Deep Learning-Based Prediction Model for Gait Recovery after a Spinal Cord Injury

Predicting how well someone will walk after a spinal cord injury (SCI) is important for planning their rehabilitation. This study uses deep learning to create a model that predicts gait recovery after SCI when patients leave the hospital. The model uses data from 405 patients with acute SCI. It looks at factors like basic information, scores from neurological tests, bladder function, initial walking ability, and nerve responses in the legs. The study found that a recurrent neural network (RNN) model was much better at predicting gait recovery than other methods like linear regression. The most important factors were leg strength and the level of the spinal cord injury.

• 1
  The recurrent neural network (RNN) model significantly outperformed linear regression, Ridge, and Lasso methods in predicting gait recovery after SCI.
• 2
  Lower-extremity motor strength (ankle dorsiflexors, knee extensors, and long toe extensors) and the neurological level of injury were identified as key predictors of gait recovery.
• 3
  Initial Functional Ambulation Category (FAC) was also a significant predictor, ranking among the top predictors for all participants, those with trauma, and those without trauma.