Machine Learning-Based Continuous Intracranial Pressure Prediction for Traumatic Injury Patients

Area of Science:

• Neurology
• Artificial Intelligence
• Intensive Care Medicine

Background:

• Abnormal intracranial pressure (ICP) elevation poses life-threatening risks in intensive care units (ICUs).
• Early detection of high ICP events is critical for patient survival.
• Machine learning (ML) applications for continuous ICP monitoring and prediction are underexplored.

Purpose of the Study:

• To develop an efficient artificial recurrent neural network (RNN) for continuous, early prediction of ICP evaluation in traumatic brain injury (TBI) patients.
• To enable real-time ICP monitoring and short-term event prediction.
• To improve clinical interventions for TBI management.

Main Methods:

• ICP data preprocessing was performed.
• A recurrent neural network (RNN) model was trained using 20-minute ICP signal history.
• The model predicted ICP signal occurrence and classified events for the subsequent 10 minutes for 13 TBI patients.

Main Results:

• The ML model demonstrated an average accuracy of 94.62%.
• Average sensitivity was 74.91%, and average specificity was 94.83%.
• The average root mean square error (RMSE) was approximately 2.18 mmHg.

Conclusions:

• The developed ML model effectively addresses the clinical challenge of managing TBI patients with ICP fluctuations.
• The model provides continuous, real-time ICP prediction, facilitating timely clinical interventions.
• The study highlights the model's high adaptive performance, accuracy, and efficiency in ICP monitoring.