Multi-task Nomogram Model for Predicting Rebleeding and Survival Outcomes in Cerebral Contusion

Area of Science:

• Neuroimaging
• Medical Informatics
• Radiology

Background:

• Cerebral contusions pose significant risks of rebleeding and mortality.
• Accurate prediction of outcomes is crucial for timely clinical management.
• Current assessment methods may lack precision in identifying high-risk patients.

Purpose of the Study:

• To develop and validate a multi-task nomogram model for predicting rebleeding and survival in cerebral contusion patients.
• To integrate clinical, radiomic, and deep transfer learning features for enhanced predictive accuracy.
• To improve early risk stratification and guide personalized treatment strategies.

Main Methods:

• Retrospective analysis of 427 patients with CT-confirmed cerebral contusions.
• Integration of clinical data (Glasgow Coma Scale), CT-based radiomics, and deep transfer learning (DenseNet121).
• Construction and validation of a multi-task nomogram model using ROC curves, calibration plots, DCA, and C-index.

Main Results:

• The nomogram achieved high predictive performance for rebleeding (AUC 0.973 training, 0.959 testing).
• Excellent prognostic accuracy was demonstrated (C-index 0.857, p < 0.0001).
• Glasgow Coma Scale scores were critical, especially in moderate TBI, with interventions improving survival by 20%.

Conclusions:

• A multi-task nomogram integrating clinical, radiomic, and DTL features is a robust tool for predicting rebleeding and survival in cerebral contusion.
• The model aids in early risk assessment and personalized treatment planning.
• Integration with GCS scores facilitates targeted interventions, particularly for moderate TBI, enhancing clinical utility and patient outcomes.