Results: Twenty-eight consecutive patients underwent thoracic, lumbar, or sacral pedicle screw placement with AR assistance. The median age at the time of surgery was 62.5 (IQR 13.8) years and the median body mass index was 31 (IQR 8.6) kg/m2. Indications for surgery included degenerative disease (n = 12, 43%); deformity correction (n = 12, 43%); tumor (n = 3, 11%); and trauma (n = 1, 4%). The majority of patients (n = 26, 93%) presented with low-back pain, 19 (68%) patients presented with radicular leg pain, and 10 (36%) patients had documented lower extremity weakness. A total of 205 screws were consecutively placed, with 112 (55%) placed in the lumbar spine, 67 (33%) in the thoracic spine, and 26 (13%) at S1. Screw placement accuracy was 98.5% for thoracic screws, 97.8% for lumbar/S1 screws, and 98.0% overall.
Conclusions: AR depicted through a unique HMD is a novel and clinically accurate technology for the navigated insertion of pedicle screws. The authors describe the first 205 AR-assisted thoracic, lumbar, and sacral pedicle screws consecutively placed at their institution with an accuracy of 98.0% as determined by a Gertzbein-Robbins grade of A or B.
Background:
Advancing age and degeneration frequently lead to low back pain, which is the most prevalent musculoskeletal disorder worldwide. Degenerative changes in intervertebral discs and musculo-ligamentous incapacity to compensate sagittal imbalance are typically amongst the sources of instability, with spinal fusion techniques being the main treatment options to relieve pain. The aims of this work were to: (i) assess the link between ligament degeneration and spinal instability by determining the role of each ligament per movement, (ii) evaluate the impact of disc height reduction in degenerative changes, and (iii) unveil the most advantageous type of posterior fixation in Oblique Lumbar Interbody Fusion to prevent adjacent disc degeneration.
Methods:
Two L3-L5 finite element models were developed, being the first in healthy condition and the second having reduced L4-L5 height. Different degrees of degeneration were tested, combined with different fixation configurations for Oblique Lumbar Interbody Fusion.
Findings:
Facet capsular ligament and anterior longitudinal ligament were the most influential ligaments for spinal stability, particularly with increasing degeneration and disc height reduction. Pre-existent degeneration had lower influence than the fusion procedure for the risk of adjacent disc degeneration, being the highest stability and minimal degeneration achieved with bilateral fixation. Right unilateral fixation was more suited to reduce disc stress than left unilateral fixation.
Interpretation:
Bilateral fixation is the best option to stabilize the spinal segment, but unilateral right fixation may suffice. This has direct implications for clinical practice, and the extension to a population-based study will allow for more efficient fusion surgeries.
Augmented reality (AR) navigation refers to novel technologies that superimpose images, such as radiographs and navigation pathways, onto a view of the operative field. The development of AR navigation has focused on improving the safety and efficacy of neurosurgical and orthopedic procedures. In this review, the authors focus on 3 types of AR technology used in spine surgery: AR surgical navigation, microscope-mediated heads-up display, and AR head-mounted displays. Microscope AR and head-mounted displays offer the advantage of reducing attention shift and line-of-sight interruptions inherent in traditional navigation systems. With the U.S. Food and Drug Administration’s recent clearance of the XVision AR system (Augmedics, Arlington Heights, IL), the adoption and refinement of AR technology by spine surgeons will only accelerate.
Objective
Readmission after spine surgery is a costly, but relatively common occurrence. Previous research has identified several risk factors for readmission however, the conclusions remain equivocal. Machine learning algorithms offer a unique perspective in the analysis of risk factors for readmission and can help predict the likelihood of this occurrence. In this investigation, a neural network, a supervised machine learning technique, is evaluated to determine whether it can predict readmission after three lumbar fusion procedures.
Methods
The American College of Surgeon’s database, the National Surgical Quality Improvement Program (NSQIP), was queried between 2009 and 2018. Patients who had undergone anterior, lateral, and/or posterior lumbar fusion were included in the study. The Python Sci-Kit Learn package was utilized to run the neural network algorithms. A multivariate regression was performed to determine risk factors for readmission.
Results
In total, 63,533 patients were analyzed (12,915 ALIF, 27,212 PLIF, and 23,406 PSF). The neural network algorithm was able to successful predict 30-day readmission for 94.6% of ALIF, 94.0% of PLIF, and 92.6% of PSF cases with AUC values of between 0.64-0.65. The multivariate regression indicated that age > 65 years and ASA > 2 were linked to increased risk for readmission for all three procedures.
Conclusion
The accurate metrics presented here indicate the capability for neural network algorithms to predict readmission after lumbar arthrodesis. Moreover, the results of this study serve as a catalyst for further research into the utility of machine learning in spine surgery.
BACKGROUND
Augmented reality mediated spine surgery is a novel technology for spine navigation. Benchmark cadaveric data have demonstrated high accuracy and precision leading to recent regulatory approval. Absence of respiratory motion in cadaveric studies may positively bias precision and accuracy results and analogous investigations are prudent in live clinical scenarios.
OBJECTIVE
To report a technical note, accuracy, precision analysis of the first in-human deployment of this technology.
METHODS
A 78-yr-old female underwent an L4-S1 decompression, pedicle screw, and rod fixation for degenerative spine disease. Six pedicle screws were inserted via AR-HMD (xvision; Augmedics, Chicago, Illinois) navigation. Intraoperative computed tomography was used for navigation registration as well as implant accuracy and precision assessment. Clinical accuracy was graded per the Gertzbein-Robbins (GS) scale by an independent neuroradiologist. Technical precision was analyzed by comparing 3-dimensional (3D) (x, y, z) virtual implant vs real implant position coordinates and reported as linear (mm) and angular (°) deviation. Present data were compared to benchmark cadaveric data.
RESULTS
Clinical accuracy (per the GS grading scale) was 100%. Technical precision analysis yielded a mean linear deviation of 2.07 mm (95% CI: 1.62-2.52 mm) and angular deviation of 2.41° (95% CI: 1.57-3.25°). In comparison to prior cadaveric data (99.1%, 2.03 ± 0.99 mm, 1.41 ± 0.61°; GS accuracy 3D linear and angular deviation, respectively), the present results were not significantly different (P > .05).
CONCLUSION
The first in human deployment of the single Food and Drug Administration approved AR-HMD stereotactic spine navigation platform demonstrated clinical accuracy and technical precision of inserted hardware comparable to previously acquired cadaveric studies.
Objective
The purpose of this study was to develop and validate a nomogram to predict overall survival (OS) for adult patients with primary intramedullary spinal cord grade II/III ependymoma (PISCGE). We also elucidated the effectiveness of postoperative radiotherapy (RT) for this disease.
Methods
Clinical data of patients with PISCGE between 1988 and 2015 were collected from The Surveillance, Epidemiology, and End Results (SEER) registry database. The independent prognostic factors were identified using univariate and multivariate Cox analyses. The nomogram was established from the results of the multivariate Cox analysis. We also use some methods to verify the superiority of the prediction model. The effectiveness of postoperative radiotherapy for PISCGE was assessed through coarsened exact matching (CEM) and survival analyses.
Results
Multivariate Cox analysis revealed that sex, age, surgical treatment, tumor grade, and marital status were independent prognostic factors of OS. The nomogram model was established based on these factors and validated internally. Calibration plots based on bootstrap resampling validation showed good consistency between the nomogram prediction and actual observation. This model also exhibited favorable discrimination characteristics. A risk classification system based on a nomogram was established to promote risk stratification of PISCGE and optimize clinical management. Moreover, we found that no association between radiation treatment and the OS for these patients (P＞0.05).
Conclusion
We built the first nomogram model and risk classification system for PISCGE patients. Our model accurately estimated the individual OS probability of these patients, and proposed different treatment approaches for patients based on the risk classification system. Furthermore, from our findings, radiotherapy confers no survival advantage to these patients.
Background:
Virtual reality surgical simulators provide a safe environment for trainees to practice specific surgical scenarios and allow for self-guided learning. Artificial intelligence technology, including artificial neural networks, offers the potential to manipulate large datasets from simulators to gain insight into the importance of specific performance metrics during simulated operative tasks.
Objective:
To distinguish performance in a virtual reality-simulated anterior cervical discectomy scenario, uncover novel performance metrics, and gain insight into the relative importance of each metric using artificial neural networks.
Methods:
Twenty-one participants performed a simulated anterior cervical discectomy on the novel virtual reality Sim-Ortho simulator. Participants were divided into 3 groups, including 9 post-resident, 5 senior, and 7 junior participants. This study focused on the discectomy portion of the task. Data were recorded and manipulated to calculate metrics of performance for each participant. Neural networks were trained and tested and the relative importance of each metric was calculated.
Results:
A total of 369 metrics spanning 4 categories (safety, efficiency, motion, and cognition) were generated. An artificial neural network was trained on 16 selected metrics and tested, achieving a training accuracy of 100% and a testing accuracy of 83.3%. Network analysis identified safety metrics, including the number of contacts on spinal dura, as highly important.
Conclusion:
Artificial neural networks classified 3 groups of participants based on expertise allowing insight into the relative importance of specific metrics of performance. This novel methodology aids in the understanding of which components of surgical performance predominantly contribute to expertise.
OBJECTIVE
Recent advances in computer vision have revolutionized many aspects of society but have yet to find significant penetrance in neurosurgery. One proposed use for this technology is to aid in the identification of implanted spinal hardware. In revision operations, knowing the manufacturer and model of previously implanted fusion systems upfront can facilitate a faster and safer procedure, but this information is frequently unavailable or incomplete. The authors present one approach for the automated, high-accuracy classification of anterior cervical hardware fusion systems using computer vision.
METHODS
Patient records were searched for those who underwent anterior-posterior (AP) cervical radiography following anterior cervical discectomy and fusion (ACDF) at the authors’ institution over a 10-year period (2008–2018). These images were then cropped and windowed to include just the cervical plating system. Images were then labeled with the appropriate manufacturer and system according to the operative record. A computer vision classifier was then constructed using the bag-of-visual-words technique and KAZE feature detection. Accuracy and validity were tested using an 80%/20% training/testing pseudorandom split over 100 iterations.
RESULTS
A total of 321 total images were isolated containing 9 different ACDF systems from 5 different companies. The correct system was identified as the top choice in 91.5% ± 3.8% of the cases and one of the top 2 or 3 choices in 97.1% ± 2.0% and 98.4 ± 13% of the cases, respectively. Performance persisted despite the inclusion of variable sizes of hardware (i.e., 1-level, 2-level, and 3-level plates). Stratification by the size of hardware did not improve performance.
CONCLUSIONS
A computer vision algorithm was trained to classify at least 9 different types of anterior cervical fusion systems using relatively sparse data sets and was demonstrated to perform with high accuracy. This represents one of many potential clinical applications of machine learning and computer vision in neurosurgical practice.
Posterior pedicle fixation technique is a common method for treating thoracolumbar burst fractures, but the effect of different fixation techniques on the postoperative spinal mechanical properties has not been clearly defined, especially on adjacent segments. A finite element model of T10-L2 with moderate T12 vertebra burst fracture was constructed to investigate biomechanical behavior of three posterior pedicle screw fixation techniques. Compared with traditional short-segment 4 pedicle screw fixation (TS-4) and intermediate long-segment 6 pedicle screw fixation (IL-6), mono-segment 4 pedicle screw fixation (MS-4) provides a safer surgical selection to prevent the secondary degeneration of adjacent segments in the long-term.
OBJECTIVE:
To establish microscope-based augmented reality (AR) support for degenerative spine surgery.
METHODS:
Head-up displays of operating microscopes were used to establish AR in a series of ten patients. Segmentation of the vertebra and additional target structures, which were visualized by AR, was based on preoperative magnetic resonance and computed tomography (CT) images, that were non-rigidly fused to low-dose intraoperative CT (iCT) data. AR registration was achieved by automatic registration applying iCT and microscope calibration.
RESULTS:
AR support could be smoothly implemented in the surgical workflow. AR allowed to visualize the target structures reliably in the surgical field, facilitating surgical orientation. Flexible placement of the reference array enabled AR implementation for anterior, lateral, posterior median, and posterior paramedian approaches. Identification of bony and artificial landmarks allowed validating registration accuracy; the measured target registration error was 1.11 ± 0.42 mm (mean ± standard deviation). The effective dose (ED) for registration scanning ranged from 0.52 to 8.71 mSv, which is in average about a third of a standard diagnostic spine scan. This depended mainly on the scan length (mean scan length cervical/thoracic/lumbar: 99 / 218 / 118 mm). Longest scan ranges were in the mid-thoracic region to ensure unambiguous vertebra assignment as prerequisite for reliable non-linear registration (mean cervical/thoracic/lumbar ED: 0.52 / 6.14 / 2.99 mSv).
CONCLUSION:
Reliable microscope-based AR support is possible due to automatic registration based on intraoperative imaging. Application of AR in degenerative spine surgery has a big potential, it might be especially helpful in complex anatomical situations and resident education.
OBJECTIVE
Augmented reality (AR) is a novel technology that has the potential to increase the technical feasibility, accuracy, and safety of conventional manual and robotic computer-navigated pedicle insertion methods. Visual data are directly projected to the operator’s retina and overlaid onto the surgical field, thereby removing the requirement to shift attention to a remote display. The objective of this study was to assess the comparative accuracy of AR-assisted pedicle screw insertion in comparison to conventional pedicle screw insertion methods.
METHODS
Five cadaveric male torsos were instrumented bilaterally from T6 to L5 for a total of 120 inserted pedicle screws. Postprocedural CT scans were obtained, and screw insertion accuracy was graded by 2 independent neuroradiologists using both the Gertzbein scale (GS) and a combination of that scale and the Heary classification, referred to in this paper as the Heary-Gertzbein scale (HGS). Non-inferiority analysis was performed, comparing the accuracy to freehand, manual computer-navigated, and robotics-assisted computer-navigated insertion accuracy rates reported in the literature. User experience analysis was conducted via a user experience questionnaire filled out by operators after the procedures.
RESULTS
The overall screw placement accuracy achieved with the AR system was 96.7% based on the HGS and 94.6% based on the GS. Insertion accuracy was non-inferior to accuracy reported for manual computer-navigated pedicle insertion based on both the GS and the HGS scores. When compared to accuracy reported for robotics-assisted computer-navigated insertion, accuracy achieved with the AR system was found to be non-inferior when assessed with the GS, but superior when assessed with the HGS. Last, accuracy results achieved with the AR system were found to be superior to results obtained with freehand insertion based on both the HGS and the GS scores. Accuracy results were not found to be inferior in any comparison. User experience analysis yielded “excellent” usability classification.
CONCLUSIONS
AR-assisted pedicle screw insertion is a technically feasible and accurate insertion method.
Purpose:
To differentiate metastatic lesions in the spine originated from primary lung cancer and other cancers using radiomics and deep learning, compared to traditional hot-spot ROI analysis.
Methods:
In a retrospective review of clinical spinal MRI database with a dynamic contrast enhanced (DCE) sequence, a total of 61 patients without prior cancer diagnosis and later confirmed to have metastases (30 lung; 31 non-lung cancers) were identified. For hot-spot analysis, a manual ROI was placed to calculate three heuristic parameters from the wash-in, maximum, and wash-out phases in the DCE kinetics. For each case, the 3D tumor mask was generated by using the normalized-cut algorithm. Radiomics analysis was performed to extract histogram and texture features from three DCE parametric maps. Deep learning was performed using these maps as inputs into a conventional convolutional neural network (CNN), as well as using all 12 sets of DCE images into a convolutional long short term memory (CLSTM) network.
Results:
For hot-spot ROI analysis, mean wash-out slope was 0.25 ± 10% for lung metastases and -9.8 ± 12.9% for other tumors. CHAID classification using a wash-out slope of -6.6% followed by wash-in enhancement ratio of 98% achieved a diagnostic accuracy of 0.79. Radiomics analysis using features representing tumor heterogeneity only reached the highest accuracy of 0.71. Classification using CNN achieved a mean accuracy of 0.71 ± 0.043, whereas a CLSTM improved accuracy to 0.81 ± 0.034.
Conclusions:
DCE-MRI machine-learning analysis methods have potential to predict lung cancer metastases in the spine, which may be used to guide subsequent workup for confirmed diagnosis.
Background:
Preoperative prognostication of short-term postoperative mortality in patients with spinal metastatic disease can improve shared decision making around end-of-life care.
Objective:
To (1) develop machine learning algorithms for prediction of short-term mortality and (2) deploy these models in an open access web application.
Methods:
The American College of Surgeons, National Surgical Quality Improvement Program was used to identify patients that underwent operative intervention for metastatic disease. Four machine learning algorithms were developed, and the algorithm with the best performance across discrimination, calibration, and overall performance was integrated into an open access web application.
Results:
The 30-d mortality for the 1790 patients undergoing surgery for spinal metastatic disease was 8.49%. Preoperative factors used for prognostication were albumin, functional status, white blood cell count, hematocrit, alkaline phosphatase, spinal location (cervical, thoracic, lumbosacral), and severity of comorbid systemic disease (American Society of Anesthesiologist Class). In this population, machine learning algorithms developed to predict 30-d mortality performed well on discrimination (c-statistic), calibration (assessed by calibration slope and intercept), Brier score, and decision analysis. An open access web application was developed for the best performing model and this web application can be found here: https://sorg-apps.shinyapps.io/spinemets/.
Conclusion:
Machine learning algorithms are promising for prediction of postoperative outcomes in spinal oncology and these algorithms can be integrated into clinically useful decision tools. As the volume of data in oncology continues to grow, creation of learning systems and deployment of these systems as accessible tools may significantly enhance prognostication and management.
Scoliosis is an abnormal sideways curvature of the spine and rib cage, which may need surgical treatments. Most of the corrective maneuvers in scoliosis surgeries are based on surgeon's experience; hence, there is great interest of understanding how the correction ratio can be influenced by the magnitude of forces and moments. Therefore, the objective of this study was to develop and validate a detailed finite element model of the thoracolumbar which can be used to simulate the scoliosis surgeries based on patient-specific clinical images. The validated models of five patients were carefully developed, and the surgery procedures were simulated and the corrective forces were estimated using inverse finite element analysis during the surgery. Furthermore, parametric studies including the influences of the corrective force magnitude and screw density were evaluated. The results showed that the maximum estimated correction force and moment were 173 (±55.43) N and 10.67 (±2.02) N m, respectively, which were aligned with measured clinical observations. The sensitivity analysis on the magnitude of applied force to the screws showed that correction ratio was slightly increased in level 1 (i.e. FB = 1.3 × F) but decreased in level 2 (i.e. FB = 1.6 × F). In addition, the parametric study on increasing the number of pedicle screws showed that there was no significant difference between lower and higher screw density. However, the stress distribution was significantly greater using higher screw density during correction maneuvers. In conclusion, this study shows a direct relationship between the applied force/moment and screw density and the correction ratio up to a border line which should be defined accurately. This detailed computational modeling can be used in clinic in hope of achieving the optimum outcome of scoliosis surgery using individual patient-specific characterization.
Study design:
Computer biomechanical simulations to analyze risk factors of proximal junctional failure (PJF) following adult scoliosis instrumentation.
Objective:
To evaluate the biomechanical effects on the proximal junctional spine of the proximal implant type, tissue dissection, and lumbar lordosis (LL) restoration.
Summary of background data:
PJF is a severe proximal junctional complication following adult spinal instrumentation requiring revision surgery. Potential risk factors have been reported in the literature, but knowledge on their biomechanics is still lacking to address the issues.
Methods:
A patient-specific multibody and finite-element hybrid modeling technique was developed for a 54-year-old patient having undergone instrumented spinal fusion for multilevel stenosis resulting in PJF. Based on the actual surgery, 30 instrumentation scenarios were derived and simulated by changing the implant type at the upper instrumented vertebra (UIV), varying the extent of proximal osteotomy and the degree of LL creation. Five functional loads were simulated, and stresses and strains were analyzed for each of the 30 tested scenarios.
Results:
There was 80% more trabecular bone with stress greater than 0.5 MPa in the UIV with screws compared to hooks. Hooks allowed 96% more mobility of the proximal instrumented functional unit compared to screws. The bilateral complete facetectomy along with posterior ligaments dissection caused a significant increase of the range of motion of the functional unit above the UIV. LL creation increased the flexion moment applied on the proximal vertebra from 7.5 to 17.5 Nm, which generated damage at the bone-screw interface that affected the screw purchase.
Conclusion:
Using hooks at UIV and reducing posterior proximal spinal element dissection lowered stress levels in the proximal junctional spinal segment and thus reduced the biomechanical risks of PJF. LL restoration was associated with increased stress levels in postoperative functional upper body flexion.
Compression-based fusionless tethers are an alternative to conventional surgical treatments of pediatric scoliosis. Anterior approaches place an anterior (ANT) tether on the anterolateral convexity of the deformed spine to modify growth. Posterior, or costo-vertebral (CV), approaches have not been assessed for biomechanical and corrective effectiveness. The objective was to biomechanically assess CV and ANT tethers using six patient-specific, finite element models of adolescent scoliotic patients (11.9 ± 0.7yrs, Cobb 34° ± 10°). A validated algorithm simulated the growth and Hueter-Volkmann growth modulation over a period of 2 years with the CV and ANT tethers at two initial tensions (100, 200N). The models without tethering also simulated deformity progression with Cobb angle increasing from 34° to 56°, axial rotation 11° to 13°, and kyphosis 28° to 32° (mean values). With the CV tether, the Cobb angle was reduced to 27° and 20° for tensions of 100N and 200N respectively, kyphosis to 21° and 19°, and no change in axial rotation. With the ANT tether, Cobb was reduced to 32° and 9° for 100N and 200N respectively, kyphosis unchanged, and axial rotation to 3° and 0°. While the CV tether mildly corrected the coronal curve over a 2-year growth period, it had sagittal lordosing effect, particularly with increasing initial axial rotation (>15°). The ANT tether achieved coronal correction, maintained kyphosis, and reduced the axial rotation, but over-correction was simulated at higher initial tensions. This biomechanical study captured the differences between a CV and ANT tether and indicated the variability arising from the patient-specific characteristics. This article is protected by copyright. All rights reserved.