Advertisement

Machine learning in clinical diagnosis, prognostication, and management of acute traumatic spinal cord injury (SCI): A systematic review

Published:October 20, 2022DOI:https://doi.org/10.1016/j.jcot.2022.102046

      Abstract

      Background

      Machine learning has been applied to improve diagnosis and prognostication of acute traumatic spinal cord injury. We investigate potential for clinical integration of machine learning in this patient population to navigate variability in injury and recovery.

      Materials and methods

      We performed a systematic review using PRISMA guidelines through PubMed database to identify studies that use machine learning algorithms for clinical application toward improvements in diagnosis, management, and predictive modeling.

      Results

      Of the 132 records identified, a total of 13 articles met inclusion criteria and were included in final analysis. Of the 13 articles, 5 focused on diagnostic accuracy and 8 were related to prognostication or management of traumatic spinal cord injury. Across studies, 1983 patients with spinal cord injury were evaluated with most classifying as ASIA C or D. Retrospective designs were used in 10 of 13 studies and 3 were prospective. Studies focused on MRI evaluation and segmentation for diagnostic accuracy and prognostication, investigation of mean arterial pressure in acute care and intraoperative settings, prediction of ambulatory and functional ability, chronic complication prevention, and psychological quality of life assessments. Decision tree, random forests (RF), support vector machines (SVM), hierarchical cluster tree analysis (HCTA), artificial neural networks ( ANN ), convolutional neural networks (CNN) machine learning subtypes were used.

      Conclusions

      Machine learning represents a platform technology with clinical application in traumatic spinal cord injury diagnosis, prognostication, management, rehabilitation, and risk prevention of chronic complications and mental illness. SVM models showed improved accuracy when compared to other ML subtypes surveyed. Inherent variability across patients with SCI offers unique opportunity for ML and personalized medicine to drive desired outcomes and assess risks in this patient population.

      Keywords

      1. Introduction

      Artificial intelligence (AI), the domain of computer science that perform tasks through human intelligence simulation,
      • Raju B.
      • Jumah F.
      • Ashraf O.
      • et al.
      Big data, machine learning, and artificial intelligence: a field guide for neurosurgeons.
      is a rapidly growing field that has broad application across industry, science, and technology.
      • Haenlein M.
      • Kaplan A.
      A brief history of artificial intelligence: on the past, present, and future of artificial intelligence.
      Within AI, Machine Learning (ML) involves creation of algorithms in which machines dynamically learn from data and aim to improve their predictive and computational accuracy over time.
      • Raju B.
      • Jumah F.
      • Ashraf O.
      • et al.
      Big data, machine learning, and artificial intelligence: a field guide for neurosurgeons.
      The nexus of ML technologies with advances in medicine has contributed to improvement in diagnosis, risk assessment, and predicting response to treatment.
      • Kaul V.
      • Enslin S.
      • Gross S.A.
      History of artificial intelligence in medicine.
      • English M.
      • Kumar C.
      • Ditterline B.L.
      • Drazin D.
      • Dietz N.
      Machine learning in neuro-oncology, epilepsy, alzheimer's disease, and schizophrenia.
      • Dietz N.
      • Sharma M.
      • Alhourani A.
      • et al.
      Evaluation of predictive models for complications following spinal surgery.
      • Dietz N.
      • Sharma M.
      • Alhourani A.
      • et al.
      Variability in the utility of predictive models in predicting patient-reported outcomes following spine surgery for degenerative conditions: a systematic review.
      • Dietz N.
      • Ruff C.
      • Giugliano R.P.
      • Mercuri M.F.
      • Antman E.M.
      Pharmacogenetic-guided and clinical warfarin dosing algorithm assessments with bleeding outcomes risk-stratified by genetic and covariate subgroups.
      As advances in basic neuroscience meet technological innovation, neurosurgery has seen unique opportunities for ML in research and clinical application to optimize patient care.
      • Raju B.
      • Jumah F.
      • Ashraf O.
      • et al.
      Big data, machine learning, and artificial intelligence: a field guide for neurosurgeons.
      ,
      • English M.
      • Kumar C.
      • Ditterline B.L.
      • Drazin D.
      • Dietz N.
      Machine learning in neuro-oncology, epilepsy, alzheimer's disease, and schizophrenia.
      ,
      • Dagi T.F.
      • Barker F.G.
      • Glass J.
      Machine learning and artificial intelligence in neurosurgery: status, prospects, and challenges.
      • Staartjes V.E.
      • Sebok M.
      • Blum P.G.
      • et al.
      Development of machine learning-based preoperative predictive analytics for unruptured intracranial aneurysm surgery: a pilot study.
      • Staartjes V.E.
      • Stumpo V.
      • Kernbach J.M.
      • et al.
      Machine learning in neurosurgery: a global survey.
      However, challenges exist for complex and non-linear data to integrate clinical importance for prognostic modeling.
      • Fallah N.
      • Noonan V.K.
      • Waheed Z.
      • et al.
      Development of a machine learning algorithm for predicting in-hospital and 1-year mortality after traumatic spinal cord injury.
      • Royston P.
      • Moons K.G.
      • Altman D.G.
      • Vergouwe Y.
      Prognosis and prognostic research: developing a prognostic model.
      • Khan O.
      • Badhiwala J.H.
      • Grasso G.
      • Fehlings M.G.
      Use of machine learning and artificial intelligence to drive personalized medicine approaches for spine care.
      Acute traumatic spinal cord injury (SCI) results in temporary or permanent impairment of motor function and sensation,
      • Silva N.A.
      • Sousa N.
      • Reis R.L.
      • Salgado A.J.
      From basics to clinical: a comprehensive review on spinal cord injury.
      and has devastating short and long term sequela with significant morbidity and mortality.
      • Singh A.
      • Tetreault L.
      • Kalsi-Ryan S.
      • Nouri A.
      • Fehlings M.G.
      Global prevalence and incidence of traumatic spinal cord injury.
      • Dietz N.
      • Sarpong K.
      • Ugiliweneza B.
      • et al.
      Longitudinal trends and prevalence of bowel management in individuals with spinal cord injury.
      • Ugiliweneza B.
      • Guest J.
      • Herrity A.
      • et al.
      A two-decade assessment of changing practice for surgical decompression and fixation after traumatic spinal cord injury - impact on healthcare utilization and cost.
      • Sharma M.
      • Dietz N.
      • Ugiliweneza B.
      • et al.
      Impact of surgical timing and approaches to health care utilization in patients undergoing surgery for acute traumatic cervical spinal cord injury.
      • Middleton J.W.
      • Lim K.
      • Taylor L.
      • Soden R.
      • Rutkowski S.
      Patterns of morbidity and rehospitalisation following spinal cord injury.
      SCI has significant implications on longevity, functional ability, psychological well-being, and socioeconomic stability.
      • Ugiliweneza B.
      • Guest J.
      • Herrity A.
      • et al.
      A two-decade assessment of changing practice for surgical decompression and fixation after traumatic spinal cord injury - impact on healthcare utilization and cost.
      ,
      • Sharma M.
      • Dietz N.
      • Ugiliweneza B.
      • et al.
      Impact of surgical timing and approaches to health care utilization in patients undergoing surgery for acute traumatic cervical spinal cord injury.
      ,
      • Migliorini C.
      • Tonge B.
      • Taleporos G.
      Spinal cord injury and mental health.
      Prevalence of chronic spinal cord injury is estimated to be almost 300,000 in the United States.
      • Center N.S.C.I.S.
      Facts and figure at a glance.
      Given the importance of efficient diagnosis and treatment in management SCI, ML technologies have the potential to advance best practices and standard of care.
      • Raju B.
      • Jumah F.
      • Ashraf O.
      • et al.
      Big data, machine learning, and artificial intelligence: a field guide for neurosurgeons.
      ,
      • Mesbah S.
      • Ball T.
      • Angeli C.
      • et al.
      Predictors of volitional motor recovery with epidural stimulation in individuals with chronic spinal cord injury.
      Further, personalized or precision medicine for patients with SCI is useful to tailor expectation and management given the inherent variability among this population in outcome, functional prognosis, and rehabilitation journey.
      • Khan O.
      • Badhiwala J.H.
      • Grasso G.
      • Fehlings M.G.
      Use of machine learning and artificial intelligence to drive personalized medicine approaches for spine care.
      ,
      • Dietz N.
      • Wagers S.
      • Harkema S.J.
      • D'Amico J.M.
      Intrathecal and oral baclofen use in adults with spinal cord injury (SCI): a systematic review of efficacy in spasticity reduction and functional improvement, dosing and adverse events.
      ,
      • Ter Wengel P.V.
      • Post M.W.M.
      • Martin E.
      • et al.
      Neurological recovery after traumatic spinal cord injury: what is meaningful? A patients' and physicians' perspective.
      This review provides an overview of current machine learning methods used in diagnosis and management of acute traumatic spinal cord injury aimed at clinical applicability and treatment outcome. Emphasis was placed on studies that aimed to inform clinical improvement and prognostic models.

      2. Materials and methods

      2.1 Data extraction

      We organized the systematic review around the PICOS model (Participants, Intervention, Comparison, Outcomes, Study Design) as previously described,
      • Dietz N.
      • Sharma M.
      • Adams S.
      • et al.
      Enhanced recovery after surgery (eras) for spine surgery: a systematic review.
      ,
      • Dietz N.
      • Sharma M.
      • Alhourani A.
      • et al.
      Bundled payment models in spine surgery: current challenges and opportunities, a systematic review.
      to define the population with spinal cord injury and associated machine learning analysis.

      2.1.1 PICOS outline

      Participants: Adult patients ≥18 years with spinal cord injury.
      Intervention: Machine learning algorithms tailored to help SCI diagnosis, management, and were of primary focus.
      Comparison: Comparison of accuracies of machine learning models and clinical were assessed as it pertains to improving patient overall quality of life.
      Outcomes: Efficacy of various machine learning techniques by way of faster, more accurate diagnosis or improvements in clinical treatment and prediction of complications or functional ability were assessed.
      Study Design: Inclusive of prospective, retrospective, case reports, case series, observational studies, comparative studies and randomized clinical trials. No reviews or meta-analyses were included.

      2.2 Search criteria

      We followed the PRISMA guidelines (2009) to construct the framework of the systematic review and conducted the search on June 3rd, 2022 using PubMed databases. Additional articles used in the references were incorporated from the references of those articles identified in the searches. We used Keyword and MeSH terms for predictive outcomes to include the following terms with numbered iterations for the database as follows: Spinal Cord Injury and Machine Learning.
      Eligibility criteria: Current publication in English language.
      • 1)
        Pubmed: Spinal Cord Injury and Machine Learning: 132 articles; 13 included
      Risk of bias evaluation: Assessment of conflict of interest, funding for study and study design were assessed according to QUADAS criteria.

      2.2.1 Exclusion criteria

      Systematic reviews and meta-analyses were excluded. Studies with nonhuman subjects, pediatric population, language other than English, case studies, brain-computer interface focus, and those without full text were also excluded.

      2.2.2 Data evaluation

      We followed the QUADAS tool to evaluate risk bias and result applicability of the studies according to 2003 guidelines. A total of 13 questions on the QUADAS survey that cover patient selection, index test, reference standard, and timing were addressed for each study and incorporated into the final analysis.

      3. Results

      Of the 132 records identified, a total of 13 articles
      • Agarwal N.
      • Aabedi A.A.
      • Torres-Espin A.
      • et al.
      Decision tree-based machine learning analysis of intraoperative vasopressor use to optimize neurological improvement in acute spinal cord injury.
      • Arslan Y.Z.
      • Demirer R.M.
      • Palamar D.
      • Ugur M.
      • Karamehmetoglu S.S.
      Comparison of the data classification approaches to diagnose spinal cord injury.
      • Chou A.
      • Torres-Espin A.
      • Kyritsis N.
      • et al.
      Expert-augmented automated machine learning optimizes hemodynamic predictors of spinal cord injury outcome.
      • DeVries Z.
      • Hoda M.
      • Rivers C.S.
      • et al.
      Development of an unsupervised machine learning algorithm for the prognostication of walking ability in spinal cord injury patients.
      • Facchinello Y.
      • Beausejour M.
      • Richard-Denis A.
      • Thompson C.
      • Mac-Thiong J.M.
      Use of regression tree analysis for predicting the functional outcome after traumatic spinal cord injury.
      • Fu J.
      • Jones M.
      • Jan Y.K.
      Development of intelligent model for personalized guidance on wheelchair tilt and recline usage for people with spinal cord injury: methodology and preliminary report.
      • Gibert K.
      • Garcia-Rudolph A.
      • Curcoll L.
      • Soler D.
      • Pla L.
      • Tormos J.M.
      Knowledge discovery about quality of life changes of spinal cord injury patients: clustering based on rules by states.
      • Inoue T.
      • Manley G.T.
      • Patel N.
      • Whetstone W.D.
      Medical and surgical management after spinal cord injury: vasopressor usage, early surgerys, and complications.
      • McCoy D.B.
      • Dupont S.M.
      • Gros C.
      • et al.
      Convolutional neural network-based automated segmentation of the spinal cord and contusion injury: deep learning biomarker correlates of motor impairment in acute spinal cord injury.
      • Okimatsu S.
      • Maki S.
      • Furuya T.
      • et al.
      Determining the short-term neurological prognosis for acute cervical spinal cord injury using machine learning.
      • Tay B.
      • Hyun J.K.
      • Oh S.
      A machine learning approach for specification of spinal cord injuries using fractional anisotropy values obtained from diffusion tensor images.
      • Torres-Espin A.
      • Haefeli J.
      • Ehsanian R.
      • et al.
      Topological network analysis of patient similarity for precision management of acute blood pressure in spinal cord injury.
      • Zariffa J.
      • Curt A.
      • Verrier M.C.
      • et al.
      Predicting task performance from upper extremity impairment measures after cervical spinal cord injury.
      were included in final analysis and met inclusion criteria, Fig. 1, Table 1, Table 2. Of the 13 articles, 5 focused on diagnostic accuracy and 8 were related to prognostication or management of traumatic spinal cord injury. Across studies, 1983 patients with spinal cord injury were evaluated with most classifying as ASIA C or D. Retrospective designs were used in 10 of the 13 studies and 3 were prospective. Studies focused on MRI evaluation and segmentation for diagnostic accuracy and prognostication, investigation of mean arterial pressure in acute care and intraoperative settings, prediction of ambulatory and functional ability, long term complication prevention, and psychological quality of life assessments. Decision tree, random forests (RF), support vector machines (SVM), hierarchical cluster tree analysis (HCTA), artificial neural networks ( ANN ), convolutional neural networks (CNN) machine learning subtypes were used. SVM was shown to have superior predictive ability.
      • Arslan Y.Z.
      • Demirer R.M.
      • Palamar D.
      • Ugur M.
      • Karamehmetoglu S.S.
      Comparison of the data classification approaches to diagnose spinal cord injury.
      ,
      • Fu J.
      • Jones M.
      • Jan Y.K.
      Development of intelligent model for personalized guidance on wheelchair tilt and recline usage for people with spinal cord injury: methodology and preliminary report.
      Fig. 1
      Fig. 1PRISMA flowchart of systemic search strategy.
      Table 1Table 1. List of studies selected for final analysis.
      AuthorYearNumber of Patients with SCIASIA Impairment Scale (AIS)TopicCohort StudyType of MLSummary
      Agarwal et al.202274A (n = 37), B (n = 8), C (n = 14), D (n = 13), E (n = 2)DRetrospectiveDeep-tree, random forestsDeep-tree based machine learning was used to assess intraoperative mean arterial pressure (MAP) and vasopressor use to optimize neurological improvement in acute spinal cord injury. MAP between 80 and 96 mmHg was associated with improved neurological function while 93 min or longer outside of MAP range of 76–104 mmHg was associated with worse outcome.
      Arslan et al.201215NADRetrospectiveSupport Vector Machines (SVM) and hierarchical cluster tree analysis (HCTA)Comparing three classification methods, SVM and HCTA attained higher recognition accuracies than the previous classification method of ANN for SCI diagnosis.
      Chou et al.202259NAMProspectiveAutomated machine learningAutomated machine learning (AutoML) was applied to spinal cord injury prognostication to enhance intraoperative hemodynamic management during decompression surgery and determine relationship of intraoperative hypertension and patient outcome.
      DeVries et al.2020862A = 257, B = 447, C = 149, D = 379MRetrospectiveUnsupervised machine learning algorithm and linear regressionComparison of unsupervised MLA and LR with total admission neurological information against prior models showed no clinically relevant differences in functional prediction. The F1-score was shown to be more reliable in algorithm assessment than area under the operating curve.
      Facchinello et al.2021172A = 68, B = 17 C = 25 D = 62MProspectiveRegression TreeFunctional outcome predictive models used 4 and 11 variables that explained up to 62.3% of variance of spinal cord independence measure. Severity of the neurological deficit at admission was most predictive variable followed by injury severity score, age, neurological level, and surgical delay.
      Fu et al.2014NANAMRetrospectiveArtificial neural network (ANN), Random Forest (RF), and Support Vector Machines (SVM)Individual wheelchair setting prediction was implemented to optimize tilt and body turning depending of the needs of the patient to mitigate frequency and severity of pressure ulcers.
      Gibert et al.2009109NAMRetrospectiveKnowledge discovery methodology, clustering based on rulesClustering based on rules is used to extrapolate patterns about the quality of life predictions for patients with spinal cord injury.
      Inoue et al.2020165A = 15, B = 38, C = 66, D = 44, E = 3DRetrospectiveExtreme gradient boosting (XGBoost), Logistic regression and decision treeReliability of machine learning tools to predict neurological outcomes after SCI was assessed. XGBoost had the highest accuracy (81.1%), followed by the logistic regression (80.6%) and the decision tree (78.8%) for neurological improvement.
      McCoy et al.201947A (n = 10), B (n = 3), C (n = 5), D (n = 21), Unknown (n = 8)MProspectiveConvolutional Neural Network2D convolutional neural networks (CNN) were used for automatic segmentation of the spinal cord and traumatic contusion injury in patients with acute spinal cord injury. The segmentation method developed in this study compares favorably with available state-of-the-art segmentation methods in a cohort of patients with acute spinal cord injury.
      Okimatsu et al.2022215A (n = 25), B (n = 19), C (n = 62), D (n = 82), E (n = 27)DRetrospectiveDeep learning-based radiomics and convolutional neural networkDeveloped deep-learning based radiomics to quantify radiographic characteristics using a convolutional neural network (CNN) to stratify prognosis of patient neurological outcomes after acute spinal cord injury (SCI).
      Tay et al.20149NADRetrospectiveSupport vector machine (SVM) and K-nearest neighbhor (KNN)Classification system that differentiates healthy individuals and patients with spinal cord injuries (SCI) via Diffusion Tensor Imaging (DTI), characterizes injury.
      Torres-Espín et al.2021118A (n = 52), B (n = 13), C (n = 16), D (n = 19), E (n = 4), Missing (n = 14)MRetrospectiveTopological network extraction and regression analysisA similarity network of patients was created to predict neurological recovery after spinal cord injury on the basis of mean arterial pressure (MAP) during surgery. The network analysis showed that time outside of an optimum MAP range (hypotension or hypertension) during surgery was associated with lower likelihood of neurological recovery.
      Zariffa et al.2016138NAMRetrospectiveClassifier, Leave-one-out cross-validationThis study defined the predictive value of impairment measures as measured by the Graded Redefined Assessment of Strength, Sensibility and Prehension (GRASSP) in traumatic cervical spinal cord injury (SCI). The study found that upper extremity motor impairment measures were highly predictive of task performance.
      Table 2QUADAS bias table.
      QUADAS QuestionYesNoUnclear
      Was the spectrum of patients representative of the patients who will receive the test in practice?1300
      Were selection criteria clearly described?1120
      Is the reference standard likely to correctly classify the target condition?1300
      Is the time period between reference standard and index test short enough to be reasonably sure that the target condition did not change between the two tests?1012
      Did the whole sample or a random selection of the sample, receive verification using a reference standard of diagnosis?0013
      Did patients receive the same reference standard regardless of the index test result?0013
      Was the reference standard independent of the index test (i.e. the index test did not form part of the reference standard)?0013
      Was the execution of the index test described in sufficient detail to permit replication of the test?1300
      Was the execution of the reference standard described in sufficient detail to permit its replication?1300
      Were the index test results interpreted without knowledge of the results of the reference standard?0013
      Were the reference standard results interpreted without knowledge of the results of the index test?0013
      Were the same clinical data available when test results were interpreted as would be available when the test is used in practice?0013
      Were uninterpretable/intermediate test results reported?922
      Were withdrawals from the study explained?0013
      QUADAS criteria represented above with tally of ratings for all 13 studies included in the final analysis: (Y): Yes; (N): No; (U): Unclear. The "reference standard” was determined to be the control cohort for those studies that incorporated randomized control trials.

      3.1 Diagnosis of SCI

      Of the 5 diagnostic-related studies on ML in SCI, all were retrospectively designed with a total of 478 patients. Classifications were 77 with ASIA A, 65 with ASIA B, 142 ASIA C, and 139 ASIA D. MRI and other diagnostic modalities were investigated to improve assessment accuracy and segmentation of cord injury.

      3.2 Management of SCI

      Most of the studies (n = 8) were focused on prognostication and management of SCI, with 3 prospective and 5 retrospectively designed. A total of 1505 patients were included with classification breakdown of 387 ASIA A, 480 ASIA B, 195 ASIA C, 481 ASIA D. Topics included blood pressure management, prognosis with predictive modeling, and psychological outcome prediction.

      4. Discussion

      The present review demonstrates ML contributions toward 1) diagnostic accuracy with advances in MRI segmentation and classification 2) acute blood pressure management for optimizing functional AISIA classification 3) quality of life clustering with predictive modeling of patient needs 4) prevention of long term complications and morbidity 5) functional prognostication with predictive modeling. A recognized purpose of ML is to identify key aspects and elements of injury or patient characteristics that drive outcomes. Given the variability in injury type and rehabilitative recovery, ML may meet the need for more personalized treatment approaches for patients with SCI.
      • Khan O.
      • Badhiwala J.H.
      • Grasso G.
      • Fehlings M.G.
      Use of machine learning and artificial intelligence to drive personalized medicine approaches for spine care.
      ,
      • Hetz S.P.
      • Latimer A.E.
      • Ginis K.A.
      Activities of daily living performed by individuals with SCI: relationships with physical fitness and leisure time physical activity.
      Machine learning may have better efficacy in uniting datasets to create accurate algorithmic models for clinical prediction in iterative generations of modeling.
      • Khan O.
      • Badhiwala J.H.
      • Grasso G.
      • Fehlings M.G.
      Use of machine learning and artificial intelligence to drive personalized medicine approaches for spine care.

      4.1 Diagnosis

      Early and accurate diagnosis and classification of spinal cord injury (complete versus incomplete) has implications on appropriate medical and surgical management for acute spinal cord injury. To date, surgical decompression within 24 h is the most effective intervention for primary injury when indicated to limit progressive damage.
      • Ugiliweneza B.
      • Guest J.
      • Herrity A.
      • et al.
      A two-decade assessment of changing practice for surgical decompression and fixation after traumatic spinal cord injury - impact on healthcare utilization and cost.
      ,
      • Fehlings M.G.
      • Vaccaro A.
      • Wilson J.R.
      • et al.
      Early versus delayed decompression for traumatic cervical spinal cord injury: results of the Surgical Timing in Acute Spinal Cord Injury Study (STASCIS).
      ,
      • Alizadeh A.
      • Dyck S.M.
      • Karimi-Abdolrezaee S.
      Traumatic spinal cord injury: an overview of pathophysiology, models and acute injury mechanisms.
      As the gold standard for spinal cord imaging, MRI aids in characterization of SCI through presence of hematoma and edema, visualization of ligamentous and disc disruption, surgical considerations, and implications for classification and prognosis.
      • Goldberg A.L.
      • Kershah S.M.
      Advances in imaging of vertebral and spinal cord injury.
      ,
      • Bozzo A.
      • Marcoux J.
      • Radhakrishna M.
      • Pelletier J.
      • Goulet B.
      The role of magnetic resonance imaging in the management of acute spinal cord injury.
      In the setting of polytrauma, patients may not be able to participate in a reliable neurological exam, potentially masking the nature of SCI. In such circumstances of suspected SCI without reliable exam, there is added emphasis on imaging techniques to characterize the nature of a spinal cord injury. Arslan and colleagues have demonstrated enhanced diagnosis of spinal cord injury ( SCI ) on unconscious and uncooperative patients using dermatomal skin impendence analysis through artificial neural network ( ANN ) to aid in diagnosis of SCI .
      • Arslan Y.Z.
      • Demirer R.M.
      • Palamar D.
      • Ugur M.
      • Karamehmetoglu S.S.
      Comparison of the data classification approaches to diagnose spinal cord injury.
      In this study, classification methods of Support Vector Machines (SVM) and hierarchical cluster tree analysis (HCTA) showed improved diagnostic rates. Comparing three classification methods, the study showed that SVM and HCTA attained higher recognition accuracies between SCI and control group using skin impedance than ANN, aiding in SCI diagnosis with reduced need for reliance on patient clinical exam. Interestingly, Tay et al. used ML to characterize diffusion tensor imaging (DTI) fractional anisotropy and fiber tracking in the evaluation of white matter tract disruption in the setting of SCI for diagnosis.
      • Tay B.
      • Hyun J.K.
      • Oh S.
      A machine learning approach for specification of spinal cord injuries using fractional anisotropy values obtained from diffusion tensor images.
      Pattern recognition algorithms were used to locate the region of cord injury with T2 and T1 sequences in different planes, and assess DTI diffusion anisotropy and fractional anisotropy to evaluate lesions and structural damage in the cord. Medical imaging will continue garner significant application of machine learning for diagnosis and prognosis,
      • Kim M.
      • Yun J.
      • Cho Y.
      • et al.
      Deep learning in medical imaging.
      and may offer particular benefit to management of spinal cord injury given its acuity, complexity variability, and multimodal treatment paradigms.

      4.2 Blood pressure management

      A cornerstone of the acute clinical care of patients with traumatic spinal cord injury is maximizing spinal cord perfusion with mean arterial pressure (MAP) goals.
      • Hawryluk G.
      • Whetstone W.
      • Saigal R.
      • et al.
      Mean arterial blood pressure correlates with neurological recovery after human spinal cord injury: analysis of high frequency physiologic data.
      Guidelines suggest maintaining a MAP of 85–90 for the first 7 days after spinal cord injury is related to optimal recovery of function, considering post-injury ASIA grading.
      • Hawryluk G.
      • Whetstone W.
      • Saigal R.
      • et al.
      Mean arterial blood pressure correlates with neurological recovery after human spinal cord injury: analysis of high frequency physiologic data.
      ,
      • Walters B.C.
      • Hadley M.N.
      • Hurlbert R.J.
      • et al.
      Guidelines for the management of acute cervical spine and spinal cord injuries: 2013 update.
      However, little high-quality clinical data confirms the duration and blood pressure parameters following acute spinal cord injury.
      • Saadeh Y.S.
      • Smith B.W.
      • Joseph J.R.
      • et al.
      The impact of blood pressure management after spinal cord injury: a systematic review of the literature.
      Several studies have explored blood pressure augmentation with one suggesting that MAP may be most beneficial in the first 2–3 days following injury in the intensive care unit.
      • Hawryluk G.
      • Whetstone W.
      • Saigal R.
      • et al.
      Mean arterial blood pressure correlates with neurological recovery after human spinal cord injury: analysis of high frequency physiologic data.
      It has also been shown that certain vasopressor agents such as levophed or phenylephrine are preferred when compared to dopamine, given its higher risk of cardiac comorbidity.
      • Inoue T.
      • Manley G.T.
      • Patel N.
      • Whetstone W.D.
      Medical and surgical management after spinal cord injury: vasopressor usage, early surgerys, and complications.
      ,
      • Saadeh Y.S.
      • Smith B.W.
      • Joseph J.R.
      • et al.
      The impact of blood pressure management after spinal cord injury: a systematic review of the literature.
      ,
      • Yue J.K.
      • Tsolinas R.E.
      • Burke J.F.
      • et al.
      Vasopressor support in managing acute spinal cord injury: current knowledge.
      In 2021, Torres-Espín et al. used a topological network machine learning of patients to predict neurological recovery after spinal cord injury (SCI) on the basis of mean arterial pressure (MAP) during surgery.
      • Torres-Espin A.
      • Haefeli J.
      • Ehsanian R.
      • et al.
      Topological network analysis of patient similarity for precision management of acute blood pressure in spinal cord injury.
      This similarity network was built using techniques including topological network extraction, dimensionality reduction, regression analysis, and prediction modeling and programming languages such as MATLAB and R. The network analysis showed that time outside of an optimal MAP range (hypotension or hypertension) during surgery was associated with lower likelihood of neurological recovery. The optimum MAP range associated with neurological recovery was found to be 76 [104–117] mmHg, such that MAPs outside this range result in lower likelihood of neurological recovery. These findings suggest updated MAP targets for therapeutic intervention given the gap in knowledge from validation of clinical MAP guidelines.
      • Saadeh Y.S.
      • Smith B.W.
      • Joseph J.R.
      • et al.
      The impact of blood pressure management after spinal cord injury: a systematic review of the literature.
      Another study associated with the Transforming Research and Clinical Knowledge for Spinal Cord Injury (TRACK-SCI) investigators published in 2022 by Chou et al. used a model-agnostic framework to further strengthen the automated machine learning (AutoML) applied to intraoperative hemodynamic monitoring and elucidated prognostic risks of intraoperative hypertension.
      • Chou A.
      • Torres-Espin A.
      • Kyritsis N.
      • et al.
      Expert-augmented automated machine learning optimizes hemodynamic predictors of spinal cord injury outcome.
      Using retrospective intraoperative records and clinical data recordings, the group demonstrated a way to build an AutoML model that quantified feature importance and interpretation, recognized feature instability, and was designed for model reproducibility and translational application. TRACK-SCI retrospectively builds on previous work and via incorporating machine learning (ML) to procure clinically actionable thresholds for intraoperative MAP management intraoperatively with Agarwal and colleagues in 2022.
      • Agarwal N.
      • Aabedi A.A.
      • Torres-Espin A.
      • et al.
      Decision tree-based machine learning analysis of intraoperative vasopressor use to optimize neurological improvement in acute spinal cord injury.
      Seventy-four patients were retrospectively analyzed following SCI. All patients had hemodynamic monitoring with recordings at 5-min intervals, which resulted in a total of 28.954 min and 5718 unique data points for each parameter, such as the type of vasopressor used, dosage, any drug related complications, the average intraoperative MAP and extreme MAP range (<76 mmHg or >104 mmHg). Outcomes were determined by ASIA grades. Factors related to improved AIS grade were determined through generation of random forests with 10,000 iterations. An average MAP value between 80 and 96 mmHg was associated with improved neurological function while an accumulated time of 93 min or longer outside of the MAP range of 76–104 mmHg was associated with worse neurological function.

      4.3 Quality of life and psychological state

      Nearly half of all patients with SCI experience mental health issues, with depression (37%) and anxiety representing the most common diagnoses.
      • Migliorini C.
      • Tonge B.
      • Taleporos G.
      Spinal cord injury and mental health.
      Complex factors underlying psycho-social state and emotional response for those with SCI affect quality of life and may present predictable trends that lend toward improved assistance from multidisciplinary teams. Gibert and colleagues use artificial intelligence (AI) to guide interpretation-oriented tools on quality of life for patients with SCI and create “Clustering based on rules” (CIBR)—a type of integral knowledge database that allows tracking and prediction of QoL.
      • Gibert K.
      • Garcia-Rudolph A.
      • Curcoll L.
      • Soler D.
      • Pla L.
      • Tormos J.M.
      Knowledge discovery about quality of life changes of spinal cord injury patients: clustering based on rules by states.
      Four types of patterns have been identified that have been used to identify psychological distress or marked decreases in QoL. Using expert opinions and interviews of 109 participants with SCI, the authors cluster mental health profiles through “knowledge discovery from data” (KDD) that informs mental and psych-social health of patients with SCI observe over time. Older age and longer time since injury were found to be associated with decreased QoL and determined as higher risk of distress and need for social support.

      4.4 Risk prevention

      Chronic sequela of SCI contribute to notable morbidity and mortality and include development of pressure ulcers, pulmonary complications, dysautonomia, cardiovascular disease, metabolic derangements, sexual dysfunction, urinary and bowel dysregulation.
      • Brienza D.
      • Krishnan S.
      • Karg P.
      • Sowa G.
      • Allegretti A.L.
      Predictors of pressure ulcer incidence following traumatic spinal cord injury: a secondary analysis of a prospective longitudinal study.
      ,
      • Sezer N.
      • Akkus S.
      • Ugurlu F.G.
      Chronic complications of spinal cord injury.
      Pressure ulcers are one of the most common complications of SCI found in as many as 85% of patients, and pose a significant risk for infection and sepsis that is responsible for 8% of deaths in those with SCI .
      • Fu J.
      • Jones M.
      • Jan Y.K.
      Development of intelligent model for personalized guidance on wheelchair tilt and recline usage for people with spinal cord injury: methodology and preliminary report.
      ,
      • Byrne D.W.
      • Salzberg C.A.
      Major risk factors for pressure ulcers in the spinal cord disabled: a literature review.
      Despite its significant cost to the healthcare system,
      • Reddy M.
      • Gill S.S.
      • Rochon P.A.
      Preventing pressure ulcers: a systematic review.
      pressure ulcers in SCI have not been recognized for the variability in individual need for positioning changes on a patient by patient basis.
      • Aissaoui R.
      • Kauffmann C.
      • Dansereau J.
      • de Guise J.A.
      Analysis of pressure distribution at the body-seat interface in able-bodied and paraplegic subjects using a deformable active contour algorithm.
      In 2014, Fu and colleagues create an intelligent model through machine learning (artificial neural network ( ANN ), random forest (RF), and support vector machines (SVM)) to identify the optimal wheelchair tilt and recline to mitigate the risk of pressure ulcers for each patient.
      • Fu J.
      • Jones M.
      • Jan Y.K.
      Development of intelligent model for personalized guidance on wheelchair tilt and recline usage for people with spinal cord injury: methodology and preliminary report.
      As wheelchair tilt and recline are the most efficacious biomechanical features of reducing ulcer risk through improving tissue blood flow, Doppler recordings of varying blood flow metrics at graded wheelchair angles allowed intelligent model algorithms to predict optimal patient positioning. Of the ML evaluations, SVM had the highest predictive accuracy for subset attributes at 78.8% after comparison with ANN, RF, C4.5. Future models in machine learning may capitalize on opportunities to address the multitude of complications following spinal cord injury and stratify risk based on personal characteristics and differing patient needs.

      4.5 Prognosis and recovery

      Functional prognosis after spinal cord injury with potential to regain walking ability and functional independence is a focus of rehabilitation for clinicians and patients in the setting of traumatic SCI.
      • van Middendorp J.J.
      • Goss B.
      • Urquhart S.
      • Atresh S.
      • Williams R.P.
      • Schuetz M.
      Diagnosis and prognosis of traumatic spinal cord injury.
      While the natural history of SCI recovery may be difficult to predict, ASIA grade with injury severity determination of complete versus incomplete motor impairments has been most frequently used as the benchmark for clinical recovery and provided inference for long-term functional prognosis.
      • van Middendorp J.J.
      • Goss B.
      • Urquhart S.
      • Atresh S.
      • Williams R.P.
      • Schuetz M.
      Diagnosis and prognosis of traumatic spinal cord injury.
      • Curt A.
      • Dietz V.
      Ambulatory capacity in spinal cord injury: significance of somatosensory evoked potentials and ASIA protocol in predicting outcome.
      • Burns A.S.
      • Ditunno J.F.
      Establishing prognosis and maximizing functional outcomes after spinal cord injury: a review of current and future directions in rehabilitation management.
      However, nuances of motor-level recovery and functional differences within ASIA grades have been shown to clinically meaningful for patients with SCI.
      • Ter Wengel P.V.
      • Post M.W.M.
      • Martin E.
      • et al.
      Neurological recovery after traumatic spinal cord injury: what is meaningful? A patients' and physicians' perspective.
      For example, improving a motor level from a C5 ASIA A grade to C7 or C8 ASIA A with improved upper extremity and hand function was shown to be perceived by patients as significant in value for QoL and independence.
      • Ter Wengel P.V.
      • Post M.W.M.
      • Martin E.
      • et al.
      Neurological recovery after traumatic spinal cord injury: what is meaningful? A patients' and physicians' perspective.
      Several guidelines for prediction of ambulation have aided in prognostication after tSCI, including an accurate and user-friendly prediction rule from 2011
      • van Middendorp J.J.
      • Hosman A.J.
      • Donders A.R.
      • et al.
      A clinical prediction rule for ambulation outcomes after traumatic spinal cord injury: a longitudinal cohort study.
      that incorporates age, lower extremity muscle group function assessments and dermatomal assessments to predict ambulatory ability through logistic regression with high accuracy. ML may further aid in ambulatory prognosis and personalize rehabilitative care and prediction of ability to regain independence for patients with SCI. In 2016, Zariffa et al. used ML to define the predictive value of impairment measures as through a Graded Redefined Assessment of Strength, Sensibility and Prehension (GRASSP) in traumatic cervical spinal cord injury (SCI).
      • Zariffa J.
      • Curt A.
      • Verrier M.C.
      • et al.
      Predicting task performance from upper extremity impairment measures after cervical spinal cord injury.
      A dataset of 138 GRASSP assessments was analyzed, and machine learning classifiers were trained to predict scores on prehension performance tasks. The study found that upper extremity motor impairment measures were highly predictive of task performance (Spearman's p of 0.84), indicating that automated assessments of upper extremity function after SCI can measure impairment and estimate other measures accordingly. In a prospective study by Facchinello et al., regression tree analysis of clinical and demographic parameters was used to predict functional outcome following traumatic spinal cord injury.
      • Facchinello Y.
      • Beausejour M.
      • Richard-Denis A.
      • Thompson C.
      • Mac-Thiong J.M.
      Use of regression tree analysis for predicting the functional outcome after traumatic spinal cord injury.
      Models using 4 and 11 predictors explained 51.4% and 62.3% of the variance of the Spinal Cord Independence Measure (SCIM) within the first year after injury after validation, respectively. More predictive variables enhanced accuracy and the most important predictor with discriminatory power was found to be the severity of the neurological deficit (ASIA grade) on admission, followed by Injury Severity Score (ISS), age, neurological level, and delay before surgery. In 2021, Okimatsu and colleagues developed a deep learning-based radiomics model (DLR) to quantify radiographic characteristics using a convolutional neural network (CNN) to stratify 1-month neurological outcomes of patients with acute spinal cord injury (SCI).
      • Okimatsu S.
      • Maki S.
      • Furuya T.
      • et al.
      Determining the short-term neurological prognosis for acute cervical spinal cord injury using machine learning.
      The goal of the study was to parse injury heterogeneity across 215 medical records and predict short term functional prognosis of patients with cervical SCI. Accuracy of the final model was 71% in prediction of outcome. In 2019, DeVries et al. developed an unsupervised ML model that assessed the functional independence measure (FIM) locomotion score to predict ambulatory ability at 1 year following injury and compare results with previous predictive models.
      • DeVries Z.
      • Hoda M.
      • Rivers C.S.
      • et al.
      Development of an unsupervised machine learning algorithm for the prognostication of walking ability in spinal cord injury patients.
      The unsupervised model used complete neurological data from index hospitalization to inform the algorithm and assess if this improved accuracy of walking prediction compared to previously described non-ML algorithms.
      • van Middendorp J.J.
      • Hosman A.J.
      • Donders A.R.
      • et al.
      A clinical prediction rule for ambulation outcomes after traumatic spinal cord injury: a longitudinal cohort study.
      Ultimately, no significant improvement in accuracy was achieved with more data in the unsupervised ML model compared to prior studies. Additionally, it was shown that F1-score for accuracy and validation was less inclined to inaccuracies from false negatives that AUC failed to capture. In 2018, McCoy and colleagues as part of the TRACK-SCI group use a prospectively derived 2D convolutional neural network (CNN) for automatic segmentation and classification of cord and contusion injury in patients with acute traumatic SCI.
      • McCoy D.B.
      • Dupont S.M.
      • Gros C.
      • et al.
      Convolutional neural network-based automated segmentation of the spinal cord and contusion injury: deep learning biomarker correlates of motor impairment in acute spinal cord injury.
      For 47 patients within 24 h of injury, an image-analysis pipeline with 2D CNNs was used for whole spinal cord and intramedullary spinal cord segmentation. Linear mixed modeling was used to compare the results of this study's CNN segmentation and current tools. The segmentation method using the BASICseg-3 algorithm developed in this study compares favorably with available state-of-the-art segmentation methods and correlated with lower extremity motor scores post-injury. As such, modeling of segmentation may stratify injury severity, classification, and contribute toward triage and prognosis in blunt SCI.

      4.6 Future directions

      In recent years, increased focus on molecular pathways of injury and recovery have elucidated the natural history of spinal cord injury and its management. Future studies may use machine learning in the basic science level to identify biological substrates, inflammatory markers, and pharmaceuticals to drive discovery in therapeutic developments. In 2015, Al-Ali used ML to locate kinases that aid in axonal regeneration following spinal cord injury in rat models.
      • Al-Ali H.
      • Lee D.H.
      • Danzi M.C.
      • et al.
      Rational polypharmacology: systematically identifying and engaging multiple drug targets to promote axon growth.
      In coming years, with the potential advent of spinal cord perfusion pressure monitoring, ML will surely have a role in validating optimal goals for perfusion pressures with consideration of the many injury and demographic characteristics. Many ML studies have evaluated functional and motor prognosis after SCI, however, an even more prominent concern for patient QoL is the need to address bowel and bladder dysregulation assessments
      • van Middendorp J.J.
      • Goss B.
      • Urquhart S.
      • Atresh S.
      • Williams R.P.
      • Schuetz M.
      Diagnosis and prognosis of traumatic spinal cord injury.
      and likelihood of recovery. ML may also aid in symptomatic management of spasticity to help navigate variability in dosing, efficacy, and complications related to anti-spastic agents such as baclofen.
      • Dietz N.
      • Wagers S.
      • Harkema S.J.
      • D'Amico J.M.
      Intrathecal and oral baclofen use in adults with spinal cord injury (SCI): a systematic review of efficacy in spasticity reduction and functional improvement, dosing and adverse events.
      Further, as interventions such as spinal cord stimulation for spinal cord injury,
      • Mesbah S.
      • Ball T.
      • Angeli C.
      • et al.
      Predictors of volitional motor recovery with epidural stimulation in individuals with chronic spinal cord injury.
      closed loop feedback and integration of sensing evoked continuous action potentials to optimize therapeutic intervention for motor control and pain may be augmented by ML capability. SVM ML modeling may be used as it has been shown to be more accurate when compared to alternative forms of ML. SVM modeling creates linear regression through a hyperplane that manages large volumes of input variables and optimizes distance between outcomes.
      • Ben-Hur A.
      • Weston J.
      A user's guide to support vector machines.
      SVM also has improved accuracy for classification in small sample sizes and has been shown to have promise in clinical application for prediction of heart failure and cancer genomics.
      • Son Y.J.
      • Kim H.G.
      • Kim E.H.
      • Choi S.
      • Lee S.K.
      Application of support vector machine for prediction of medication adherence in heart failure patients.
      ,
      • Huang S.
      • Cai N.
      • Pacheco P.P.
      • Narrandes S.
      • Wang Y.
      • Xu W.
      Applications of support vector machine (SVM) learning in cancer genomics.
      Given its early comparative success in predicting diagnostic and prognostic management in SCI, SVM will likely see a continued role for execution in the evolution of later clinical models. F1-score for efficacy assessment may also be considered to evaluate ML models given its predictive value against standard area under the curve.
      • DeVries Z.
      • Hoda M.
      • Rivers C.S.
      • et al.
      Development of an unsupervised machine learning algorithm for the prognostication of walking ability in spinal cord injury patients.
      Clinical application of ML tools requires a judicious iterative process to evolve algorithms and build healthcare's trust on reliability and efficacy to improve patient care.

      5. Strength and limitations

      While a few studies involved prospective methods,
      • Facchinello Y.
      • Beausejour M.
      • Richard-Denis A.
      • Thompson C.
      • Mac-Thiong J.M.
      Use of regression tree analysis for predicting the functional outcome after traumatic spinal cord injury.
      most of the studies included in final analysis were retrospective in design. A limitation of the analysis is the nature of retrospective data collection, especially for informing predictive modeling that is likely best created through prospective methods. Additionally, a small cohort of studies exists in the literature regarding applications of ML in diagnosis and management of SCI. While we anticipate its growth, current literature on this topic is limited and variable in topic and design, and not conducive to meta-analysis or direct comparison. Future studies may validate ML-derived predictive scoring systems,
      • Fallah N.
      • Noonan V.K.
      • Waheed Z.
      • et al.
      Development of a machine learning algorithm for predicting in-hospital and 1-year mortality after traumatic spinal cord injury.
      for example, or compare algorithms as more investigations are completed. Also, variability in ASIA score and type for assessment and prediction should be carefully assessed and may deviate from traditional linear regression modeling for predictive algorithms. Large numbers of variables derived from small datasets may also feed unstable model behavior without clinically meaningful ouctome.
      • Chou A.
      • Torres-Espin A.
      • Kyritsis N.
      • et al.
      Expert-augmented automated machine learning optimizes hemodynamic predictors of spinal cord injury outcome.
      Indeed, severity of injury was shown to be one of the most important factors in determining long-term functional outcome.
      • Facchinello Y.
      • Beausejour M.
      • Richard-Denis A.
      • Thompson C.
      • Mac-Thiong J.M.
      Use of regression tree analysis for predicting the functional outcome after traumatic spinal cord injury.
      Finally, models should be externally validated and carefully applied before clinical use and reliance.

      6. Conclusions

      Machine learning represents a platform technology with clinical application in traumatic spinal cord injury diagnosis, prognostication, management, rehabilitation, and risk prevention of chronic complications and mental illness. SVM models showed improved accuracy when compared to other ML subtypes surveyed. Inherent variability across patients with SCI offers unique opportunity for ML and personalized medicine to drive desired outcomes and assess risks in this patient population.

      Declaration of competing interest

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

      References

        • Raju B.
        • Jumah F.
        • Ashraf O.
        • et al.
        Big data, machine learning, and artificial intelligence: a field guide for neurosurgeons.
        J Neurosurg. 2020; : 1-11
        • Haenlein M.
        • Kaplan A.
        A brief history of artificial intelligence: on the past, present, and future of artificial intelligence.
        Calif Manag Rev. 2019; 61: 5-14
        • Kaul V.
        • Enslin S.
        • Gross S.A.
        History of artificial intelligence in medicine.
        Gastrointest Endosc. 2020; 92: 807-812
        • English M.
        • Kumar C.
        • Ditterline B.L.
        • Drazin D.
        • Dietz N.
        Machine learning in neuro-oncology, epilepsy, alzheimer's disease, and schizophrenia.
        Acta Neurochir Suppl. 2022; 134: 349-361
        • Dietz N.
        • Sharma M.
        • Alhourani A.
        • et al.
        Evaluation of predictive models for complications following spinal surgery.
        J Neurol Surg Cent Eur Neurosurg. 2020; 81: 535-545
        • Dietz N.
        • Sharma M.
        • Alhourani A.
        • et al.
        Variability in the utility of predictive models in predicting patient-reported outcomes following spine surgery for degenerative conditions: a systematic review.
        Neurosurg Focus. 2018; 45: E10
        • Dietz N.
        • Ruff C.
        • Giugliano R.P.
        • Mercuri M.F.
        • Antman E.M.
        Pharmacogenetic-guided and clinical warfarin dosing algorithm assessments with bleeding outcomes risk-stratified by genetic and covariate subgroups.
        Int J Cardiol. 2020; 317: 159-166
        • Dagi T.F.
        • Barker F.G.
        • Glass J.
        Machine learning and artificial intelligence in neurosurgery: status, prospects, and challenges.
        Neurosurgery. 2021; 89: 133-142
        • Staartjes V.E.
        • Sebok M.
        • Blum P.G.
        • et al.
        Development of machine learning-based preoperative predictive analytics for unruptured intracranial aneurysm surgery: a pilot study.
        Acta Neurochir. 2020; 162: 2759-2765
        • Staartjes V.E.
        • Stumpo V.
        • Kernbach J.M.
        • et al.
        Machine learning in neurosurgery: a global survey.
        Acta Neurochir. 2020; 162: 3081-3091
        • Fallah N.
        • Noonan V.K.
        • Waheed Z.
        • et al.
        Development of a machine learning algorithm for predicting in-hospital and 1-year mortality after traumatic spinal cord injury.
        Spine J. 2022; 22: 329-336
        • Royston P.
        • Moons K.G.
        • Altman D.G.
        • Vergouwe Y.
        Prognosis and prognostic research: developing a prognostic model.
        BMJ. 2009; 338: b604
        • Khan O.
        • Badhiwala J.H.
        • Grasso G.
        • Fehlings M.G.
        Use of machine learning and artificial intelligence to drive personalized medicine approaches for spine care.
        World Neurosurg. 2020; 140: 512-518
        • Silva N.A.
        • Sousa N.
        • Reis R.L.
        • Salgado A.J.
        From basics to clinical: a comprehensive review on spinal cord injury.
        Prog Neurobiol. 2014; 114: 25-57
        • Singh A.
        • Tetreault L.
        • Kalsi-Ryan S.
        • Nouri A.
        • Fehlings M.G.
        Global prevalence and incidence of traumatic spinal cord injury.
        Clin Epidemiol. 2014; 6: 309-331
        • Dietz N.
        • Sarpong K.
        • Ugiliweneza B.
        • et al.
        Longitudinal trends and prevalence of bowel management in individuals with spinal cord injury.
        Top Spinal Cord Inj Rehabil. 2021; 27: 53-67
        • Ugiliweneza B.
        • Guest J.
        • Herrity A.
        • et al.
        A two-decade assessment of changing practice for surgical decompression and fixation after traumatic spinal cord injury - impact on healthcare utilization and cost.
        Cureus. 2019; 11e6156
        • Sharma M.
        • Dietz N.
        • Ugiliweneza B.
        • et al.
        Impact of surgical timing and approaches to health care utilization in patients undergoing surgery for acute traumatic cervical spinal cord injury.
        Cureus. 2019; 11e6166
        • Middleton J.W.
        • Lim K.
        • Taylor L.
        • Soden R.
        • Rutkowski S.
        Patterns of morbidity and rehospitalisation following spinal cord injury.
        Spinal Cord. 2004; 42: 359-367
        • Migliorini C.
        • Tonge B.
        • Taleporos G.
        Spinal cord injury and mental health.
        Aust N Z J Psychiatr. 2008; 42: 309-314
        • Center N.S.C.I.S.
        Facts and figure at a glance.
        in: Birmingham Alabama Alabama Io. 2019
        • Mesbah S.
        • Ball T.
        • Angeli C.
        • et al.
        Predictors of volitional motor recovery with epidural stimulation in individuals with chronic spinal cord injury.
        Brain. 2021; 144: 420-433
        • Dietz N.
        • Wagers S.
        • Harkema S.J.
        • D'Amico J.M.
        Intrathecal and oral baclofen use in adults with spinal cord injury (SCI): a systematic review of efficacy in spasticity reduction and functional improvement, dosing and adverse events.
        Arch Phys Med Rehabil. 2022;
        • Ter Wengel P.V.
        • Post M.W.M.
        • Martin E.
        • et al.
        Neurological recovery after traumatic spinal cord injury: what is meaningful? A patients' and physicians' perspective.
        Spinal Cord. 2020; 58: 865-872
        • Dietz N.
        • Sharma M.
        • Adams S.
        • et al.
        Enhanced recovery after surgery (eras) for spine surgery: a systematic review.
        World Neurosurg. 2019; 130: 415-426
        • Dietz N.
        • Sharma M.
        • Alhourani A.
        • et al.
        Bundled payment models in spine surgery: current challenges and opportunities, a systematic review.
        World Neurosurg. 2019; 123: 177-183
        • Agarwal N.
        • Aabedi A.A.
        • Torres-Espin A.
        • et al.
        Decision tree-based machine learning analysis of intraoperative vasopressor use to optimize neurological improvement in acute spinal cord injury.
        Neurosurg Focus. 2022; 52: E9
        • Arslan Y.Z.
        • Demirer R.M.
        • Palamar D.
        • Ugur M.
        • Karamehmetoglu S.S.
        Comparison of the data classification approaches to diagnose spinal cord injury.
        Comput Math Methods Med. 2012; 2012803980
        • Chou A.
        • Torres-Espin A.
        • Kyritsis N.
        • et al.
        Expert-augmented automated machine learning optimizes hemodynamic predictors of spinal cord injury outcome.
        PLoS One. 2022; 17e0265254
        • DeVries Z.
        • Hoda M.
        • Rivers C.S.
        • et al.
        Development of an unsupervised machine learning algorithm for the prognostication of walking ability in spinal cord injury patients.
        Spine J. 2020; 20: 213-224
        • Facchinello Y.
        • Beausejour M.
        • Richard-Denis A.
        • Thompson C.
        • Mac-Thiong J.M.
        Use of regression tree analysis for predicting the functional outcome after traumatic spinal cord injury.
        J Neurotrauma. 2021; 38: 1285-1291
        • Fu J.
        • Jones M.
        • Jan Y.K.
        Development of intelligent model for personalized guidance on wheelchair tilt and recline usage for people with spinal cord injury: methodology and preliminary report.
        J Rehabil Res Dev. 2014; 51: 775-788
        • Gibert K.
        • Garcia-Rudolph A.
        • Curcoll L.
        • Soler D.
        • Pla L.
        • Tormos J.M.
        Knowledge discovery about quality of life changes of spinal cord injury patients: clustering based on rules by states.
        Stud Health Technol Inf. 2009; 150: 579-583
        • Inoue T.
        • Manley G.T.
        • Patel N.
        • Whetstone W.D.
        Medical and surgical management after spinal cord injury: vasopressor usage, early surgerys, and complications.
        J Neurotrauma. 2014; 31: 284-291
        • McCoy D.B.
        • Dupont S.M.
        • Gros C.
        • et al.
        Convolutional neural network-based automated segmentation of the spinal cord and contusion injury: deep learning biomarker correlates of motor impairment in acute spinal cord injury.
        AJNR Am J Neuroradiol. 2019; 40: 737-744
        • Okimatsu S.
        • Maki S.
        • Furuya T.
        • et al.
        Determining the short-term neurological prognosis for acute cervical spinal cord injury using machine learning.
        J Clin Neurosci. 2022; 96: 74-79
        • Tay B.
        • Hyun J.K.
        • Oh S.
        A machine learning approach for specification of spinal cord injuries using fractional anisotropy values obtained from diffusion tensor images.
        Comput Math Methods Med. 2014; 2014276589
        • Torres-Espin A.
        • Haefeli J.
        • Ehsanian R.
        • et al.
        Topological network analysis of patient similarity for precision management of acute blood pressure in spinal cord injury.
        Elife. 2021; 10
        • Zariffa J.
        • Curt A.
        • Verrier M.C.
        • et al.
        Predicting task performance from upper extremity impairment measures after cervical spinal cord injury.
        Spinal Cord. 2016; 54: 1145-1151
        • Hetz S.P.
        • Latimer A.E.
        • Ginis K.A.
        Activities of daily living performed by individuals with SCI: relationships with physical fitness and leisure time physical activity.
        Spinal Cord. 2009; 47: 550-554
        • Fehlings M.G.
        • Vaccaro A.
        • Wilson J.R.
        • et al.
        Early versus delayed decompression for traumatic cervical spinal cord injury: results of the Surgical Timing in Acute Spinal Cord Injury Study (STASCIS).
        PLoS One. 2012; 7e32037
        • Alizadeh A.
        • Dyck S.M.
        • Karimi-Abdolrezaee S.
        Traumatic spinal cord injury: an overview of pathophysiology, models and acute injury mechanisms.
        Front Neurol. 2019; 10: 282
        • Goldberg A.L.
        • Kershah S.M.
        Advances in imaging of vertebral and spinal cord injury.
        J Spinal Cord Med. 2010; 33: 105-116
        • Bozzo A.
        • Marcoux J.
        • Radhakrishna M.
        • Pelletier J.
        • Goulet B.
        The role of magnetic resonance imaging in the management of acute spinal cord injury.
        J Neurotrauma. 2011; 28: 1401-1411
        • Kim M.
        • Yun J.
        • Cho Y.
        • et al.
        Deep learning in medical imaging.
        Neurospine. 2019; 16: 657-668
        • Hawryluk G.
        • Whetstone W.
        • Saigal R.
        • et al.
        Mean arterial blood pressure correlates with neurological recovery after human spinal cord injury: analysis of high frequency physiologic data.
        J Neurotrauma. 2015; 32: 1958-1967
        • Walters B.C.
        • Hadley M.N.
        • Hurlbert R.J.
        • et al.
        Guidelines for the management of acute cervical spine and spinal cord injuries: 2013 update.
        Neurosurgery. 2013; 60: 82-91
        • Saadeh Y.S.
        • Smith B.W.
        • Joseph J.R.
        • et al.
        The impact of blood pressure management after spinal cord injury: a systematic review of the literature.
        Neurosurg Focus. 2017; 43: E20
        • Yue J.K.
        • Tsolinas R.E.
        • Burke J.F.
        • et al.
        Vasopressor support in managing acute spinal cord injury: current knowledge.
        J Neurosurg Sci. 2019; 63: 308-317
        • Brienza D.
        • Krishnan S.
        • Karg P.
        • Sowa G.
        • Allegretti A.L.
        Predictors of pressure ulcer incidence following traumatic spinal cord injury: a secondary analysis of a prospective longitudinal study.
        Spinal Cord. 2018; 56: 28-34
        • Sezer N.
        • Akkus S.
        • Ugurlu F.G.
        Chronic complications of spinal cord injury.
        World J Orthoped. 2015; 6: 24-33
        • Byrne D.W.
        • Salzberg C.A.
        Major risk factors for pressure ulcers in the spinal cord disabled: a literature review.
        Spinal Cord. 1996; 34: 255-263
        • Reddy M.
        • Gill S.S.
        • Rochon P.A.
        Preventing pressure ulcers: a systematic review.
        JAMA. 2006; 296: 974-984
        • Aissaoui R.
        • Kauffmann C.
        • Dansereau J.
        • de Guise J.A.
        Analysis of pressure distribution at the body-seat interface in able-bodied and paraplegic subjects using a deformable active contour algorithm.
        Med Eng Phys. 2001; 23: 359-367
        • van Middendorp J.J.
        • Goss B.
        • Urquhart S.
        • Atresh S.
        • Williams R.P.
        • Schuetz M.
        Diagnosis and prognosis of traumatic spinal cord injury.
        Global Spine J. 2011; 1: 1-8
        • Curt A.
        • Dietz V.
        Ambulatory capacity in spinal cord injury: significance of somatosensory evoked potentials and ASIA protocol in predicting outcome.
        Arch Phys Med Rehabil. 1997; 78: 39-43
        • Burns A.S.
        • Ditunno J.F.
        Establishing prognosis and maximizing functional outcomes after spinal cord injury: a review of current and future directions in rehabilitation management.
        Spine. 2001; 26: S137-S145
        • van Middendorp J.J.
        • Hosman A.J.
        • Donders A.R.
        • et al.
        A clinical prediction rule for ambulation outcomes after traumatic spinal cord injury: a longitudinal cohort study.
        Lancet. 2011; 377: 1004-1010
        • Al-Ali H.
        • Lee D.H.
        • Danzi M.C.
        • et al.
        Rational polypharmacology: systematically identifying and engaging multiple drug targets to promote axon growth.
        ACS Chem Biol. 2015; 10: 1939-1951
        • Ben-Hur A.
        • Weston J.
        A user's guide to support vector machines.
        Methods Mol Biol. 2010; 609: 223-239
        • Son Y.J.
        • Kim H.G.
        • Kim E.H.
        • Choi S.
        • Lee S.K.
        Application of support vector machine for prediction of medication adherence in heart failure patients.
        Healthc Inform Res. 2010; 16: 253-259
        • Huang S.
        • Cai N.
        • Pacheco P.P.
        • Narrandes S.
        • Wang Y.
        • Xu W.
        Applications of support vector machine (SVM) learning in cancer genomics.
        Cancer Genomics Proteomics. 2018; 15: 41-51