Research on Rehabilitation Effect Prediction for Patients with SCI Based on Machine Learning.

Area of Science:

• Biomedical Engineering
• Rehabilitation Medicine
• Data Science

Background:

• Spinal cord injury (SCI) patients present complex conditions making accurate prediction of discharged activity of daily living (ADL) scores challenging.
• Accurate ADL score prediction is crucial for assessing rehabilitation effectiveness and guiding patient recovery post-SCI.

Purpose of the Study:

• To develop and evaluate a machine learning-based prediction model for discharged ADL scores in SCI patients.
• To enhance the accuracy of ADL score prediction to better inform rehabilitation strategies.

Main Methods:

• Collected and preprocessed medical records from 1231 SCI patients.
• Utilized Pearson correlation and Random Forest (RF) for feature selection, identifying 6 key predictors.
• Developed and compared RF and Harris Hawks Optimizer-optimized RF (HHO-RF) models for ADL score prediction.

Main Results:

• The HHO-RF model achieved superior prediction performance compared to the standard RF model.
• HHO-RF demonstrated improved evaluation metrics: MAE of 0.0821, RMSE of 0.1089, and R² of 0.8537.
• Key predictors included admission ADL score, age, injury segment, reason, position, and degree.

Conclusions:

• The HHO-RF model offers a significant advancement in accurately predicting discharged ADL scores for SCI patients.
• This predictive capability can guide clinical decisions and optimize rehabilitation program selection for improved patient outcomes.