Deep Learning Based Evaluation of Surgical Candidacy for Cervical Spinal Cord Decompression

Deep Learning Based Evaluation of Surgical Candidacy for Cervical Spinal Cord Decompression | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Deep Learning Based Evaluation of Surgical Candidacy for Cervical Spinal Cord Decompression Anshul Ratnaparkhi, Bayard Wilson, David Zarrin BSE, Abhinav Suri, and 9 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4385667/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Many patients who present to their primary care physician for neck pain undergo magnetic resonance imaging (MRI) as part of their diagnostic workup. The physician is then tasked with deciding if the findings of the MRI and workup warrant referral a spine surgery, an intricate task complicated by the high rates of background findings. This results in a high number of non-surgical patients being referred to surgery. Although there are a multitude of reasons for non-surgical patients to still see a subspecialist, deep learning has the potential to help inform physicians of their patients’ surgical candidacy. The preset work describes a proof-of-concept model for evaluating operative candidacy for cervical stenosis only using data from outpatient elective magnetic resonance imaging (MRI) scans. This deep-learning algorithm was trained to automatically segment the areas of both the spinal canal and spinal cord on 100 axial cervical spine MRIs. Once segmented, the model used these areas to generate a biomarker for cervical stenosis, calculated as the minimum difference in cross-sectional area between the spinal canal and the spinal cord within the cervical spine. Following training, the model and its biomarker were tested against a cohort of 147 consecutive patients evaluated in the outpatient setting by a group of board-certified neurosurgeons at our institution for complaints related to their cervical spines. Our automated model determined that the mean minimum difference in cross-sectional area between the spinal canal and spinal cord for our cohort was 35.90±25.00 mm 2 for patients who ultimately underwent surgical decompression and 48.55±33.52 mm 2 for patients who did not (P=0.005). Using this biomarker, the model distinguished between surgical and non-surgical patients with relatively high accuracy (AUC 0.79). When tested against a novel cohort of outpatient spine surgery clinic patients, the described algorithm determined whether the patient underwent decompression for cervical stenosis using data acquired solely from their cervical spine MRI scans. These findings support a proof-of-concept for our automated deep-learning model and biomarker, which could significantly improve the efficiency of the referral process for patients with neck complaints to a surgical subspecialist. Health sciences/Biomarkers/Diagnostic markers Health sciences/Biomarkers/Predictive markers Health sciences/Health care/Diagnosis Health sciences/Health care/Medical imaging cervical spine deep learning machine learning surgical candidacy triage. Figures Figure 1 Figure 2 Figure 3 INTRODUCTION Neck pain evaluations comprise a substantial number of primary care consultations each year 1,2 . While current evidence affords these patients with a variety of options — including exercise programs, medical therapies, diagnostic studies, and procedural interventions — the first question often posed to their referring physicians is whether the patient’s complaints warrant referral to a surgeon. In many instances, a patient’s clinical history and physical exam findings will prompt a physician to obtain magnetic resonance imaging (MRI) studies of the cervical spine. Given the cost of obtaining an MRI, considerable efforts have been made in the primary care setting to ensure they are ordered only for the appropriate clinical indications 3,4 . Nevertheless, once completed the referring provider is tasked with determining whether the results of the MRI can be managed alone or require the input of a surgical subspecialist. Due to the relative nuance of interpreting spine MR images, referring providers often rely on radiological reports to determine the degree of pathological severity for a given scan. In some cases, a referral to a surgeon may be suggested by the reading radiologist in the text of the radiology report. In others, the ordering provider must make that determination using the qualifiers within the report, few of which speak to whether or not a particular finding is treatable with surgery. Automated image analysis has significant potential in this context as it can provide ordering physicians with additional information to inform their decisions regarding whether or not to make a referral to a surgical specialist. In this study, we present a proof of concept for automated image analysis using deep learning to automatically determine whether a patient’s cervical stenosis warrants surgical decompression using an imaging biomarker derived solely from axial non-contrasted T2-weight MR images of the cervical spine ( Figure 1 ). We discuss the relative components of the model and suggest that our model, and others like it, could measurably improve the efficiency of referrals to spine surgeons by reducing the number of non-operative referrals. RESULTS Baseline characteristics for the test cohort are summarized in Table 1. Of the 205 patients initially identified, 58 were excluded for having inaccessible MRI data (obtained at an outside institution), leaving 147 patient images against which to test our model. Across this cohort, the mean minimum difference in area between the spinal cord and spinal canal (i.e. our biomarker) differed significantly between patients who ultimately underwent surgical decompression (35.90±25.00 mm 2 ) and patients for whom surgery was deemed unnecessary during their clinic evaluation (48.55±33.52 mm 2 , p-value 5.290 x 10 -3 , see Table 2 and Figure 2 ). Table 1: Comparison of age and gender among surgical and non-sugical patients. Non- Surgical (n=80) Surgical (n=67) Age 54.63 ± 14.13 60.08 ± 12.87 Male Sex 45 34 Female Sex 35 33 Table 2 : Comparison of the mean of the proposed biomarker between nonsurgical and surgical cohorts. Mean (mm 2 ) Stdev (mm 2 ) Statistics Surgical (n=67) 35.90 25.00 p-value: 5.290 x 10 -3 Non-surgical (n=80) 48.55 33.52 t-statistic: -2.829 By solely relying on the radiographic biomarker generated for each automatically segmented scan, our model was able to prospectively predict (based on retrospectively collected data) whether a patient would ultimately undergo surgery for decompression of their cervical stenosis with relatively high accuracy (AUC 0.79). Of the 147 patients included in our analysis, 67 were deemed surgical candidates and 80 deemed nonsurgical. The Receiver Operating Characteristic (ROC) analysis and the corresponding confusion matrix for our model when tested against this cohort, are presented in Figure 3 and Table 3 , respectively. Table 3: Confusion matrix corresponding to the prediction of surgical status using the proposed biomarker. DISCUSSION In this study, we describe a proof of concept for an automated image analysis tool capable of identifying operative cervical stenosis using only data from MR images. Our model distinguished between surgical and non-surgical cervical stenosis with high accuracy (AUC 0.79) by relying on an automated biomarker representing the degree of space between the spinal cord and surrounding spinal canal. This biomarker was constructed because it is both physiologic and intuitive; it follows that as the area of the spinal canal approaches that of the spinal cord, spinal cord compression becomes more severe. Moreover, the results suggest — perhaps unsurprisingly — that increases in compression severity correlate with increases in surgical candidacy. The success of our model in predicting whether patients ultimately underwent surgical decompression for cervical stenosis is worth discussing both in terms of the components it does and does not include. Currently, no model for surgical decision-making should rely on imaging data alone; however, the accuracy of our biomarker suggests that, in many instances, radiographic interpretation is not only necessary to determine whether a patient’s cervical stenosis is operative, but also sufficient. Previous work by our group used deep learning to generate a parallel biomarker for lumbar stenosis to predict whether a patient’s lumbar pathology warranted surgical decompression 5 . Compared to our cervical model, our lumbar biomarker managed to achieve a higher degree of predictive accuracy (AUC 0.88). This is partially a consequence of the inherent differences in anatomy between these two spinal regions, the most obvious of which is the presence of the spinal cord coursing within the cervical spinal canal. As such, in cases of cervical stenosis, variables such as cord deformity and changes in T2 signal within the cord itself can often play a role in surgical decision-making, and for the moment, these variables remain absent from our model. Ongoing work by our group hopes to include these components as part of a more comprehensive model. As a proof of concept, our model is currently not intended to assist in the clinical decision making of spine surgeons. Rather, the real value of the present algorithm nd the iterations that build upon it lies in optimizing the referral process to spine surgeons. As any spine surgeon can attest, a substantial number of patients — more than half in many cases — referred to see a spine surgeon never undergo surgery 6 – 8 . While some of these non-operative patients still benefit from time spent in consultation with a surgeon, the degree of inefficiency with the referral process can lead to longer wait times for patients with operative pathology and delays in care for those non-operative pathology (i.e. delaying referral to a physiatrist or physical therapy) 9 , 10 . Spine patients are particularly vulnerable to this inefficiency due in part to the relative nuance of radiographic interpretation of spinal pathology. Given both the incidence of neck and back-related complaints during visits to primary care practicioners 11 , 12 and the prevalence of non-operative degenerative changes on routine MRIs of the spine 13 , 14 , it is not surprising that so many patients arrive at the office of a spine surgeon only to be reassured that their findings do not require surgery. This problem is not novel, and various efforts have been made to improve the efficiency of referrals to spine surgeons with some success. Current efforts include the development of multidisciplinary pathways 15 and clinical practice guidelines 16 ; however, no studies have focused on the problem of variability among referring providers with respect to discerning operative from non-operative spinal pathology. The present model could provide referring physicians tasked with reviewing a patient’s cervical spine MRI a greater degree of confidence when making the decision to refer to a spine surgeon. Our stenosis biomarker could be used to generate a surgical likelihood score, informing referring providers of how likely a patient’s stenosis is to require surgery. To be clear, our model would not serve as a replacement for the clinical judgement of a physician, but instead help provide additional actionable information — alongside a patient’s clinical history, signs, symptoms and physical exam findings — with which to make the decision to refer to a specialist. This additional information could help increase the number of patients with surgical pathology who are referred to a surgeon, as well as decrease the number referred without surgical pathology who could instead be referred for non-operative treatment modalities. LIMITATIONS There are several limitations to our study. Firstly, our model is susceptible to the domain effect since the decision regarding surgical candidacy was limited to a group of physicians at a single institution, limiting its generalizability. Secondly, as a proof of concept our model focused on cases of cervical stenosis to the exclusion of other surgical pathologies of the cervical spine, and so fails to address other forms of cervical pathology warranting referral to a subspecialist (foraminal stenosis, tumor, or trauma) 17 . Thirdly, as an automated image analysis tool, our model knowingly omits any details related to patients’ clinical presentation or history. Nevertheless, that it manages to successfully predict surgical candidacy for decompressive surgery with nearly 80% accuracy speaks to the relative importance of radiographic findings in surgical decision making. Lastly, our model does not capture those patients without surgical pathology for whom surgical referral is warranted for reasons beyond clinical history or physical exam, such as for reassurance or to gather more information about their condition. Such cases go unnoticed by our model, but are important cases for referral. These limitations notwithstanding, we feel our model performs well in predicting surgical candidacy using imaging data from MR images of the cervical spine alone. CONCLUSION This work presents a novel, fully autonomous model capable of predicting a patient’s surgical candidacy for cervical decompression based solely on the degree of cervical stenosis calculated on axial MR images with relatively high accuracy (AUC 0.79). We argue that this model could serve as a valuable tool for referring physicians when considering whether to consult a spine surgeon for patients with cervical stenosis, and its incorporation along with other clinical data could help optimize the process by which patients are referred for evaluation in spine surgery clinic. METHODS IRB and Ethical Statement: The present study was conducted under the World Declaration of Helsinki: ethical principles for medical research involving human subjects 18 . The present work study was conducted according to the rules and regulations of our institution, as approved by the Institutional Review Board (IRB #16-000196). Waiver of informed consent was granted for the purposes of data collection by our institution’s IRB. Data Availability: The datasets used and analyzed during the current study are available from the corresponding author at reasonable request. Manual Segmentation for training network: Our machine learning model was trained on manually segmented MR images. One hundredT2-axial MRI scans were randomly selected from our institution’s picture archiving and communications system (PACS). Images were downloaded as DICOM files, anonymized by removing header information, converted to NIFTI format, resampled to 256 x 256 pixels in the plane of acquisition, and, finally, histogram matched to a common template to standardize the maximum and minimum pixel values to 1 and 0 respectively. Following this pre-processing, clinically trained human expert raters delineated cervical canals and spinal cords in the axial plane on these scans using ITK-SNAP, an open-source application for medical image segmentation ( Figure 1 ) 5 . Segmentations were stored as NIFTI files alongside the pre-processed training data for deep learning training. Deep Learning Model Training: Our ensemble contained three U-Nets trained on image patches extracted from pre-processed axial T2-MRIs. The first U-Net was trained using patches 160x160px, the second using patches 128x128px, and the third using patches 96x96px. Patches were extracted in the plane of acquisition at regular strides of length 80px, 64 px, and 48px, respectively. Extracted image patches, along with the corresponding extracted region from the segmentation images, were augmented three times using affine transformations and flipped up-down as well as left-right. The collection of augmented and original patches and segmentations was then used to train the network. The three networks in our ensemble were identical in all hyperparameters except the input image size. The relevant hyperparameters were a kernel size of 3x3px with a stride of 1px in each dimension, ReLU activation, and Glorot initialization for all convolutional layers. The starting channel depth was 16 for all networks, and our networks had five downsampling and upsampling units. Each unit consisted of a max pooling block using a 2x2px kernel sandwiched between two convolutional blocks. Skip connections were added in accordance with the standard U-Net dogma. Channel depth was doubled with each downsampling and halved with each upsampling according to standard U-Net dogma as well. The final layer of all networks was a sigmoid to constrain the output between 0 and 1 for segmentation purposes. We also use dropout regularization with a probability of 0.25 and the Adam optimizer with a fixed learning rate of 1e-5 for training, alongside the Dice coefficient as a loss function. Segmentations generated using all three networks (96x96, 128 x 128, and 160 x 160) in the ensemble were combined using pixel-wise majority voting to generate the final segmentation. The results presented here are based on two ensembles: the first trained to segment spinal canals automatically and the second trained to segment spinal cords automatically in the axial plane on cervical spine MR images. Biomarker Generation: Canal and cord areas were measured in square millimeters (mm 2 ) at all cervical levels except the most cranial 20% of each image due to the presence of brain regions that were irrelevant to our research question. The difference in area between the spinal canal and spinal cord was computed at every level, and the minimum value for this difference in area along the entire cervical spinal canal served as our biomarker. Segmentation failure for a particular axial slice was defined when irrelevant regions of the spinal canal (i.e. the medulla) were segmented or when the model failed to segment either the spinal canal or spinal cord region completely. In such cases, the specific axial slice where failure occurred was excluded. Patient Data Collection: To test our model in the clinical setting, we then collected and stored MR imaging data and clinical metadata for 205 consecutive patients presenting with complaints related to cervical stenosis at our institution between 2019 and 2020. As part of their treatment course, all of these patients had been independently evaluated by one of four board-certified neurosurgeons during that same time period. Patients were excluded if their imaging data could not be accessed or processed – an issue that might occur if their imaging had been obtained at an outside institution. Accessible MR images were downloaded in DICOM format and pre-processed as described above. Further processing and measurements were performed using these pre-processed images. Model Evaluation: We then allowed our model to perform automated segmentations on these MRI scans using previously described trained deep learning ensemble segmentation algorithms. The corresponding biomarker generated by these automated segmentations was tested against the outcome of a subsequent surgical decompression for cervical stenosis. Declarations Disclosures: Dr. Macyszyn discloses a financial interest in Theseus Artificial Intelligence, which has signed a license agreement with the University of California, Los Angeles for the technology utilized in this research. The remaining authors have no personal, financial, or institutional interest in any part of the research presented in this work. Author Contribution A.R. and B.W. co-authored the main manuscript text. D.Z. and T.F. were responsible for editing the manuscript. A.R., D.Z, K.C, A.L., T.F, B.Y., and B.S were responsible for data acquisition and data analysis..B.G, A.S, were responsible for designing the machine learning architecture. K.C. and D.B prepared the figures.B.G., L.M., and J.B. were responsible for the study conception, funding, supervision, and overall direction of the project. All authors reviewed the final manuscript. Acknowledgement Acknowledgments: The authors would like to dedicate this work in honor of Dr. Bilwaj Gaonkar, Ph.D., who passed away during the course of this study. Dr. Gaonkar, a co-author of this manuscript, contributed substantially to the research described. The present study would not have been possible without him. Data Availability The datasets used and analyzed during the current study are available from the corresponding author at reasonable request. References Bot SDM, van der Waal JM, Terwee CB, et al. Incidence and prevalence of complaints of the neck and upper extremity in general practice. Ann Rheum Dis . 2005;64(1):118-123. Picavet HSJ, Schouten JSAG. Musculoskeletal pain in the Netherlands: prevalences, consequences and risk groups, the DMC3-study. Pain . 2003;102(1):167-178. Izzo R, Popolizio T, Balzano RF, Pennelli AM, Simeone A, Muto M. Imaging of cervical spine traumas. Eur J Radiol . 2019;117:75-88. Malhotra A, Wu X, Kalra VB, et al. Utility of MRI for cervical spine clearance after blunt traumatic injury: a meta-analysis. Eur Radiol . 2017;27(3):1148-1160. Wilson B, Gaonkar B, Yoo B, et al. Predicting Spinal Surgery Candidacy From Imaging Data Using Machine Learning. Neurosurgery . 2021;89(1):116-121. Debono B, Sabatier P, Koudsie A, Buffenoir K, Hamel O. Managing spine surgery referrals: The consultation of neurosurgery and its nuances. Neurochirurgie . 2017;63(4):267-272. Kidane B, Gandhi R, Sarro A, Valiante TA, Harvey BJ, Rampersaud YR. Is referral to a spine surgeon a double-edged sword?: patient concerns before consultation. Can Fam Physician . 2011;57(7):803-810. Mayman D, Yen D. Maximizing use of a surgical clinic for referrals of patients having back problems. Can J Surg . 1999;42(2):117-119. Findlay JM, Deis N. Appropriateness of lumbar spine referrals to a neurosurgical service. Can J Neurol Sci . 2010;37(6):843-848. Braybrooke J, Ahn H, Gallant A, et al. The impact of surgical wait time on patient-based outcomes in posterior lumbar spinal surgery. Eur Spine J . 2007;16(11):1832-1839. Childress MA, Stuek SJ. Neck pain: Initial evaluation and management. Am Fam Physician . 2020;102(3):150-156. Bartholomeeusen S, Van Zundert J, Truyers C, Buntinx F, Paulus D. Higher incidence of common diagnoses in patients with low back pain in primary care. Pain Pract . 2012;12(1):1-6. Jensen MC, Brant-Zawadzki MN, Obuchowski N, Modic MT, Malkasian D, Ross JS. Magnetic resonance imaging of the lumbar spine in people without back pain. N Engl J Med . 1994;331(2):69-73. Kasch R, Truthmann J, Hancock MJ, et al. Association of lumbar MRI findings with current and future back pain in a population-based cohort study. Spine (Phila Pa 1976) . 2022;47(3):201-211. Wilgenbusch CS, Wu AS, Fourney DR. Triage of spine surgery referrals through a multidisciplinary care pathway: a value-based comparison with conventional referral processes. Spine . 2014;39(22 Suppl 1):S129-35. Layne EI, Roffey DM, Coyle MJ, Phan P, Kingwell SP, Wai EK. Activities performed and treatments conducted before consultation with a spine surgeon: are patients and clinicians following evidence-based clinical practice guidelines? Spine J . 2018;18(4):614-619. Fernández de Rota JJ, Meschian S, Fernández de Rota A, Urbano V, Baron M. Cervical spondylotic myelopathy due to chronic compression: the role of signal intensity changes in magnetic resonance images. J Neurosurg Spine . 2007;6(1):17-22. World medical association declaration of Helsinki. JAMA . 2013;310(20):2191. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4385667","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":304746071,"identity":"99a9e167-db0e-449e-90aa-b136b290d5ba","order_by":0,"name":"Anshul Ratnaparkhi","email":"","orcid":"","institution":"University of Miami","correspondingAuthor":false,"prefix":"","firstName":"Anshul","middleName":"","lastName":"Ratnaparkhi","suffix":""},{"id":304746072,"identity":"c8dbc001-ec6c-418f-85e5-0422c32dc5a5","order_by":1,"name":"Bayard Wilson","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Bayard","middleName":"","lastName":"Wilson","suffix":""},{"id":304746074,"identity":"7c1b4cf3-db31-43f1-9481-9918d4d90952","order_by":2,"name":"David Zarrin BSE","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"David","middleName":"Zarrin","lastName":"BSE","suffix":""},{"id":304746075,"identity":"d5f5e413-0ce9-41fe-93d6-857bfb30bc78","order_by":3,"name":"Abhinav Suri","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Abhinav","middleName":"","lastName":"Suri","suffix":""},{"id":304746077,"identity":"1ed2b887-fbd0-4f74-9ed9-ba643681faec","order_by":4,"name":"Bryan Yoo","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Bryan","middleName":"","lastName":"Yoo","suffix":""},{"id":304746080,"identity":"7b75cfa9-ffad-4b03-8c9c-4be9f690e2e9","order_by":5,"name":"Banafsheh Salehi","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Banafsheh","middleName":"","lastName":"Salehi","suffix":""},{"id":304746081,"identity":"2a3c2755-bc1c-42f2-aa7a-2df83737b36e","order_by":6,"name":"David Berin","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"David","middleName":"","lastName":"Berin","suffix":""},{"id":304746084,"identity":"455368f3-7f05-4615-bc63-eafe6bce043c","order_by":7,"name":"Kirstin Cook","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Kirstin","middleName":"","lastName":"Cook","suffix":""},{"id":304746085,"identity":"49cbad5b-bd4b-415c-85ae-516d980695a0","order_by":8,"name":"TJ Florence","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"TJ","middleName":"","lastName":"Florence","suffix":""},{"id":304746086,"identity":"cbc92ec4-f138-4970-9313-5fee3b4a54c1","order_by":9,"name":"Azim Laiwalla","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Azim","middleName":"","lastName":"Laiwalla","suffix":""},{"id":304746087,"identity":"46723fdd-40f4-4243-9e72-73b0b53e7b76","order_by":10,"name":"Bilwaj Gaonkar","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Bilwaj","middleName":"","lastName":"Gaonkar","suffix":""},{"id":304746088,"identity":"a0cb3384-8f31-458e-8941-4e5f91f54290","order_by":11,"name":"Luke Macyszyn","email":"","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":false,"prefix":"","firstName":"Luke","middleName":"","lastName":"Macyszyn","suffix":""},{"id":304746089,"identity":"2f5c5242-2798-473b-8846-c8560c5836c0","order_by":12,"name":"Joel Beckett","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABBUlEQVRIie3PsYqDMBjA8QShk55rROq9Qo6AFHyZitAuQm90kCOHkOkeIGAfon2DimCXzKXg5Bu0S7mhyCXpTQfxbrwhf/gkmvxEAbDZ/mGQy6Fy4ctBj2eOund+IxgFVBMMvl9hJuqYJvjwV+LUVXPlxf2NnNqmfy3uAB+b9wsokpSaPmzbZcFOYBT3qyzhAgMs0goBsTYTnsdwYIrkcegxSc6QIsjaCbK5wWHEiNSbW+iNmlSfcJwi+Qzuqfz9MJ+FHtWEIUgnyLYjAVfTr0jidsQNRMoWy25NTOSlrobrRxn5fp0NvVtG0dOxbc+XMpkbyc8NV1+XhuOq54k9m81msz36AgCnWXm5Cq5bAAAAAElFTkSuQmCC","orcid":"","institution":"University of California Los Angeles","correspondingAuthor":true,"prefix":"","firstName":"Joel","middleName":"","lastName":"Beckett","suffix":""}],"badges":[],"createdAt":"2024-05-08 00:53:40","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4385667/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4385667/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":57429327,"identity":"856d5822-911b-499b-8233-25e8844efd2f","added_by":"auto","created_at":"2024-05-30 14:47:04","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":72113,"visible":true,"origin":"","legend":"\u003cp\u003eSegmentations of the cervical spinal cord (red) and cervical spinal column (green) generated automated using our machine-learning model on MR images on a non-surgical patient (A, B) and surgical patient (C, D)\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4385667/v1/5cd093d11570837d0af5dd7c.jpg"},{"id":57429328,"identity":"d373feb8-0ce5-400c-801d-4d25a94a2c6e","added_by":"auto","created_at":"2024-05-30 14:47:05","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":28612,"visible":true,"origin":"","legend":"\u003cp\u003eBoxplot showing the statistically significant difference in the distribution of proposed biomarkers in surgical and non-surgical cohorts\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4385667/v1/b6902786f34f1b90b40c45cf.jpg"},{"id":57430614,"identity":"ef7aa017-3e1f-4a90-b758-fc208569badc","added_by":"auto","created_at":"2024-05-30 14:55:05","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":52420,"visible":true,"origin":"","legend":"\u003cp\u003eROC curve corresponding to the prediction of surgical status using the proposed biomarker.\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4385667/v1/439eb48c072ab173aa58749f.jpg"},{"id":77411009,"identity":"f26c2765-4962-49c8-96d1-c9b203fcd0fb","added_by":"auto","created_at":"2025-02-28 10:17:03","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":615625,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4385667/v1/7dcc9b19-4c9f-4780-8223-c103951e10d2.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Deep Learning Based Evaluation of Surgical Candidacy for Cervical Spinal Cord Decompression","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eNeck pain evaluations comprise a substantial number of primary care consultations each year\u0026nbsp;\u003csup\u003e1,2\u003c/sup\u003e. While current evidence affords these patients with a variety of options \u0026mdash; including exercise programs, medical therapies, diagnostic studies, and procedural interventions\u0026nbsp;\u0026mdash; the first question often posed to their referring physicians is whether the patient\u0026rsquo;s complaints warrant referral to a surgeon.\u003c/p\u003e\n\u003cp\u003eIn many instances, a patient\u0026rsquo;s clinical history and physical exam findings will prompt a physician to obtain magnetic resonance imaging (MRI) studies of the cervical spine. \u0026nbsp;Given the cost of obtaining an MRI, considerable efforts have been made in the primary care setting to ensure they are ordered only for the appropriate clinical indications\u0026nbsp;\u003csup\u003e3,4\u003c/sup\u003e. Nevertheless, once completed the referring provider is tasked with determining whether the results of the MRI can be managed alone or require the input of a surgical subspecialist.\u003c/p\u003e\n\u003cp\u003eDue to the relative nuance of interpreting spine MR images, referring providers often rely on radiological reports to determine the degree of pathological severity for a given scan. In some cases, a referral to a surgeon may be suggested by the reading radiologist in the text of the radiology report. In others, the ordering provider must make that determination using the qualifiers within the report, few of which speak to whether or not a particular finding is treatable with surgery. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAutomated image analysis has significant potential in this context as it can provide ordering physicians with additional information to inform their decisions regarding whether or not to make a referral to a surgical specialist. In this study, we present a proof of concept for automated image analysis using deep learning to automatically determine whether a patient\u0026rsquo;s cervical stenosis warrants surgical decompression using an imaging biomarker derived solely from axial non-contrasted T2-weight MR images of the cervical spine (\u003cstrong\u003eFigure 1\u003c/strong\u003e). \u0026nbsp;We discuss the relative components of the model and suggest that our model, and others like it, could measurably improve the efficiency of referrals to spine surgeons by reducing the number of non-operative referrals. \u0026nbsp;\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cp\u003eBaseline characteristics for the test cohort are summarized in \u003cstrong\u003eTable 1. \u0026nbsp;\u003c/strong\u003eOf the 205 patients initially identified, 58 were excluded for having inaccessible MRI data (obtained at an outside institution), leaving 147 patient images against which to test our model. \u0026nbsp;Across this cohort, the mean minimum difference in area between the spinal cord and spinal canal (i.e. our biomarker) differed significantly between patients who ultimately underwent surgical decompression (35.90\u0026plusmn;25.00 mm\u003csup\u003e2\u003c/sup\u003e) and patients for whom surgery was deemed unnecessary during their clinic evaluation (48.55\u0026plusmn;33.52 mm\u003csup\u003e2\u003c/sup\u003e, p-value 5.290 x 10\u003csup\u003e-3\u003c/sup\u003e, see \u003cstrong\u003eTable 2\u0026nbsp;\u003c/strong\u003eand \u003cstrong\u003eFigure 2\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1:\u003c/strong\u003e Comparison of age and gender among surgical and non-sugical patients.\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"394\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.380710659898476%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.131979695431475%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eNon- Surgical (n=80)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"32.48730964467005%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSurgical (n=67)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.380710659898476%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eAge\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.131979695431475%\" valign=\"top\"\u003e\n \u003cp\u003e54.63 \u0026plusmn; 14.13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"32.48730964467005%\" valign=\"top\"\u003e\n \u003cp\u003e60.08 \u0026plusmn; 12.87\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.380710659898476%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMale Sex\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.131979695431475%\" valign=\"top\"\u003e\n \u003cp\u003e45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"32.48730964467005%\" valign=\"top\"\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"25.380710659898476%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eFemale Sex\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"42.131979695431475%\" valign=\"top\"\u003e\n \u003cp\u003e35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"32.48730964467005%\" valign=\"top\"\u003e\n \u003cp\u003e33\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2\u003c/strong\u003e: Comparison of the mean of the proposed biomarker between nonsurgical and surgical cohorts.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"29.50191570881226%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.689655172413794%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean (mm\u003csup\u003e2\u003c/sup\u003e)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.49808429118774%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eStdev (mm\u003csup\u003e2\u003c/sup\u003e)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.310344827586206%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eStatistics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"29.50191570881226%\" valign=\"top\"\u003e\n \u003cp\u003eSurgical (n=67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.689655172413794%\" valign=\"top\"\u003e\n \u003cp\u003e35.90\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.49808429118774%\" valign=\"top\"\u003e\n \u003cp\u003e25.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.310344827586206%\" valign=\"top\"\u003e\n \u003cp\u003ep-value: 5.290 x 10\u003csup\u003e-3\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"29.50191570881226%\" valign=\"top\"\u003e\n \u003cp\u003eNon-surgical (n=80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.689655172413794%\" valign=\"top\"\u003e\n \u003cp\u003e48.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.49808429118774%\" valign=\"top\"\u003e\n \u003cp\u003e33.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"29.310344827586206%\" valign=\"top\"\u003e\n \u003cp\u003et-statistic: -2.829\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eBy solely relying on the radiographic biomarker generated for each automatically segmented scan, our model was able to prospectively predict (based on retrospectively collected data) whether a patient would ultimately undergo surgery for decompression of their cervical stenosis with relatively high accuracy (AUC 0.79). Of the 147 patients included in our analysis, 67 were deemed surgical candidates and 80 deemed nonsurgical. The Receiver Operating Characteristic (ROC) analysis and the corresponding confusion matrix for our model when tested against this cohort, are presented in \u003cstrong\u003eFigure 3\u0026nbsp;\u003c/strong\u003eand \u003cstrong\u003eTable 3\u003c/strong\u003e, respectively.\u0026nbsp;\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eTable 3:\u003c/strong\u003e Confusion matrix corresponding to the prediction of surgical status using the proposed biomarker.\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAgsAAAD3CAYAAACTkBeFAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAAFiUAABYlAUlSJPAAAABhaVRYdFNuaXBNZXRhZGF0YQAAAAAAeyJjbGlwUG9pbnRzIjpbeyJ4IjowLCJ5IjowfSx7IngiOjUyNCwieSI6MH0seyJ4Ijo1MjQsInkiOjI0N30seyJ4IjowLCJ5IjoyNDd9XX1p8ZABAAA9B0lEQVR4Xu2dfYhdRZr/7/ziXzJRsyKyjIxgIHEwBDT4gjKoqBEzBAccEzSojFEJvsCgKIas6zsZDBsW1BAcdXAkSjITMASNb4vbDJF1xASCshpIYGaFIYivAf/q4O9+yvPtPF1d59zb3Tfpe/t+P/B0nVun6px6e6qeU1Xn9E9+aNMyxhhjjKnh/1WuMcYYY0wRGwvGGGOMacTGgjHGGGMasbFgjDHGmEZsLBhjjDGmERsLxhhjjGnExoIxxhhjGvF3FswEjhw5Uh2ZXjBnzpzqyAwi1oeZw7rTP3hmwRhjjDGN2FgwxhhjTCM2FowxxhjTiI0FY4wxxjRiY8EYY4wxjdhYMF1xwgknNMppp53WuvHGG1v79u2rYhxfNmzYkNKBG/nyyy+TX56uuvDGdOK5554ba/ed2s/u3btTuGXLllU+g8PBgwdbd999d/XrKOSFPJE3MzzYWDAd6aZT+Prrr1vbtm1rXXHFFamTOd689957yb344ouTC6Tj7LPPbq1du7a1ePHiyvdHSuGN6YZoeD7//PPVUZn3338/uUuWLEnuoLBr167WggULirr89ttvJ/eSSy5JrhkObCyYjsQOb3R0tCgjIyOt+fPnJ6Nh48aNKfzx5I033kjpiB3YZ599ltKzdOnSyucopfDGdMOHH36Y3BUrVrQOHDiQBtY69u7dm9xFixYld1CQzl9++eXJjUjnzXBhY8F0RE/hpUFXMOg+9thj6XgmZhZKDOpTnelfWNb66KOP0vGDDz6Y3Jdffjm5JRT2F7/4RXIHBc0eDJqRY44dNhZMR/Qk1WnK/owzzqiOxkMH+9BDD7UWLlw4ttZ74YUXtrZu3VqFmAhLH+yBUHiE36UlEa6j86B14vXr16ffuPzW+nJd+KZ1ZcXJw2hdlz0bnNd1u1m6MYPHp59+mlwMZ5a2mE1j+a1kIOPHzMO8efMmLIMBbQo9ULtBP2ij6EsduV7QHrUsor0EeVpK+ld3L/w4LyNn+fLl6bfaM22d3+zbAIWXbpVQnDxMnhd0iLD98rBhMvjcszGR0dHRMdmzZw+fA0+yf//+cedyaQ/KKdyKFSvG/Ijf7izHrpHLmjVrxl0D2blzZzGsZMuWLePCr127Nvlzf34rHblw3VL4Q4cOjYXRNXNpDwrp/MjIyJgf6VC8kiidZrBRfSNqW7QhfudtKYracduwGOdPe1uyZMlYO8mFc4SJcZCm9ka7xEXXYpxO+pffi7SWwimM0i09qMujhD6D86Qh3ge917VzISzpJpzpH2wsmAlIoZFNmzYlBWawjP65xE5JgyQdhfzoZNTB0GnoujG8RAMzHYo6GK6lDibvePIODJGRQ1j5SUrhlc6SQaROOnaI0aAhXYpXSqcZbFTnCIYwdas2q3ZW0g8ZFrkhofZHnNj2aVNqh3mc2N44p/ZPfOIoXjTUEd0Lfw3ASLwXuhjjSDfza3FP/BH5NeUfkS7E/MjA4v4xL1xL6ZWumf7BxoKZgJQaUeeI0kd/CQqOwqvjQdl1Tk8p+MXBXUK8PA6CHxIHc4kMiXhO4WO40gAvKYVXWpvuqc6WvNR1tBJ1eqTDDDaxXlXvceBVXTMAx7BqU9FfbZ42VdIJGQVx8O3U3miznEPioCwdyPVLorTEOEhpgEd0n1yn8EOiH4LhjH/Ma0xrLEMJ4XSe+KZ/sLFgJhCVV51UNxKNAjoC+Zee1pHYcUR/dbJ0MvHJqyS6Rt4h1k0P14VXB1n3lBWNJfnVdcJI7IjNYKM61eCHTsS61qCcP4lLd2L7l19uWEThPKLfaktN7U1xaN+l8yWRnuU6wn1K11I6tAQjyY1piR40og7LL79GlGi4m/7BGxxNLWw04tXDJtodS6s9kLbaHULrgw8+aJ166qnJ/6233kpuu3NonXXWWem4W37/+9+njWNsDlu1alXjxq9PPvkkueeff35yhTZo5Zsy68KfeeaZyf3222+TC9xv3bp16fjee+9NLrz22mvJ5R7anJVLuzNMYczsgVdxIW87V155ZXK3bds21kalO23jYKz9s6FP+qSNgyXJ+ctf/pLc1atXJ7cJvisS4Z5sbmTzo0SbceveeJDu5K8V170Giq7C4cOHkwvcl/Lg3MqVKyvfH8sI2obHhHxLlC7TX9hYMLXoLQg6vLbFXxQMhGeeeWZchwB63fKyyy5LbhNcP8LOcTpmDBCMEYwGBl86wvxLjPqd7zZXh5N3nnXhZVQo3fDiiy+mzp2OLRo87syGk7pvD2AgYzDD9u3bkyvdiYaF4k8WDd4XXXRRcutgYJaxDrxZcOmll6b2S5uVAEa8+NnPflYd/TjIA3qXo3Tkr4GqPGL+nnzyyeSiQ0LXNoOJjQVTy1//+tfkXnXVVcmdCuecc051NJE333wzufmTmsAAwRgZGRlJr6oxcPOFyDjD8M477yQ3dqSxw4udJ5TCw9y5c5Ora+M+9dRTyZC59dZbk1/OoUOHigZUlPvvv78KbQYdDZalbw/ccMMNyX3hhReS+/HHHye39FEjDN9SW8klp/T6JeijUHGA57XMzZs3p2Puhw6pvX7xxRetu+66K52DeF0N+OhbpOk10JNPPjm53333XXLRP4wSrlH66BnpzPNaEn8wrb+wsWBq0cDazexAHZ9//nl1NB4GY3Vmt9xyS3L1nnj+PjadBl9c5MkJg+Hdd99N/lyDDgxiB1a31FAXHvRbA4JmFR544IEJBoeISxZm9qOn8gsuuCC5Edoo7ZP2w8CqdlT6NomWIiZL3fcHnn766eSee+65yYWXXnopucwqPP744yl9sR3LUM+NgrqlhrolGNADgfKsWQUt4eV888031ZEZJGwsmCJ6koBO058lNG2/Y8eO5OY8/PDDqdPkSUVrvuLvf/97dTSeU045JbkapOMHciJ1Sw114YWezMg7swp0/qVZBYXTlHOOPjTT9KEaM1ioTeVT/ZHbbrstuRizWoaIy2AagHnq1wxWhHvQbthTEM9zTyi1N64lI6ZkmJRmQbiPDPU4GwEa8POlhqaZEuWRNGtWgWWOfGZA4ehXSksSxNeHo7xk0YdUGx2NGWN0dPx73fyerLCTWfF5kyC+JdEerMfOxV3h2lWOsPNau8iJy2+dk7/88p3Vun7chY3UhZcoXl18id6GaBs6496eIC+K2+6Ex/JsBptY5/GtmFz0toTaQHuQnxCGNqMwase4tE2dq2u3nNe5XCcQtTeEdOJHO9RbCpwnvu6DcA3FQeSvtEmUp7q3OBRP4fL4Er0NQdnobQulCz/OxTI2/YONBTMBlJQBVcovxZ2s6Bp1UhqM1dnUSYyjjie/Tn4Nna8LL4mdb6mjj0InrLC5REMBMYMNdajBNxqHJVEbQzjOz0cjvCSltklbqmtvaut5e8VAiEZBFPw1MNcZCxIN+rpWnREQda6TQVWXLiSPa/oHL0OYIpraLE07dgtrpe3OcdzO63YnlXaOtzuNCW9QAHsT2h3muKWCdueSrsEmrRinbsqUVy/bnWs6Jq7O14UXP//5z6uj8bu4S7Dxst3RjpvG5bg9mIx7hdTMDrSs0LRhF6699trqaPweAnHNNde02gN5as+0TcBFJ/Av6QRtCb1go6Li0NbQE+33ie0QWIL7r//6r6Lu0T7lr02JgmsSDnBZTmRZTkuGda9By58wjz76aDouQTiWA0mH7gOkh76CN6tMf/ITLIbq2JjEkSNHqiPTC+bMmVMdmUGkn/WB1yPZf4CRescdd1S+swfrTv/gmQVjjOlT2CTLhj8+rJTDhkA2OEK+SdiYXmNjwRhj+hR9w4DZg/iGAG808KoxywMs2dUtDxjTK7wMYSbgZYje4qnUwWYm9YHZA145rPs2A3sV2M8wW/fIWHf6B88sGGNMn4IRUNoQyDEbbGezoWD6C88sGGOMMaYRzywYY4wxphEbC8YYY4xpxMaCMcYYYxqxsWCMMcaYRmwsGGOMMaYRGwvGGGOMacTGgjHGGGMasbFgjDHGmEZsLBhjjDGmERsLxhhjjGnExoIxxhhjGrGxYIwxxphGiv9I6pFHHqmOjDHGGDOb6WbM93+dHAJoCPfdd1/1yxhjjDnK3Llzq6N6vAxhjDHGmEZsLBhjjDGmERsLxhhjjGnExoIxxhhjGrGxYIwxxphGbCwYY4wxphEbC8YYY4xpxMaCMcYYYxqZ9keZjhw5Uh2ZXjBnzpzqqHdM5aNMJ554YnVkzGDw/fffV0e9xbpgBpHJ6IM/ymSMMcaYaWNjwRhjjDGN2FgwxhhjTCM2FowxxhjTiI0FY4wxxjRyTIyFE044oVFOO+201o033tjat29fFeP4smHDhpQO3MiXX36Z/PJ01YWfzSjPk5Xdu3dXVzDdQHmVyjGK9GXr1q1VrOOP0pLXL78feuih6tdR6sIPMspTNzLdvkL6t2zZsspnOCC/eVnmcuGFF6Y2d/DgwSrW8aWubjR+5G1+ttRlz42FbjqHr7/+urVt27bWFVdcMSMV/t577yX34osvTi6QjrPPPru1du3a1uLFiyvfHymFN+Z4IX1ZtWpV6+677658Zx50/dJLL2199NFHlY8xxx7a2/r165PRMFMPnCVuuummNH7MVnpuLLz//vvJXbJkSWt0dLQoIyMjrfnz56dOcOPGjSn88eSNN95I6bjkkksqn1brs88+S+lZunRp5XOUUvjZzv333z+uziQqH5S1dH6YyqjXoBelMt2/f39rzZo1KczmzZtnZIZBaem2ficbfpCoq6co6I+ZOvQzpXJFdu7c2Zo3b17qr6+//voqxvFDfSPjQjdMNny/0nNjQU/hpUFX0IE89thj6XimppJyopFjTD9x1llntZ555pkxnZoJA9uYfuGaa65pvfbaa+n4wIEDrV27dqVjc2zpubHw4YcfJrfTlP0ZZ5xRHY2HdR/WoxYuXDhujarpaYrpUNZ0FR7hd2lJhOvoPBCG3zwpAy6/teZYF75p/Ulx8jAYRkwjswbNeV23lM5BR2uP5I36jPmN51XOOU3n68pxtncal19+eXJL0/5Mx062TGin6JbCo3NcozS1qzBqq9QPSxDw9ttvj50XpfD8fu6559LvEtL5PM1cg7zomuSxLp39Cv0aeVc5SCh/2jjnu0Hr4nn/iI7VXaMuTlNd9Dtxxurjjz+ujo5C3vK23VRGpT6FuiqVEWWp80D75Dd6AOgFv9V31YXnXnVoDCEPkRmtSz73PB1GR0fHZM+ePXw6Osn+/fvHnculPSincCtWrBjzI/68efPGrpHLmjVrxl0D2blzZzGsZMuWLePCr127Nvlzf34rHblw3VL4Q4cOjYXRNXOZP39+Oj8yMjLmRzoUryRK57Hg4Ycf/uG7776blCjdubSfblN6VR51onBLlixJLr8pF9Vhp+vUnaec1EZwCad7IKU20s9CG1HaY3spSWyr0T+2LZWJ2iBSKhP8dJ7wSNS9PC25P3qhctc9kbrwSiNxFCaK+g6uFf2lfwh5yvMmvUFK7bgXouvrnp3qKRf6jNhGVVYxH7HsENV19I/XiWWua+BHmHgdyjW/T4zD9fI4MylKG27pfBTlIfYReVlznJcRZRKvE8cdtbG8vmL4vG6Iz7GuoXuqbZbqUnWicSYXxkXOb9q0acxvsnVZast10g09NRbIGIkmQ9E/l1g5KlCMi1jYUkgyr+vG8BIVHp2fCoprqTPkmrEA1QiiwpMehZWfpBRe6SwZROoUY8OIBg3pUrxSOo8FM2ksIFEhVBedrlM6H9tIHo/60bm8jfSzkG7SjMQ2VhKVSTSwY/zYsehcqbzU3tGdvA2rPdLuo7/uEdOoe5OuGBbJw1Pv8ivpjYwCXPlJb8hD3qnGPkHXK7XjXojuqfvFMuhGYpnGvghRf4FQL/KnvvCLZas8U//xOhyrn4rlh8g/j0OZ6Rzpi3FmUtTGS20qitoeEstN8clbbGfkPe9rdU4Dc96nxHEqtr9S3SC6d94+SuHV3ktlT9o4h8R0TrYuS225Trqhp8sQ//3f/53cq666Krk5TBsyhcJbEGxOaWewtXLlynSOKSD5sRFE00ynnnpq64477hhbJsjXa1mzghtuuCGFBa3xtjvDdM1PP/00+YOmcOM01v/+7/8m9/zzz09upBRe4f75z38mN/Lv//7vyf3973+fXKaNbr755nTcVvaULtIHSid5Jp3vvvtu8p9NtBUkrTEK1dFUoO4pp7ayTNhARv1QlqA6mC1oGl7TnOy6Fs8++2xy251E0pNILJOnnnoqufDWW28lF/1QWxSPPvpocqdTTyW4HmmEUjvftm1bcuOGtd/97nfJffLJJ8e1ISCvut7x3sOhaeY60dKL+Oqrr9KGvD/84Q8TypX+D/0H9UN1aJ3+3HPPHXcdjvlncdwjwlQ2/Rf1/PTTT4+LQ72/+uqr6ZhNs/2yd6wTpJN8/frXv06/KTu9vcb4Ih0hb7Ftk/c4Jmzfvr06c7Tt5UvnXJc2Rpz2gFr59obf/va3ySUvOdIP+jnVWT/UZU+NhXfeeSe5JLikROedd156tSQaBZBXciwIoYrM12sZjODWW2+dUPC84dC2sMYGeimxlFNozUtrwqIuvBrhJ598klzBuhHGCw1MDZhGqfzmnbn4zW9+k9x//OMfyZ1N5GU6HVS/cbCMyPCkDgZpPVvUDUL4q0OjbcWBU/51bYsyYRChDao9S5fQOYz32Lmge+jMsdi5/atf/Sq5L7zwQnIFexSoMzrC2PHrQeC6665Lbg4PCFDqcPuJV155pfXFF19MeCVbqL/rpP/SJQw/8hzX32kT3OPxxx+vfFqtHTt2JDcOOhH6MfWf/fagQtss6cKCBQvSK8S0ZxlgQkYw+Y2GQkT/fVeGF6gM7r333gn7ZShPxhH1Lb2C9DEmkI/8njJ+r7322uRCP9Rlz4wFOhwy3gSFQ2e3ZcuW1gcffDCW6W4quQ6e4Olk6FhoRGz8oAMsbWTR4J7PIMgAyS3LuvBnnnlmcr/99tvkAvdbt25dOqbRCTVK7lFq/AgGlOmM2hdPUWwWKok4fPhwdTT40L7Rm5GRkbGZAogGUd1ABGq/euMH41kdC22PDlib5PKn4l7CgEYHjy5EA+X1119Pbvw36rH+MA5LdS0969Tv9BrqAYOqTuIsZA79BGXMYE8/RT70oNQJHohk+NHXnX766Sk+Dykl4/ibb75JLgZlLLco2pAe+7J+h7bLLC0zxrHd7927N7nMvNRxzjnnJDeWufpt2uXy5cvTxkNm8qijY/WUDqtXr06u2j9wP9JBPUcDpR/qsmfGghJKJksKhGAg0NnlVppet7zsssuS2wTXj9BYsPwwQDBGMBroAPnAUq5A+p13rGo4xInUhZdRoXTDiy++mJSY5ZJo8HTbEZjuQZko15IMMnWDEO0bvckHoekYRC+//HJqqxgiQJnym1kMDO6mtyimA0YPxKcfzQxceeWVyc0p1TNCmgcF7WBngKeMGezpp8hH3qfVwcMVfShlqDjEv/POO9OsLQNGyWigT4zlFuV4G1rdgkFQ0gWEWS9m0vInbA2okwW9Qvd4WAXKhEGZOsKQZom89PA5XTRjFmfGpBd1MxkzWZc9Mxb++te/Jrduv0I3yOor8eabbya3tK8AKFwUiUqnoVFw7I2Ilaxlkosuuii5EJca8sZXCg9z585Nrq6Ny9QgCoz1X+LQoUPFhh/FH3Lpjm7KsunpbragdtgtJ598cnX048BDe8MQ2b9/f3pSU2dJh8QT1rGYZdCeBC1FYJSgq+hsaVYRvSzVby79DIMNhgHlSj4xyni4kXFY16eVoIwwHFly2LNnT7oWZQQMGHmfBzt37pxQXrnMhr7nlFNOqY46kxto9BcsF1EWlBf1JUOaZfW6pc/pgA6ic7R/GefSCy2x5cxkXfbMWNDA2s3sQB2ff/55dTQeGj8VBrfccktysaKZwsdij1DpWJ5UNJUgS41roKwQZwrqlhrqwoN+68lGswoPPPDABINDDNI0Xz+Qd3ggBf/b3/6W3GEntsvSE6XQrF+dMc4AxJManSWGmDpJLVv0EtLM9dEdplw1BSu9FjKECFdqC4MC9aK+i8GdvonOnIeb6Rq0lCXX4iGJawP9kDZ0a/A8FvXYj2j5QcsRJer6+wjLZdqrgBENGGLHAu1LQA+0BIF+5G2jH+qyJ8YCmdTAmj+Fd4OeKLSJI+fhhx9OSsBgkU9V/v3vf6+OxqPC1SAtBcKyj9QtNdSFF7LmyTuzClRwaVZB4eLu24g+NpMbPbMd1U9pl7GUJkdTc+wGLkE8ypL1d46HAc0G1H2UhSlO6Y46IH30rBQHY3c6s4PdoL0JGPJ1SxAyKgBjvARxqW90qF+JS0V5HwO0024HIh6QWE8vGYaxvIQGIoyVOoOLdnAsl52OJ1dffXVyt23bVqv///Ef/5FcvU1BWapcS9QtjfUK+jR0k7asB1vpdKQf6rInxgIWmCgpRCc05UIlx/UhVaQs8z/96U9jT+56EuEcA60aB3H5rcFGlV33OWfFi1O0UBdeKB165ZPPV5dmFbSJBYMids5UKHkjz9yjbvlithIbf+z8qI+6KTjt+KdzzdcRYzzqoTSlPRu56667kks55oM/ywiUEzDrJRYtWpSMezZ25UsN1IUG8HzDbx16UOgW6SRr7egOa/Al3ZFRwZRwU96aNrTNNHGpSOUqyENdWy9Bm6a8br/99gmDIdemHhh4tPeKcsaAIA59TYyD7lB+xOHVzgsuuKA6M7gw9ujhjnKtyy9lpP0CxMGPMsr7FNCbCXUPjSXqZsjrwGDg/ugD6LXKSF/UZfW9hSkzOnr0AxPtAh37IMRkRdeok9KHdrhfKawkxmlbaxP8StfQ+brwkvXVhzaQdiUWw0jaxsBY2Fw4Fz+wcSyYyY8y1YUjz7FcCK84bWUe+4BKHp/6UBzCESdeJy/PfpeR8HEZjkthOkmpTGiT8qMs8zgqa4Qyy8sxjyP/mEbKWf66RlP4KPH+O7MPLkVRO0DIU1PeSu24F6Lr656TradSWSsPuOpr6AMVR/1LLFPKm/rVtTiXl0feX8UPC8U4+s05wsQ4MylKG27pfCfJ+xWVd1N+ow5KfxCVWx6nVDeIxjBdQ3VRF14S7096S2GQydZlqS3XSTf0xFhQ5dQNDN0KnYYUB0EJ6Azil7hyoULyAuMauUJLofKGwm+lPxZ2XXhJ7KC76Twom7wR80W2PNyxoB+NBQTFpn5V1pQ/v/GXgpXi0x5iPITyJCxx8/D9LLGj6KYd1QlxKZPYmaAHTQMx7S/qTlMcnc/TiB7EepC+1IWXSH9Ib+l8FK5BumLeSLc6Y0mpHfdCdP1OeaoTtee69kp548d5xVH7J5/5tRiUYl8ivanrq3T/GId7Eaepb50JUXvM8z0ZIb+07Ty/lBvnSnEou7xPqYvTVDdx/OK4KXwU3bc0JkSZTF2W2nKddMNP+NO+4ZQ5cuRIdWR6wZw5c6qj3sF3CeI77N1w4oknVkfGDAbff/99ddRbrAtmEJmMPnTzZlXP3oYwxhhjzOzExoIxxhhjGrGxYIwxxphGbCwYY4wxppFpb3A0/c9UNjgaY4wZDrzB0RhjjDHTxsaCMcYYYxqxsWCMMcaYRmwsGGOMMaYRGwvGGGOMacTGgjHGGGMasbFgjDHGmEZsLBhjjDGmERsLxhhjjGnExoIxxhhjGrGxYIwxxphGbCwYY4wxphH/I6khYCr/SOrEE0+sjkzk+++/r47MsGBdKGNdmD34H0kZY4wxZtrYWDDGGGNMIzYWjDHGGNOIjQVjjDHGNGJjwRhjjDGN2FgwXbNs2bLWhRdeWP2aOXbt2pXScsIJJyQ57bTTWnfffXfr4MGDVYhmvvzyyxSHa5TYvXt3ui73MaZEv+hCZMOGDand0n6b2Lp16zj9WbhwYYqLXuRYF4ywsWC6gs7k7bffbm3cuLHymRkwCpYvX57SsmTJktbSpUuT/+bNm1PnvW/fvvS7DjpEOsqvv/668pnIJZdc0lqxYkXr5ptvLnagZrjpF12IYACsXbu2+lUP+rNq1aqU/vnz5yf9+eqrr1Lciy++eIL+WBeMsLFgOsIT+1NPPZU6FjqPmeK5555LRsG8efNaIyMjrQ8++KD1xhtvtD799NPUoWEAPPjgg1XoidARYih89NFHlU89TzzxRLrePffcU/kY0z+6EMF4wQDoBOHQH9i0aVPrs88+S/rzxRdftNasWdM6cOBA6/bbb0/nI9YFAzYWTEd4gqKzWLduXeUzM+j+r7322riO+tRTT209/fTT6ZgnpvwJiN90lOedd15XhgKcddZZqQPdtm1bx2ldMzz0iy4A7ZLZtG5mFAAjB9avX9+644470rF45plnkgGEfjBLEbEuGLCxYBrhSYqnEab88ycpBmDWM3EJxxQnewG0DsrvXk1dsmZKJ11KB2AwjI6OJuE4ctNNN411qHSUSDeoQ3322WeTa4abftEFYNC+9NJL0+BOevbs2VOdKcOsGvoDt956a3JzbrnlluTu2LEjuRHrgrGxYBr54x//mNzVq1cnt8TevXvTEw4d6fnnn5+eUJjS5DfT/rGTVKfarYiPP/44ub/5zW+SOxlOOeWUZCzs37+/df/991e+nVm8eHFa1+WJigHADDf9oguCa2/ZsiUtx9FWmzh8+HB19KNhXeKMM85Ibmn2zbpgbCyYRugc4KKLLkpuCcIwVclgzBoowpMOewvoeLZv316FnDrvvfdechctWjTpJ7dXXnml9fjjj6c0Tparrroque+++25yzfDSL7oAzGxw7ZUrV1Y+0+fzzz9PLsZNCevCcGNjwdTCoKyOo9OTy6uvvjpuMCY865wQd1jzZK/lgm4k5//+7/9qn9zYzd3rp55f/vKXyWWfhBle+lEXJgPGBQYL5HsSxEsvvVQdlbEuDDc2Fkwt7JYGBuQmWDMtPbWfdNJJye3lAH7nnXdOeHLjmDTQmTPD0Es0Nfvhhx8m1wwn/agLk+WBBx5ILjqSb1RkSYTNwU1YF4YbGwumFu0T6ETdGuixgKcjDITYIXPM0xzQ4eXvik+Hs88+O7naHGaGk37UhcnCTAbGDm2ZzZHM0LGPguU89vSw/6EJ68JwY2PBHFemu6mLddNSh4zBoKe+t956K7m9oJ87fzPYTFcXpgKGNm8DMQPCHgqMa3SK75awrAdarsixLgw3NhbMQMAbDXDuuecm1xgzNZhh4A0K7YVgAzB7GrTUIqPBmIiNBXNcmeqmrssuuyy5eiuiiZNPPrk6mj69fDfemMjx3ODYDa+//npymXUoYV0YbmwsmFp4TbFfuPLKK5NbtyeBjWPaoKWwvYBPSUPd1KwZDvpJF6YK+xNYzih9hRH90VsS119/fXJzrAvDjY0FUwvfL4BOu6SPB+xJ0OtnfL8+Ggx0dDfccEM6JkxpN/pU+eSTT5Krd8zNcNJPujBVLr/88uTee++942YJpD9sXER/6l4NtS4MNzYWTC0Muny1DXr5hsFUefTRR9MmRjZm8X8etJt7wYIFyY/pU8L0EuVbyyBmOOk3XZgKfOZZGxtPP/30pDuT0R/rwnBjY8E0cttttyW3l28YTBV2Y7Obm1e84m5ujtnhzble79jW1GwvlzbMYNJPujAVpD/oCoYPuiP90Wejm/THujDc/OSHNtWxmaU88sgjrfvuu6/61R0nnnhicpmi5MmDzkW7pYcF/nnV8uXL09Qs/5UPvv/+++Sa4cG6YF2Y7cydO7c6qsczC6YR7RXg64jD9u9ptTtc+yHMcGNdsC4MMzYWTEfYEMUO6CeffLLymf3wFMn/m1ixYkXxX2Kb4cS6YF0YVmwsmI7wRMV35VnfHJYnqo0bN6ZB4Yknnqh8jLEumOHFexaGgOnsWYiwc5pXrtgINZthEODb+Wz6yv8FsNdphw/rgnVhttPNngUbC0PAVIwFY4wxw4E3OBpjjDFm2thYMMYYY0wjNhaMMcYY04iNBWOMMcY0YmPBGGOMMY3YWDDGGGNMIzYWjDHGGNOIjQVjjDHGNGJjwRhjjDGN2FgwxhhjTCM2FowxxhjTiI0FY4wxxjRiY8EYY4wxjfi/Tg4B/NdJxBhjjJkKnlkwxhhjTCM2FowxxhjTiI0FY4wxxjRiY8EYY4wxjdhYMMYYY0wjNhaMMcYY04iNBWOMMcY0YmPBGGOMMY1M+6NMR44cqY7M8WbOnDnVUTNT+SiT69UMGt3qw2SxLphBpNf64JkFY4wxxjRiY8EYY4wxjdhYMMYYY0wjNhaMMcYY04iNBWOMMcY00nNj4bnnnmudcMIJSTZs2FD5ltm9e3cKt2zZsspncDh48GDr7rvvrn4dhbyQJ/I2G1BddiOd6rsTxOc6g9gectQOTjvttNaXX35Z+ZYZZD0g7Q899FD16yhqE7NBD9QuJyuzpQ+YLmrfSDd9xKD2oeg5+cvTPVv6tZ4bC/v27auOWq3nn3++Oirz/vvvJ3fJkiXJHRR27drVWrBgQTIYct5+++3kXnLJJck1w83XX3/duueee6pfsws6xUsvvbT10UcfVT7GNLN27dpxY8Rs4qabbkr5m6303Fj48MMPk7tixYrWgQMH0sBax969e5O7aNGi5A4KMnIuv/zy5EZGR0eTzDZGRkbG8lYn999/fxXaRLZt29aoB7MRtYnZYDTTrmM7lyxdujSdX79+ffG8HxjK3H777dXRcKD288Ybb1Q+g0lPjQWmYfSU8eCDDyb35ZdfTm4Jhf3FL36R3EFBsweDZuSY48+8efOSe/PNN3dcjjBmtoM+0O93sxxh+oueGguffvppcrG4Fy9e3Jo/f356qipN1+PHzAONh7A5W7dubV144YVja10LFy5MDaypw2Va9MYbbxyLwxqRpry0Dpanheux5sr1O91La08ycpYvX55+a42KPQz8Zt8GKHyTYihOHibPC2vfhC2VZb9C+VEWKnsJ9Voq3zoIR/hYR1yDequ7Rl0c1c3xYs2aNUkPprMcUdcWmqZz8zjkHZ3Cn9+l9VPaFmUa9U5hiRvBjyUIwHhWWKHf0g21gabyV13lszBTyX8/obyTD8pX+SBP8XxdP9F0njqjLCiTeN1+ncl68sknkzvV5Yip6HUeJ/alsW5yaPOUZSxbrpH3w9IpPUSiF/xWfeHym3uBwnPdOrg3YchbZEb7NT73PB1GR0fHZP369Xw6+od2Q0i/cfmNfwyH7Ny5M51rGxbj/A8dOvTDkiVL0rmScI4wMQ6yZcuWYnhkZGQkuW3DZFycPXv2JL88vCS/F2kthVMYpZv78bsuj5L9+/en86Qh3qc9wIxdOxfCkm7CdcvDDz9cHXWP0qL7Kk/dSl6PlAHSHjjH+cU4aj/RP16HvOs6ukZedgjlk98nxqlrQ70U3Y88qf0htNM8rM4TJz8nHULIE2Fi3krXi7qgOGrnKsv8XlEXYjnLD4l6TLpK9aLziqN2ozQRR2GicH9dK/pPJv/HipieKKSFNJT6tygKF8uePKDn8XzdderOk/e8znQPRNefaYntn9+xPEp6qPN5nzMVvY79h+LoN2Wm6+X3in0w4WM8xdW9SBfnVRcKr7ZJveGPn66v+zJGyC/KihUr0vlNmzaN+U02/72mp8aCMqhCUgdABmM4RAWYK4AqhDixI6BQVRl5HA3KOqcCkzIpHumL8XQv/Emr/OO9YmUh/C5di3vij8ivKf+IGmTMjzpH7h/zwrWUXjW6bpkJY0F5yxswQr3ourHcyS9+UaliecfrcKzyoMzkj8R6jXEwznTuWHekUmLVbaxX0hHDqjON+UbUromTdyoqFyReL3bMUX8oh9gB5vdSJ1QqF8UjHdG/Lt2I7qN2w/3ll+cfUfnEupxs/o8V8b5R8jquE4VDYj7UNjtdp3SePFMupXiUuc7FNjBTEtskv2Pac91FlF+1HclU9FrX4nyME8sov5f6J87H/gnht+LlY0Nduqkf/DkvP7X3kr5FXYlpnmz+e01PjQUVYixgZSJXdhVs9Feh0nHFwpCo84iDL+HqKg+JDTUqlRoE6YvhJUpLrojqOHN/3Sc2CAQ/JPohVDD+Ma8xrXkjRQin88Tvll4YC50kVxAadEnZJGoXsTNTmccyVDvJyxuhPXCP2OGoXuvakMod4Tg/3yvJ001aNCDnbaSu7Sh8qV0jaouxg6Dc431zKaWBcuA3ZRnDSvJ2J/+6dCMKH9uF0lvKj9IV28tk83+syO8rId/cv66sJQpXKqd4vu46pfPKO/Udw0qiHpTOH0+J/Zr8qFP55X2H8hv9p6LXtCV+065LcXRNJN6LcuY+de2ubqAvpRvhevjH+ld6SzqndMW6nUr+e03P9iywhsO6bDvz4/Yg3HvvvcnNNzrqrQnWXsRTTz2V3P/8z/9snXrqqek4cs011ySXvQ7ixRdfTPdtDz6tO+64o/I9StyRfPHFF1dHrdbKlSvTDtUPPvig8hnPe++9Vx2NR+mO14K610DblZvcfH3u3/7t35L72GOPjeX12WefTW67MRb3cRCu3eDS8T//+c/k9iuvvPJK64svvijmA5Tnf/zjH8mtQ2+c0DZYx4t7FGgP3OPxxx+vfFqtHTt2JLetaMU2dNZZZ42V4bvvvpvc4wFp+fOf/5yOWdvstMZIe1E7v+6665Kbc8MNNyQ37ifYtm1bcuvi3HfffdXRUSgTdmpTliViOU6n3f3qV79K7gsvvJBcwfo6eUVX1F6mmv9+pvT21FRRnnldrwT9G1CG/bi3g74aHYVbb711nF6XmIpev/XWW8mlLEpx8Ge8yuHthc8++6w4nsBJJ52U3LhvYbKQXsYKxq58f8nGjRuTe+211yYX+qFf65mxQOHC+eefn1xx5ZVXJpdOTA0iGhZkEtj0gR9o42BJcv7yl78kd/Xq1clt4uyzz66OfkQbjth4ItFmlro3HrS5MX8tqu41UBkLhw8fTi5wX8qDc1JqUEfftkQn5FuidB1v2tbyuNfCcml6TYx6J890cGzOoZy7zQcdCe2EtrFq1arW6aefnuIz2JY6wW+++Sa5lGWs1ygy+L799tvkHi8YCKlbWLduXWNnE9sLA0IpHzLEpTexPKRXOeecc051VA/por4oY2147AUYd9QlOhTz/vrrryc3GjJTyf8woTzzr+dLZYOIWJb9xNNPP53aAwYND4hNTEWv1SefeeaZyS2Rj1cl0AUGdPouNjzqoXa6aMxS+wf0Av2gXOLY0A/9Ws+MhbpvD2AFrVmzJh1v3749ucpUrCjFnywavC+66KLk1sHAHC0ydrSya5XOm4FLArJ44Wc/+1l19GOjgXz2AJSO/DVQlUfMn3YEMysidO3ZhHbtMsBT1gz2zJpQzihDN1BnzP7QhhSH+HfeeWfrvPPOS0pSMhrogGK9RpnJwYWnFj1R0Aa7oZQHRG1OTGdQwKDTrno+OEZ9UcboR5NRM1nUF8SnHz0l68Eip5R3JM//MEIZlMoG6XfQ7T/96U/pmHbWTR84Gb3WADsVMA70VhC6wAMsfReDda/QjFmcGZNeREMhMpP9Ws+MBSlu6dsDmi7U9OPHH3+c3NK0HBVSenLNJaduultTPHGAp3I2b96cjrkfT82HDh1K12Uq9q677krnIF5XA76mewSdKZXIYJan4+STT07ud999l1wUgorlGqWncdKZ57Uk/f7BFwYeypZyIa90Blu2bBmboejGohc8JT/zzDOpbvbs2ZOupfqkLK+44oo02EV27tw5ocxymamPSP3hD39ILmnvtBzRbXuYDpQdRpd0AmOZMqYMKe+65YmpcP311ydXfQH6SSdHGynNhhyP/A8y6reapJ/7CmabJrMccTz0mvEB44AxjfZHP7Zp06axceKBBx6oQk4PjCXyTvvXOCW90JiZM5P9Ws+MBTo+uOCCC5IbobHyZE/ha5oF8nV/mKp1VPf0w1QXnHvuucmFl156Kbl0iKx3k7446/Dmm28mNzcK6pYa6pZgQNO+yrNmFZiGLjEda7hf4ElfAw+DDevhNGCs5el2XBhjXIvZBq4NtBl94+OUU05J7lRnqo4H+XLE559/no4jc+fOTS7tplMHKmLZ1q1Tf/LJJ9XRUZjx4z7oKOXIfhPKmI68zgifKlwv9gWagr3llluSK6aS/9lKKf+aZfvb3/6W3EGmm+WIqei1HkbVb5fQLHdEM34YCPQzjBHsX8jHiV6gfQnogcZG9CPvJ/uhX+uJsaCOKZ/qj9x2223JZZpFFRT3EGgAxqorKQf3YEqIadJ4nnuCljgiXEtGTMkwKc2CxIFOT6+CioR8qaFppkR5JM2aVcCazBuDwqEwpek44utDHN1M180kcTq8NNigFKqXTvDES52XBj8NPBEpH3VYakdAOSKy5mcCBmPaF4ZOaTki5i0uV0Vo37QHfdwH9JSmzV05+eZC0Bpnnf5yn16ivQn0Bbp2vgQx1fwPIhoINPsYiQ9XEU1T62Eoh3iUDVPpHPcztLm4HFEawKei11dffXVy33nnnWIczWrlyK9uf0+n/3k0GahHDCXaspYgpMORfujXemIs/M///E9yr7rqquSW0PrMa6+9lioj75i0+YlzbGhSA8dl7ZupZmA6OsaTEaLd8kBhEoc1chENE013sjlIgxBxiM991Fi061UwkMNPf/rT5Aq9OVEyPpRWFF6zCk888URyI4SL03EyCJQujB3uz5rvdJ/OjzV6KgTViSBfdVNsJagr6oPvyeedHtemTGg3ql8GHdoWcTA0YhzKkoGZOF999VVxFux4ouUItbccDapMg+bLFZSjjIw4a6YlNOLEslfeSwOPlsropHOjjGuUjJmI9KJbZBiwJ4K806ajToup5H8QiQNBLH/abp2uaKc+RjflQP2KGI9yrdvs2k/E5YiSPkxFrzE4mR1WnFhGtB8+wV5CszavvvpqcgX3xDDt1N5LM4VNYDCQRvQBfvvb3yY30hf9WvUK5ZQZHW1+fzpKuzGMvQvKcX5+Z/UdhTrhXdM8Du+ctp/QiuHbDSW57UIeF4f3b9sNYkJ4BH/Cc9y2csfFy8PqfVZdS79zUToQyqoUBiF+XbqQPG63zMRHmWKeqR9+q1xx1RbaA8FYHMobP8LKj/qNZcK5eC0kbxd5/SqOfnOOMDFOr0X3y9tQLsqz0pmfl24h5DnPe6k96Z3sGEflIV2J96orY/lxrHjoaIynOKpjnZN/XbshrMLEa+YymfwfK2J6oigPneq4U7i8DyO84lAHKoM8fqxnwhEnXodjrh3jzITQBpSm0nlJ3g7ztjMVvS6VrX5H/3ivqJNqcwob6wO/eC/6MoUhjvolXQ+/GF4Syye/ZpTJ5r/X9MRYUEHWdQyS2LjrFIfMMpCoUFQ5eSOIQoOgohSH9HAv3a9kmOg+Sg+NgvswYKvS40CGcD11VLj4EZ7f3DuGjaLGRZhOyst5wscOkXSWOtRumanPPVPHMR/UC36ck2GockTqlEr1q3aGdGoXun+Mw71Ux6U4vRQpcV07j6I0NnUmUScUVp1RSSjf2JFwD/zqypgyoWzq7iGdIEyMF3UCUX3od127kW426Y2k2/wfK+I9oqh8O9VxN+FyvVf7Vjuui696q9OzPPxMCPWntJXOR1G/gJTazlT0uqls6+5F26q7R4wX74l/HFM4xl/1l+tcFKWt0wP3ZPLfa37Cn/YNp8yRI0eqo/6DqRmm9tioUveBjUFmzpw51VEzLLcgk6Gf69VMHZbn2gN/mvpnSW820a0+TBbrwuyFfR3QHmgHYrlmMvRaH3r2NsRMQMdHZfPhmBzWcbRmm2+eMma2wnomoj0vOdpfU9p4asxsAh1gw5/2tuRIR+bNO/pxQFPPQBsL2pjF7EHsHNkkRIfJZpClNe9vGzMboa2z6Y2vG8YNXcAmQc7ROdZ9QtmY2cK//uu/pk1/jA/5Rms2COoLoL36bsJsZ6CXIegM2QWPUVBiyZIl6R3/0k7r2YCXIUxOrhMYy8CbDvLbsmVL7RfiBhkvQ5gcLbvB/Pnzk6AjeitoNo8RXoYIUMF8RIb1VxqB4Hj9+vWz2lAwpoR0gvZPR8hMAvIv//IvSU9Ym52NhoIxJfieycjISHotk9cK0QUMBYxojGY+uuQxojumPbNg+p+pzCwYY4wxYqBnFowxxhhz7LGxYIwxxphGbCwYY4wxphEbC8YYY4xpxMaCMcYYYxqxsWCMMcaYRmwsGGOMMaYRGwvGGGOMacTGgjHGGGMasbFgjDHGmEZsLBhjjDGmERsLxhhjjGnE/0hqCPC/qO4dx+rfIJv+xbpQxrowXHhmwRhjjDGN2FgwxhhjTCM2FowxxhjTiI0FY4wxxjRiY8EYY4wxjdhYMF2zbNmy1oUXXlj96g82bNjQOuGEE1q7d++ufMps3bo1pZ+wyMKFC1PcL7/8sgpxFK5FmF27dlU+xoynH3WBtnzaaad11AfO3XjjjWO6QJy77767dfDgwSrEUawLZgxenTSzm4cffrg66p7R0dFxsn79el6x/WFkZGTCuZmSLVu2pDR1SteaNWvGws2fP/+HpUuX/jBv3ryx33v27JkQZ8WKFSnMoUOHxvmb4SPWP9KPuoDQZtXO69IWdSbXBVzrgqnDxsIQMF1jYf/+/amzoGOJ/jMp6rAldZ1jDLdp06Zx52RELFmyZJw/Qp45R0cZ/c3wEeu/H3UBiUYAUtIHtWkk6gJGgHQBAyLGQawLBrwMYTqycePG1tdff91at25d5TNzMC3K9O/atWsrn2aeeuqp5LaNhtYdd9yRjsUzzzzTanf6rY8++igtU0TOOuusVrsDbW3btq1xStcMF/2kC4LlA5YR2gN95VNm+/btyaXNR1049dRTW48++mg6PnDgwIT2bl0wYGPBNEJHtHnz5lb76bt1ySWXVL4/ov0CuOqwtGbKngB+l/YETBU6qksvvTQN7qRnz5491Zky+/btSx073HrrrcnNueWWW5K7Y8eO5EbUoT777LPJNcNNP+lC5IYbbkjt/M9//nPl0wzpz8FgwIiow7pgbCyYRv74xz8md/Xq1cktsXfv3vS0T0d6/vnnp06HJxR+sxEsdpLqVLuVHK69ZcuW1gcffNBavHhx5Vvm8OHD1dGPnWGJM844I7kYIDlcn6c1nqgYAMxw02+6AFyDtsvMWSd9OPnkk5NLe84hXW+//XY6njt3bnIj1gVjY8E0oo7loosuSm4JwjBVuX///tYbb7yRhKf+efPmpY5M05/Thac5rr1y5crKZ/p8/vnnyaVDL3HVVVcl9913302uGV76SReAmTaW45gpuP/++yvfeq677roUlrbOTIcGfWbgMGRgxYoVtUaHdWG4sbFgaqEz0SDa6anl1VdfTZ2kIDzrnEBnJOjURkdHu5bpgHFBJw35ngTx0ksvVUdlfvnLXyb3tddeS64ZTvpNF5gJYGmN9s39uoHZNYwXDAJmOhYsWJBmLM4777yx2YlXXnmlCj0R68JwY2PB1PLZZ58lt2ktE3haiZ2jOOmkk5I7k9OWDzzwQHJ5kso3ZzGFq6nXOrRM8eGHHybXDCf9pgv33HNPMl7YpFu6Xx0vvvji2AwJaSU/2hj5/PPPN35Pwbow3NhYMLV8/PHH1VEzdfsB+gGe3ugQ2QDG5kjWk5lyZfMZU7jsf2ji7LPPTq42SprhpJ90gVkyBnxmCCazJPfQQw+lNs9sxMjISNr3w0wDhtCmTZuS8bF8+fLaNx6sC8ONjQVzXOnFpq7JQofIFCtPUky3MpvA+isdJpvQQMsVOf1sCJnBZiq6wMwEs2S016effjr5dQPLFugAsIyQv83B2w46f++99yY3x7ow3NhYMEMBMww8SWn9l7VZOkxNL8toMKafYYMkT/bI6aefXjQogFk0fmOQwKeffppcyA0FcfXVVycXg9qYHBsL5rhyPDc4dsPrr7+eXGYdSsRX3YzpJVPRhZ///OdpWa1OhPYjEL5bfvrTn1ZHZawLw42NBVPLokWLqqPBhf0JPGGV1mGZ0tVbEtdff31yc/REVrdMYYaDftEF9ijolcySCL40yW/tadB+A6jbk6DXOusMZ+vCcGNjwdTCl+eg0xsD/czll1+eXNZh45MRhoK+fMdrbXWvw33yySfJ1TvmZjgZdF1gv4E+kf7rX/96gsGA0azzjzzySHJzrAvDjY0FUwuvZOm1qvh++CDBu+ja2MgaLzMNCO+Y48c5fRe/hPJ92WWXJdcMJ7NBF373u9+lpQkMZPY0YAChC7irVq1KYTAYrrnmmnScY10YbmwsmEZuu+225L711lvJHTR4omI6lp3edPY8GSIYCfpsdNMuby1TXHnllck1w8ts0QXaPUYDr0qiC1999VV6DZO3gx5//PEq9ESsC8PNT/jXk9WxmaUwrVg3tVjHkSNHkst0PU/hDLR6c2BY4AM1vHfOMgUfv4E5c+Yk1wwP1gXrgvHMgukA0690EDyF1G2Mmq3oTQn2NhhjXbAuDDM2FkxH2BzIDugnn3yy8pn98BTJ9/OZnq17L90MH9YF68KwYmPBdIQnKv7HAuubw/JExatnDApPPPFE5WOMdcEML96zMARMZ89ChJ3TvH7IpsDZDIMAu8XZCJZ/e9/rtMOHdcG6YGwsDAVTMRaMMcYY4WUIY4wxxjRiY8EYY4wxjdhYMMYYY0wjNhaMMcYY04iNBWOMMcY0YmPBGGOMMY3YWDDGGGNMIzYWjDHGGNOIjQVjjDHGNGJjwRhjjDGN2FgwxhhjTCM2FowxxhjTiI0FY4wxxjTi/zppjDHGmEY8s2CMMcaYRmwsGGOMMaYRGwvGGGOMacTGgjHGGGMasbFgjDHGmEZsLBhjjDGmERsLxhhjjGnExoIxxhhjGrGxYIwxxphGbCwYY4wxphEbC8YYY4xpxMaCMcYYYxpotf4/6/uGOTMN1fUAAAAASUVORK5CYII=\"\u003e\u003cbr\u003e\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eIn this study, we describe a proof of concept for an automated image analysis tool capable of identifying operative cervical stenosis using only data from MR images. Our model distinguished between surgical and non-surgical cervical stenosis with high accuracy (AUC 0.79) by relying on an automated biomarker representing the degree of space between the spinal cord and surrounding spinal canal. This biomarker was constructed because it is both physiologic and intuitive; it follows that as the area of the spinal canal approaches that of the spinal cord, spinal cord compression becomes more severe. Moreover, the results suggest \u0026mdash; perhaps unsurprisingly \u0026mdash; that increases in compression severity correlate with increases in surgical candidacy. The success of our model in predicting whether patients ultimately underwent surgical decompression for cervical stenosis is worth discussing both in terms of the components it does and does not include. Currently, no model for surgical decision-making should rely on imaging data alone; however, the accuracy of our biomarker suggests that, in many instances, radiographic interpretation is not only necessary to determine whether a patient\u0026rsquo;s cervical stenosis is operative, but also sufficient.\u003c/p\u003e \u003cp\u003ePrevious work by our group used deep learning to generate a parallel biomarker for lumbar stenosis to predict whether a patient\u0026rsquo;s lumbar pathology warranted surgical decompression \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. Compared to our cervical model, our lumbar biomarker managed to achieve a higher degree of predictive accuracy (AUC 0.88). This is partially a consequence of the inherent differences in anatomy between these two spinal regions, the most obvious of which is the presence of the spinal cord coursing within the cervical spinal canal. As such, in cases of cervical stenosis, variables such as cord deformity and changes in T2 signal within the cord itself can often play a role in surgical decision-making, and for the moment, these variables remain absent from our model. Ongoing work by our group hopes to include these components as part of a more comprehensive model.\u003c/p\u003e \u003cp\u003eAs a proof of concept, our model is currently not intended to assist in the clinical decision making of spine surgeons. Rather, the real value of the present algorithm nd the iterations that build upon it lies in optimizing the referral process to spine surgeons. As any spine surgeon can attest, a substantial number of patients \u0026mdash; more than half in many cases \u0026mdash; referred to see a spine surgeon never undergo surgery \u003csup\u003e\u003cspan additionalcitationids=\"CR7\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. While some of these non-operative patients still benefit from time spent in consultation with a surgeon, the degree of inefficiency with the referral process can lead to longer wait times for patients with operative pathology and delays in care for those non-operative pathology (i.e. delaying referral to a physiatrist or physical therapy)\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e,\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. Spine patients are particularly vulnerable to this inefficiency due in part to the relative nuance of radiographic interpretation of spinal pathology. Given both the incidence of neck and back-related complaints during visits to primary care practicioners\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e and the prevalence of non-operative degenerative changes on routine MRIs of the spine\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e, it is not surprising that so many patients arrive at the office of a spine surgeon only to be reassured that their findings do not require surgery. This problem is not novel, and various efforts have been made to improve the efficiency of referrals to spine surgeons with some success. Current efforts include the development of multidisciplinary pathways\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e and clinical practice guidelines\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e; however, no studies have focused on the problem of variability among referring providers with respect to discerning operative from non-operative spinal pathology.\u003c/p\u003e \u003cp\u003eThe present model could provide referring physicians tasked with reviewing a patient\u0026rsquo;s cervical spine MRI a greater degree of confidence when making the decision to refer to a spine surgeon. Our stenosis biomarker could be used to generate a surgical likelihood score, informing referring providers of how likely a patient\u0026rsquo;s stenosis is to require surgery. To be clear, our model would not serve as a replacement for the clinical judgement of a physician, but instead help provide additional actionable information \u0026mdash; alongside a patient\u0026rsquo;s clinical history, signs, symptoms and physical exam findings \u0026mdash; with which to make the decision to refer to a specialist. This additional information could help increase the number of patients with surgical pathology who are referred to a surgeon, as well as decrease the number referred without surgical pathology who could instead be referred for non-operative treatment modalities.\u003c/p\u003e\n\u003ch3\u003eLIMITATIONS\u003c/h3\u003e\n\u003cp\u003eThere are several limitations to our study. Firstly, our model is susceptible to the domain effect since the decision regarding surgical candidacy was limited to a group of physicians at a single institution, limiting its generalizability. Secondly, as a proof of concept our model focused on cases of cervical stenosis to the exclusion of other surgical pathologies of the cervical spine, and so fails to address other forms of cervical pathology warranting referral to a subspecialist (foraminal stenosis, tumor, or trauma) \u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. Thirdly, as an automated image analysis tool, our model knowingly omits any details related to patients\u0026rsquo; clinical presentation or history. Nevertheless, that it manages to successfully predict surgical candidacy for decompressive surgery with nearly 80% accuracy speaks to the relative importance of radiographic findings in surgical decision making. Lastly, our model does not capture those patients without surgical pathology for whom surgical referral is warranted for reasons beyond clinical history or physical exam, such as for reassurance or to gather more information about their condition. Such cases go unnoticed by our model, but are important cases for referral. These limitations notwithstanding, we feel our model performs well in predicting surgical candidacy using imaging data from MR images of the cervical spine alone.\u003c/p\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eThis work presents a novel, fully autonomous model capable of predicting a patient\u0026rsquo;s surgical candidacy for cervical decompression based solely on the degree of cervical stenosis calculated on axial MR images with relatively high accuracy (AUC 0.79). We argue that this model could serve as a valuable tool for referring physicians when considering whether to consult a spine surgeon for patients with cervical stenosis, and its incorporation along with other clinical data could help optimize the process by which patients are referred for evaluation in spine surgery clinic.\u003c/p\u003e"},{"header":"METHODS","content":"\u003cp\u003e\u003cu\u003eIRB and Ethical Statement:\u003c/u\u003e The present study was conducted under the World Declaration of Helsinki: \u0026nbsp;ethical principles for medical research involving human subjects\u0026nbsp;\u003csup\u003e18\u003c/sup\u003e. The present work study was conducted according to the rules and regulations of our institution, as approved by the Institutional Review Board (IRB #16-000196). Waiver of informed consent was granted for the purposes of data collection by our institution\u0026rsquo;s IRB.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eData Availability:\u003c/u\u003e The datasets used and analyzed during the current study are available from the corresponding author at reasonable request.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eManual Segmentation for training network:\u003c/u\u003e Our machine learning model was trained on manually segmented MR images. \u0026nbsp;One hundredT2-axial MRI scans were randomly selected from our institution\u0026rsquo;s picture archiving and communications system (PACS). Images were downloaded as DICOM files, anonymized by removing header information, converted to NIFTI format, resampled to 256 x 256 pixels in the plane of acquisition, and, finally, histogram matched to a common template to standardize the maximum and minimum pixel values to 1 and 0 respectively. Following this pre-processing, clinically trained human expert raters delineated cervical canals and spinal cords in the axial plane on these scans using ITK-SNAP, an open-source application for medical image segmentation (\u003cstrong\u003eFigure 1\u003c/strong\u003e)\u003csup\u003e5\u003c/sup\u003e. Segmentations were stored as NIFTI files alongside the pre-processed training data for deep learning training.\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eDeep Learning Model Training:\u003c/u\u003eOur ensemble contained three U-Nets trained on image patches extracted from pre-processed axial T2-MRIs. The first U-Net was trained using patches 160x160px, the second using patches 128x128px, and the third using patches 96x96px. Patches were extracted in the plane of acquisition at regular strides of length 80px, 64 px, and 48px, respectively. Extracted image patches, along with the corresponding extracted region from the segmentation images, were augmented three times using affine transformations and flipped up-down as well as left-right. The collection of augmented and original patches and segmentations was then used to train the network.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe three networks in our ensemble were identical in all hyperparameters except the input image size. The relevant hyperparameters were a kernel size of 3x3px with a stride of 1px in each dimension, ReLU activation, and Glorot initialization for all convolutional layers. The starting channel depth was 16 for all networks, and our networks had five downsampling and upsampling units. Each unit consisted of a max pooling block using a 2x2px kernel sandwiched between two convolutional blocks. Skip connections were added in accordance with the standard U-Net dogma. Channel depth was doubled with each downsampling and halved with each upsampling according to standard U-Net dogma as well. The final layer of all networks was a sigmoid to constrain the output between 0 and 1 for segmentation purposes. We also use dropout regularization with a probability of 0.25 and the Adam optimizer with a fixed learning rate of 1e-5 for training, alongside the Dice coefficient as a loss function. Segmentations generated using all three networks (96x96, 128 x 128, and 160 x 160) in the ensemble were combined using pixel-wise majority voting to generate the final segmentation. The results presented here are based on two ensembles: the first trained to segment spinal canals automatically and the second trained to segment spinal cords automatically in the axial plane on cervical spine MR images.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eBiomarker Generation:\u003c/u\u003e\u0026nbsp; \u0026nbsp;Canal and cord areas were measured in square millimeters (mm\u003csup\u003e2\u003c/sup\u003e) at all cervical levels except the most cranial 20% of each image due to the presence of brain regions that were irrelevant to our research question. The difference in area between the spinal canal and spinal cord was computed at every level, and the minimum value for this difference in area along the entire cervical spinal canal served as our biomarker. Segmentation failure for a particular axial slice was defined when irrelevant regions of the spinal canal (i.e. the medulla) were segmented or when the model failed to segment either the spinal canal or spinal cord region completely. In such cases, the specific axial slice where failure occurred was excluded.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cu\u003ePatient Data Collection:\u0026nbsp;\u003c/u\u003eTo test our model in the clinical setting, we then collected and stored MR imaging data and clinical metadata for 205 consecutive patients presenting with complaints related to cervical stenosis at our institution between 2019 and 2020. As part of their treatment course, all of these patients had been independently evaluated by one of four board-certified neurosurgeons during that same time period. \u0026nbsp;Patients were excluded if their imaging data could not be accessed or processed \u0026ndash; an issue that might occur if their imaging had been obtained at an outside institution. \u0026nbsp; Accessible MR images were downloaded in DICOM format and pre-processed as described above. Further processing and measurements were performed using these pre-processed images.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cu\u003eModel Evaluation:\u003c/u\u003e We then allowed our model to perform automated segmentations on these MRI scans using previously described trained deep learning ensemble segmentation algorithms. \u0026nbsp;The corresponding biomarker generated by these automated segmentations was tested against the outcome of a subsequent surgical decompression for cervical stenosis. \u0026nbsp;\u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003cstrong\u003eDisclosures:\u003c/strong\u003e \u003cp\u003eDr. Macyszyn discloses a financial interest in Theseus Artificial Intelligence, which has signed a license agreement with the University of California, Los Angeles for the technology utilized in this research. The remaining authors have no personal, financial, or institutional interest in any part of the research presented in this work.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eA.R. and B.W. co-authored the main manuscript text. D.Z. and T.F. were responsible for editing the manuscript. A.R., D.Z, K.C, A.L., T.F, B.Y., and B.S were responsible for data acquisition and data analysis..B.G, A.S, were responsible for designing the machine learning architecture. K.C. and D.B prepared the figures.B.G., L.M., and J.B. were responsible for the study conception, funding, supervision, and overall direction of the project. All authors reviewed the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eAcknowledgments: The authors would like to dedicate this work in honor of Dr. Bilwaj Gaonkar, Ph.D., who passed away during the course of this study. Dr. Gaonkar, a co-author of this manuscript, contributed substantially to the research described. The present study would not have been possible without him.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets used and analyzed during the current study are available from the corresponding author at reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBot SDM, van der Waal JM, Terwee CB, et al. Incidence and prevalence of complaints of the neck and upper extremity in general practice. \u003cem\u003eAnn Rheum Dis\u003c/em\u003e. 2005;64(1):118-123.\u003c/li\u003e\n\u003cli\u003ePicavet HSJ, Schouten JSAG. Musculoskeletal pain in the Netherlands: prevalences, consequences and risk groups, the DMC3-study. \u003cem\u003ePain\u003c/em\u003e. 2003;102(1):167-178.\u003c/li\u003e\n\u003cli\u003eIzzo R, Popolizio T, Balzano RF, Pennelli AM, Simeone A, Muto M. Imaging of cervical spine traumas. \u003cem\u003eEur J Radiol\u003c/em\u003e. 2019;117:75-88.\u003c/li\u003e\n\u003cli\u003eMalhotra A, Wu X, Kalra VB, et al. Utility of MRI for cervical spine clearance after blunt traumatic injury: a meta-analysis. \u003cem\u003eEur Radiol\u003c/em\u003e. 2017;27(3):1148-1160.\u003c/li\u003e\n\u003cli\u003eWilson B, Gaonkar B, Yoo B, et al. Predicting Spinal Surgery Candidacy From Imaging Data Using Machine Learning. \u003cem\u003eNeurosurgery\u003c/em\u003e. 2021;89(1):116-121.\u003c/li\u003e\n\u003cli\u003eDebono B, Sabatier P, Koudsie A, Buffenoir K, Hamel O. Managing spine surgery referrals: The consultation of neurosurgery and its nuances. \u003cem\u003eNeurochirurgie\u003c/em\u003e. 2017;63(4):267-272.\u003c/li\u003e\n\u003cli\u003eKidane B, Gandhi R, Sarro A, Valiante TA, Harvey BJ, Rampersaud YR. Is referral to a spine surgeon a double-edged sword?: patient concerns before consultation. \u003cem\u003eCan Fam Physician\u003c/em\u003e. 2011;57(7):803-810.\u003c/li\u003e\n\u003cli\u003eMayman D, Yen D. Maximizing use of a surgical clinic for referrals of patients having back problems. \u003cem\u003eCan J Surg\u003c/em\u003e. 1999;42(2):117-119.\u003c/li\u003e\n\u003cli\u003eFindlay JM, Deis N. Appropriateness of lumbar spine referrals to a neurosurgical service. \u003cem\u003eCan J Neurol Sci\u003c/em\u003e. 2010;37(6):843-848.\u003c/li\u003e\n\u003cli\u003eBraybrooke J, Ahn H, Gallant A, et al. The impact of surgical wait time on patient-based outcomes in posterior lumbar spinal surgery. \u003cem\u003eEur Spine J\u003c/em\u003e. 2007;16(11):1832-1839.\u003c/li\u003e\n\u003cli\u003eChildress MA, Stuek SJ. Neck pain: Initial evaluation and management. \u003cem\u003eAm Fam Physician\u003c/em\u003e. 2020;102(3):150-156.\u003c/li\u003e\n\u003cli\u003eBartholomeeusen S, Van Zundert J, Truyers C, Buntinx F, Paulus D. Higher incidence of common diagnoses in patients with low back pain in primary care. \u003cem\u003ePain Pract\u003c/em\u003e. 2012;12(1):1-6.\u003c/li\u003e\n\u003cli\u003eJensen MC, Brant-Zawadzki MN, Obuchowski N, Modic MT, Malkasian D, Ross JS. Magnetic resonance imaging of the lumbar spine in people without back pain. \u003cem\u003eN Engl J Med\u003c/em\u003e. 1994;331(2):69-73.\u003c/li\u003e\n\u003cli\u003eKasch R, Truthmann J, Hancock MJ, et al. Association of lumbar MRI findings with current and future back pain in a population-based cohort study. \u003cem\u003eSpine (Phila Pa 1976)\u003c/em\u003e. 2022;47(3):201-211.\u003c/li\u003e\n\u003cli\u003eWilgenbusch CS, Wu AS, Fourney DR. Triage of spine surgery referrals through a multidisciplinary care pathway: a value-based comparison with conventional referral processes. \u003cem\u003eSpine \u003c/em\u003e. 2014;39(22 Suppl 1):S129-35.\u003c/li\u003e\n\u003cli\u003eLayne EI, Roffey DM, Coyle MJ, Phan P, Kingwell SP, Wai EK. Activities performed and treatments conducted before consultation with a spine surgeon: are patients and clinicians following evidence-based clinical practice guidelines? \u003cem\u003eSpine J\u003c/em\u003e. 2018;18(4):614-619.\u003c/li\u003e\n\u003cli\u003eFern\u0026aacute;ndez de Rota JJ, Meschian S, Fern\u0026aacute;ndez de Rota A, Urbano V, Baron M. Cervical spondylotic myelopathy due to chronic compression: the role of signal intensity changes in magnetic resonance images. \u003cem\u003eJ Neurosurg Spine\u003c/em\u003e. 2007;6(1):17-22.\u003c/li\u003e\n\u003cli\u003eWorld medical association declaration of Helsinki. \u003cem\u003eJAMA\u003c/em\u003e. 2013;310(20):2191.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"cervical spine, deep learning, machine learning, surgical candidacy, triage. ","lastPublishedDoi":"10.21203/rs.3.rs-4385667/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4385667/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMany patients who present to their primary care physician for neck pain undergo magnetic resonance imaging (MRI) as part of their diagnostic workup. The physician is then tasked with deciding if the findings of the MRI and workup warrant referral a spine surgery, an intricate task complicated by the high rates of background findings. This results in a high number of non-surgical patients being referred to surgery. Although there are a multitude of reasons for non-surgical patients to still see a subspecialist, deep learning has the potential to help inform physicians of their patients’ surgical candidacy. The preset work describes a proof-of-concept model for evaluating operative candidacy for cervical stenosis only using data from outpatient elective magnetic resonance imaging (MRI) scans. This deep-learning algorithm was trained to automatically segment the areas of both the spinal canal and spinal cord on 100 axial cervical spine MRIs. Once segmented, the model used these areas to generate a biomarker for cervical stenosis, calculated as the minimum difference in cross-sectional area between the spinal canal and the spinal cord within the cervical spine. Following training, the model and its biomarker were tested against a cohort of 147 consecutive patients evaluated in the outpatient setting by a group of board-certified neurosurgeons at our institution for complaints related to their cervical spines.\u0026nbsp; Our automated model determined that the mean minimum difference in cross-sectional area between the spinal canal and spinal cord for our cohort was 35.90±25.00 mm\u003csup\u003e2\u003c/sup\u003e for patients who ultimately underwent surgical decompression and 48.55±33.52 mm\u003csup\u003e2\u0026nbsp; \u003c/sup\u003efor patients who did not (P=0.005). Using this biomarker, the model distinguished between surgical and non-surgical patients with relatively high accuracy (AUC 0.79). When tested against a novel cohort of outpatient spine surgery clinic patients, the described algorithm determined whether the patient underwent decompression for cervical stenosis using data acquired solely from their cervical spine MRI scans.\u0026nbsp; These findings support a proof-of-concept for our automated deep-learning model and biomarker, which could significantly improve the efficiency of the referral process for patients with neck complaints to a surgical subspecialist.\u003c/p\u003e","manuscriptTitle":"Deep Learning Based Evaluation of Surgical Candidacy for Cervical Spinal Cord Decompression","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-05-30 14:47:00","doi":"10.21203/rs.3.rs-4385667/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"cb61c873-1e8c-4521-8b79-d54f396792e3","owner":[],"postedDate":"May 30th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":32173948,"name":"Health sciences/Biomarkers/Diagnostic markers"},{"id":32173949,"name":"Health sciences/Biomarkers/Predictive markers"},{"id":32173950,"name":"Health sciences/Health care/Diagnosis"},{"id":32173951,"name":"Health sciences/Health care/Medical imaging"}],"tags":[],"updatedAt":"2025-02-28T10:08:48+00:00","versionOfRecord":[],"versionCreatedAt":"2024-05-30 14:47:00","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4385667","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4385667","identity":"rs-4385667","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}