The Reliability of the AOSpine Thoracolumbar Classification System in Children: Results of a Multicenter Study

Evaluation and Management of Thoracolumbar Spine Trauma in Pediatric Patients: A Critical Analysis Review

» Pediatric thoracolumbar trauma, though rare, is an important cause of morbidity and mortality and necessitates early, accurate diagnosis and management.» Obtaining a detailed history and physical examination in the pediatric population can be difficult. Therefore, the threshold for advanced imaging, such as magnetic resonance imaging, is low and should be performed in patients with head injuries, altered mental status, inability to cooperate with examination, and fractures involving more than 1 column of the spine.» The classification of pediatric thoracolumbar trauma is based primarily on adult studies and there is little high-level evidence examining validity and accuracy in pediatric populations.» Injury pattern and neurologic status of the patient are the most important factors when determining whether to proceed with operative management.

The Reliability of the AO Spine Upper Cervical Classification System in Children: Results of a Multi-Center Study

BACKGROUND

There is no uniform classification system for traumatic upper cervical spine injuries in children. This study assesses the reliability and reproducibility of the AO Upper Cervical Spine Classification System (UCCS), which was developed and validated in adults, to children.

METHODS

Twenty-six patients under 18 years old with operative and nonoperative upper cervical injuries, defined as from the occipital condyle to the C2-C3 joint, were identified from 2000 to 2018. Inclusion criteria included the availability of computed tomography and magnetic resonance imaging at the time of injury. Patients with significant comorbidities were excluded. Each case was reviewed by a single senior surgeon to determine eligibility. Educational videos, schematics describing the UCCS, and imaging from 26 cases were sent to 9 pediatric orthopaedic surgeons. The surgeons classified each case into 3 categories: A, B, and C. Inter-rater reliability was assessed for the initial reading across all 9 raters by Fleiss's kappa coefficient (kF) along with 95% confidence intervals. One month later, the surgeons repeated the classification, and intra-rater reliability was calculated. All images were de-identified and randomized for each read independently. Intra-rater reproducibility across both reads was assessed using Fleiss's kappa. Interpretations for reliability estimates were based on Landis and Koch (1977): 0 to 0.2, slight; 0.2 to 0.4, fair; 0.4 to 0.6, moderate; 0.6 to 0.8, substantial; and >0.8, almost perfect agreement.

RESULTS

Twenty-six cases were read by 9 raters twice. Sub-classification agreement was moderate to substantial with α κ estimates from 0.55 for the first read and 0.70 for the second read. Inter-rater agreement was moderate (kF 0.56 to 0.58) with respect to fracture location and fair (kF 0.24 to 0.3) with respect to primary classification (A, B, and C). Krippendorff's alpha for intra-rater reliability overall sub-classifications ranged from 0.41 to 0.88, with 0.75 overall raters.

CONCLUSION

Traumatic upper cervical injuries are rare in the pediatric population. A uniform classification system can be vital to guide diagnosis and treatment. This study is the first to evaluate the use of the UCCS in the pediatric population. While moderate to substantial agreement was found, limitations to applying the UCCS to the pediatric population exist, and thus the UCCS can be considered a starting point for developing a pediatric classification.

LEVEL OF EVIDENCE

Level III.

The reliability of the AOSpine Thoracolumbar Spine Injury Classification System in children: an international validation study

PURPOSE

To evaluate the AOSpine Thoracolumbar Spine Injury Classification System and if it is reliable and reproducible when applied to the paediatric population globally.

METHODS

A total of 12 paediatric orthopaedic surgeons were asked to review MRI and CT imaging of 25 paediatric patients with thoracolumbar spine traumatic injuries, in order to determine the classification of the lesions observed. The evaluators classified injuries into primary categories: A, B and C. Interobserver reliability was assessed for the initial reading by Fleiss's kappa coefficient (k_F) along with 95% confidence intervals (CI). For A and B type injuries, sub-classification was conducted including A0-A4 and B1-B2 subtypes. Interobserver reliability across subclasses was assessed using Krippendorff's alpha (α_k) along with bootstrapped 95% CIs. A second round of classification was performed one-month later. Intraobserver reproducibility was assessed for the primary classifications using Fleiss's kappa and sub-classification reproducibility was assessed by Krippendorff's alpha (α_k) along with 95% CIs.

RESULTS

In total, 25 cases were read for a total of 300 initial and 300 repeated evaluations. Adjusted interobserver reliability was almost perfect (kF = 0.74; 95% CI 0.71 to 0.78) across all observers. Sub-classification reliability was substantial (α_k= 0.67; 95% CI 0.51 to 0.81), Adjusted intraobserver reproducibility was almost perfect (kF = 0.91; 95% CI 0.83 to 0.99) for both primary classifications and for sub-classifications (α_k = 0.88; 95% CI 0.83 to 0.93).

CONCLUSION

The inter- and intraobserver reliability for the AOSpine Thoracolumbar Spine Injury Classification System was high amongst paediatric orthopaedic surgeons. The AOSpine Thoracolumbar Spine Injury Classification System is a promising option as a uniform fracture classification in children.

LEVEL OF EVIDENCE

III.

Feasibility of Deep Learning Algorithms for Reporting in Routine Spine Magnetic Resonance Imaging

BACKGROUND

Artificial intelligence is gaining traction in automated medical imaging analysis. Development of more accurate magnetic resonance imaging (MRI) predictors of successful clinical outcomes is necessary to better define indications for surgery, improve clinical outcomes with targeted minimally invasive and endoscopic procedures, and realize cost savings by avoiding more invasive spine care.

OBJECTIVE

To demonstrate the ability for deep learning neural network models to identify features in MRI DICOM datasets that represent varying intensities or severities of common spinal pathologies and injuries and to demonstrate the feasibility of generating automated verbal MRI reports comparable to those produced by reading radiologists.

METHODS

A 3-dimensional (3D) anatomical model of the lumbar spine was fitted to each of the patient's MRIs by a team of technicians. MRI T1, T2, sagittal, axial, and transverse reconstruction image series were used to train segmentation models by the intersection of the 3D model through these image sequences. Class definitions were extracted from the radiologist report for the central canal: (0) no disc bulge/protrusion/canal stenosis, (1) disc bulge without canal stenosis, (2) disc bulge resulting in canal stenosis, and (3) disc herniation/protrusion/extrusion resulting in canal stenosis. Both the left and right neural foramina were assessed with either (0) neural foraminal stenosis absent, or (1) neural foramina stenosis present. Reporting criteria for the pathologies at each disc level and, when available, the grading of severity were extracted, and a natural language processing model was used to generate a verbal and written report. These data were then used to train a set of very deep convolutional neural network models, optimizing for minimal binary cross-entropy for each classification.

RESULTS

The initial prediction validation of the implemented deep learning algorithm was done on 20% of the dataset, which was not used for artificial intelligence training. Of the 17,800 total disc locations for which MRI images and radiology reports were available, 14,720 were used to train the model, and 3560 were used to validate against. The convergence of validation accuracy achieved with the deep learning algorithm for the foraminal stenosis detector was 81% (sensitivity = 72.4.4%, specificity = 83.1%) after 25 complete iterations through the entire training dataset (epoch). The accuracy was 86.2% (sensitivity = 91.1%, specificity = 82.5%) for the central stenosis detector and 85.2% (sensitivity = 81.8%, specificity = 87.4%) for the disc herniation detector.

CONCLUSIONS

Deep learning algorithms may be used for routine reporting in spine MRI. There was a minimal disparity among accuracy, sensitivity, and specificity, indicating that the data were not overfitted to the training set. We concluded that variability in the training data tends to reduce overfitting and overtraining as the deep neural network models learn to focus on the common pathologies. Future studies should demonstrate the accuracy of deep neural network models and the predictive value of favorable clinical outcomes with intervention and surgery.

LEVEL OF EVIDENCE

3.

CLINICAL RELEVANCE

Feasibility, clinical teaching, and evaluation study.