Predicting Complete Injury in Spinal Cord Injury Patients by Applying Machine Learning Methods to Heart Rate...

Area of Science:

• Biomedical Engineering
• Computational Medicine
• Data Science in Healthcare

Background:

• Spinal cord injury (SCI) diagnosis and monitoring are critical for patient outcomes.
• Heart rate variation (HRV) parameters offer potential biomarkers for physiological assessment.
• Machine learning (ML) presents opportunities for developing novel diagnostic tools.

Purpose of the Study:

• To develop and evaluate ML models for diagnosing complete and incomplete SCI.
• To utilize HRV parameters and demographic data for injury classification.
• To identify key features contributing to accurate SCI diagnosis.

Main Methods:

• Trained multiple ML algorithms including Random Forest, SVM, and neural networks on 296 patient datasets.
• Included 11 HRV parameters and demographic data in the model training.
• Employed feature selection techniques to optimize model performance and identify significant predictors.

Main Results:

• A feature-selected MLPClassifier achieved 85.33% accuracy (AUC 0.8590).
• A feedforward neural network model reached 86.67% accuracy (AUC 0.9608).
• Key predictive features included spinal injury location, age, mean heart rate, and mean R-R interval.

Conclusions:

• ML algorithms trained on HRV data show significant potential for SCI diagnosis.
• These models can serve as valuable tools for monitoring SCI patients.
• Improved diagnosis and monitoring can contribute to enhancing the quality of life for individuals with SCI.