A Bayesian Framework for Early Risk Prediction in Traumatic Brain Injury.

Area of Science:

• Neuroscience
• Radiology
• Medical Imaging

Background:

• Early risk detection is critical for managing traumatic brain injury (TBI) treatment.
• Computed tomography (CT) scans show prognostic value, but robust predictions in large-scale TBI analysis are lacking due to incomplete records and dataset bias.
• Existing clinical data and Marshall CT classification have limitations in predicting TBI outcomes.

Purpose of the Study:

• To develop a robust Bayesian framework for risk prediction in TBI using imaging and clinical data.
• To evaluate the predictive power of CT-derived imaging features versus clinical measures for TBI outcomes.
• To assess the combined prognostic value of imaging and clinical data in TBI patients.

Main Methods:

• A Bayesian framework with mutual information-based forward feature selection was employed.
• 154 image-based features (intensity, volume, texture) were extracted from 1791 CT scans across 22 regions of interest (ROIs) using multi-atlas segmentation.
• Image features were combined with 14 clinical parameters to generate risk likelihood scores.

Main Results:

• CT imaging data alone predicted discharge disposition better (AUC 0.81) than clinical data (AUC 0.67).
• Clinical data alone predicted discharge Glasgow Coma Scale (GCS) score better (AUC 0.85) than imaging data (AUC 0.65).
• Combining imaging and clinical data improved predictions for discharge disposition (AUC 0.86) and GCS score (AUC 0.88).

Conclusions:

• CT imaging data possess significant prognostic value for TBI patients, extending beyond current clinical measures.
• The integration of CT imaging features with clinical data enhances the accuracy of TBI outcome prediction.
• This approach offers a more comprehensive risk assessment for traumatic brain injury management.