A CT-based multimodal fusion model for predicting outcomes in blunt chest trauma: A multicenter study

Area of Science:

• Medical imaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Blunt chest trauma (BCT) poses significant prognostic challenges.
• Current prognostic models may lack comprehensive predictive power.

Purpose of the Study:

• To develop and validate a multimodal predictive model for BCT.
• Integrate clinical variables, radiomic features (RFs), and deep learning-based features (DLFs) for improved prognostic accuracy.

Main Methods:

• Retrospective analysis of 337 BCT patients from three centers.
• Extraction of clinical, radiomic, deep learning, and delta features from CT scans.
• Development of clinical-only, imaging-only, and fused models using LASSO regression for feature selection.

Main Results:

• Independent prognostic factors identified: rib fracture count, multiple injuries, hemopneumothorax-to-lung ratio (HPR).
• The delta-clinical-DLR fusion model achieved high AUCs (0.85 training, 0.86 test).
• Significant improvements in net reclassification (NRI up to 0.76) and integrated discrimination (IDI up to 0.26) were observed.

Conclusions:

• Multimodal fusion of multi-timepoint CT imaging and clinical variables enhances BCT prognostic performance.
• The developed model provides a robust tool for individualized risk prediction.
• This approach supports improved clinical decision-making in BCT management.