Paraspinal Muscle Fat Infiltration as a Key Predictor of Symptomatic Intravertebral Vacuum Cleft: A Machine Learning...

Area of Science:

• Spine surgery
• Radiology
• Machine Learning

Background:

• Symptomatic intravertebral vacuum cleft (SIVC) is a painful complication of vertebral compression fractures (VCFs).
• Predicting SIVC is challenging due to its complex causes.
• Paraspinal muscle fat infiltration's role in SIVC prediction is underexplored.

Purpose of the Study:

• Develop machine learning (ML) models to predict SIVC.
• Assess the impact of muscle-related variables on SIVC prediction accuracy.

Main Methods:

• Collected demographic, radiological, and muscle-related data.
• Trained and tested ML models (Logistic Regression, Random Forest, XGBoost, Multi-Layer Perceptron) with and without muscle variables.
• Evaluated models using accuracy, AUC, and feature importance.

Main Results:

• The Random Forest model with muscle variables achieved 96.6% accuracy and 0.956 AUC.
• Multifidus and erector spinae fatty infiltration were key predictors.
• Including muscle variables significantly enhanced all ML models' performance.

Conclusions:

• ML models, especially Random Forest, accurately predict SIVC when muscle fat infiltration is considered.
• Paraspinal muscle fat infiltration is a crucial SIVC predictor.
• Integrating muscle fat infiltration into risk assessments can improve SIVC diagnosis and management.