Pseudarthrosis following adult spinal deformity surgery may be predicted with preoperative MRI adipose tissue features: an artificial intelligence study on raw 3D imaging

Abstract

Aim: The purpose of this study is to investigate the utility of incorporating magnetic resonance imaging (MRI) into an artificial intelligence (AI) model to preoperatively predict pseudarthrosis for patients undergoing adult spinal deformity (ASD) surgery.

Methods: A retrospective cohort study was conducted on patients undergoing ASD surgery at Vanderbilt University Medical Center with at least 2 years of follow-up. We first collected demographic variables and measured traditional radiographic variables with Surgimap software. The primary outcome of interest was pseudarthrosis, defined as mechanical pain without evidence of bony union with or without a rod fracture. Next, cohort differences between patients diagnosed with and without pseudarthrosis were evaluated with t-tests for continuous variables and chi-squared tests for categorical variables using Bonferroni-Holm multiple comparison correction. Using a subpopulation of patients with preoperative thoracic MRI available, a three-dimensional convolutional neural network (3D-CNN) with five-fold nested cross-validation was developed to predict pseudarthrosis - accuracy was evaluated with the Youden index. Finally, class activation mapping (CAM) was conducted to visualize the MRI features utilized by the model for accurate classifications.

Results: Of 191 patients undergoing ASD surgery, the demographic and traditional radiographic variables were collected, and only age was observed to be significantly different between the patients diagnosed with pseudarthrosis (69.9 ± 10.1 years old) and those without (60.9 ± 19.9), with a t-test P-value of 0.003. The 3D-CNN demonstrated an average Youden index of 0.49 ± 0.25 on the withheld data, with a P-value of 5.50e-3 compared to an equivocal null model. Finally, CAM consistently revealed posterior adipose tissue to be most important in preoperatively predicting pseudarthrosis.

Conclusion: Adipose tissue features in MRI, independent of body mass index (BMI), may be useful for preoperatively predicting pseudarthrosis. Overall, this work demonstrates the capabilities of raw imaging AI in spine surgery and can serve as the basis for a deeper biological inquiry into the pathogenesis of pseudarthrosis.

Keywords

Adult spinal deformity / artificial intelligence / deep learning / machine learning / magnetic resonance imaging / pseudarthrosis

References

[1] How NE,Dvorak MF.Pseudarthrosis in adult and pediatric spinal deformity surgery: a systematic review of the literature and meta-analysis of incidence, characteristics, and risk factors.Neurosurg Rev2019;42:319-36

[2] Deyo RA,Martin BI,Goodman DC.Trends, major medical complications, and charges associated with surgery for lumbar spinal stenosis in older adults.JAMA2010;303:1259-65 PMCID:PMC2885954

[3] Buchholz AL,Shaffrey CI. Postoperative spinal deformities: kyphosis, nonunion, and loss of motion segment. In: Complications in neurosurgery. 2019. pp. 325-30. Available from: http://103.203.175.90:81/fdScript/RootOfEBooks/E%20Book%20Collection%202021%20-%20A/ENGLISH/Complications%20in%20Neurosurgery.pdf. [Last accessed on 7 Nov 2024]

[4] Kim YJ,Lenke LG,Cheh G.Pseudarthrosis in long adult spinal deformity instrumentation and fusion to the sacrum: prevalence and risk factor analysis of 144 cases.Spine2006;31:2329-36

[5] Kim YJ,Lenke LG,Edwards C 2nd.Pseudarthrosis in primary fusions for adult idiopathic scoliosis: incidence, risk factors, and outcome analysis.Spine2005;30:468-74

[6] Kim YJ,Lenke LG,Edwards CC 2nd.Pseudarthrosis in adult spinal deformity following multisegmental instrumentation and arthrodesis.J Bone Joint Surg Am2006;88:721-8

[7] Pateder DB,Kebaish KM.Spinal fusion after revision surgery for pseudarthrosis in adult scoliosis.Spine2006;31:E314-9

[8] Joshi RS,Lau D.Artificial intelligence for adult spinal deformity.Neurospine2019;16:686-94 PMCID:PMC6944987

[9] Joshi RS,Ames CP.Artificial intelligence for adult spinal deformity: current state and future directions.Spine J2021;21:1626-34

[10] Lenke LG.Commentary: artificial intelligence for adult spinal deformity.Neurospine2019;16:695-6 PMCID:PMC6944988

[11] Perez-Breva L.Artificial intelligence in neurosurgery: a comment on the possibilities.Neurospine2019;16:640-2 PMCID:PMC6944990

[12] Durand WM, Lafage R, Hamilton DK, et al; International Spine Study Group (ISSG). Artificial intelligence clustering of adult spinal deformity sagittal plane morphology predicts surgical characteristics, alignment, and outcomes. Eur Spine J 2021;30:2157-66.

[13] Durand WM, Daniels AH, Hamilton DK, et al; International Spine Study Group. Artificial intelligence models predict operative versus nonoperative management of patients with adult spinal deformity with 86% accuracy. World Neurosurg 2020;141:e239-53.

[14] Scheer JK, Smith JS, Schwab F, et al; International Spine Study Group. Development of a preoperative predictive model for major complications following adult spinal deformity surgery. J Neurosurg Spine 2017;26:736-43.

[15] Scheer JK, Osorio JA, Smith JS, et al; International Spine Study Group. Development of validated computer-based preoperative predictive model for proximal junction failure (PJF) or clinically significant PJK with 86% accuracy based on 510 ASD patients with 2-year follow-up. Spine 2016;41:E1328-35.

[16] Pellisé F, Serra-Burriel M, Smith JS, et al; International Spine Study Group, European Spine Study Group. Development and validation of risk stratification models for adult spinal deformity surgery. J Neurosurg Spine 2019;31:587-99.

[17] Scheer JK, Oh T, Smith JS, et al; International Spine Study Group. Development of a validated computer-based preoperative predictive model for pseudarthrosis with 91% accuracy in 336 adult spinal deformity patients. Neurosurg Focus 2018;45:E11.

[18] Johnson GW,Ali MA.Artificial intelligence to preoperatively predict proximal junction kyphosis following adult spinal deformity surgery: soft tissue imaging may be necessary for accurate models.Spine2023;48:1688-95 PMCID:PMC11101214

[19] Kernbach JM.Foundations of machine learning-based clinical prediction modeling: part II - generalization and overfitting. In: Staartjes VE, Regli L, Serra C, editors. Machine learning in clinical neuroscience. Springer, Cham; 2022. pp. 15-21.

[20] Youden WJ.Index for rating diagnostic tests.Cancer1950;3:32-5

[21] Selvaraju RR,Das A,Parikh D.Grad-CAM: visual explanations from deep networks via gradient-based localization.Int J Comput Vis2020;128:336-59

[22] Adogwa O,Lenke LG.Comparison of rod fracture rates in long spinal deformity constructs after transforaminal versus anterior lumbar interbody fusions: a single-institution analysis.J Neurosurg Spine2020;32:42-9

[23] Passias PG,Alas H.Alcoholism as a predictor for pseudarthrosis in primary spine fusion: an analysis of risk factors and 30-day outcomes for 52,402 patients from 2005 to 2013.J Orthop2019;16:36-40 PMCID:PMC6324756

[24] Marques MF, Fiere V, Obeid I, et al; Société Française de Chirurgie Rachidienne, SFCR. Pseudarthrosis in adult spine deformity surgery: risk factors and treatment options. Eur Spine J 2021;30:3225-32.

[25] Kawabata A,Sakai K.Identification of predictive factors for mechanical complications after adult spinal deformity surgery: a multi-institutional retrospective study.Spine2020;45:1185-92

[26] Kuo CC,Aguirre AO.Vertebral bone quality score independently predicts proximal junctional kyphosis and/or failure after adult spinal deformity surgery.Neurosurgery2023;92:945-54

[27] Akbik OS,Pernik MN.A comparative analysis of frailty, disability, and sarcopenia with patient characteristics and outcomes in adult spinal deformity surgery.Global Spine J2023;13:2345-56 PMCID:PMC10538316