LASSO regression-based machine learning model for differentiating spinal tuberculosis, pyogenic spondylitis, and...

Area of Science:

• Medical Imaging and Diagnostics
• Machine Learning in Healthcare
• Spinal Diseases

Background:

• Spinal tuberculosis, pyogenic spondylitis, and endplate osteochondritis present similar clinical and radiological features, complicating differential diagnosis.
• Accurate diagnosis is crucial for effective treatment and patient outcomes.

Purpose of the Study:

• To develop and validate an interpretable machine learning (ML) model for the differential diagnosis of spinal tuberculosis, pyogenic spondylitis, and endplate osteochondritis.
• To enhance clinical decision-making by providing an accessible diagnostic tool.

Main Methods:

• Retrospective analysis of 481 patients with confirmed diagnoses.
• Development and evaluation of six ML algorithms, selecting the optimal LASSO regression model based on 12 core features.
• Utilized SHAP analysis for model interpretability and feature importance assessment.
• Created an open-access web-based prediction calculator.

Main Results:

• The LASSO model demonstrated strong diagnostic performance in the validation cohort: AUC of 0.838 for spinal tuberculosis, 0.683 for pyogenic spondylitis, and 0.897 for endplate osteochondritis.
• SHAP analysis provided insights into feature contributions, enhancing model transparency.
• A functional online prediction tool was developed.

Conclusions:

• An interpretable ML model with good diagnostic efficiency for spinal tuberculosis and endplate osteochondritis was successfully established.
• The developed web calculator serves as a practical auxiliary tool for clinicians in differentiating these three common spinal conditions.