Medical Student University of Missouri - Kansas City Kansas City, MO, US
Introduction: The ligamentum flavum, a key anatomical structure in the spinal canal, has been implicated in the pathogenesis of spinal stenosis and related conditions. Thickening of the ligamentum flavum can lead to compression of neural structures, resulting in various clinical symptoms. However, the relationship between ligamentum flavum thickness and different types of spinal deformities remains incompletely understood.
Methods: Ligamentum flavum thickness measurements were collected at multiple spinal levels ranging from T4-S1. The thickness classifications were determined using the mean and standard deviation of measurements at each spinal level and were categorized into normal, mild, and severe thickening and plotted on a bell curve. Several machine learning models used demographic variables and thickness measurements from the dataset for training and testing. Machine learning model performance was compared using accuracy, precision, recall, F1-score, and ROC-AUC on the testing dataset.
Results: A total of 54 patients with adult spinal deformity were included in this study. The thickness of the ligamentum flavum at each level was calculated using the mean and standard deviation of each measurement at each level (L4-L5 mean thickness = 3.34 mm [95% CI 3.12, 3.57]). Thickness at each level was classified into three categories: Normal, Mild, and Severe. At the L4-L5 level, normal ligamentum flavum thickness was classified as < 3.10 mm. Mild thickening ranged from 3.10 – 3.89 mm, and severe thickening was > 3.89 mm. The percentage of patients included in our study who fell into each threshold was; Normal = 43.4%, Mild 35.8%, and Severe = 20.8%. Multiple machine learning models were evaluated on their ability to classify ligamentum flavum thickness. The Gradient Boosting Model performed the best with an Accuracy = 0.82, Precision = 0.82, Recall = 0.82, F1 Score = 0.81, and an ROC AUC = 0.93.
Conclusion : This study demonstrates the potential of machine learning techniques, specifically clustering algorithms, in categorizing patients with spinal deformities based on ligamentum flavum thickness. The Gradient Boosting Model's high performance (Accuracy = 0.82, ROC AUC = 0.93) in classifying ligamentum flavum thickness suggests that this approach could be a valuable tool in the assessment and management of spinal stenosis and related conditions.