Human intention recognition for trauma resuscitation: An interpretable deep learning approach for medical process...

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Decision Support Systems
• Emergency Medicine

Background:

• Trauma resuscitation is a complex, time-critical process with multiple simultaneous goals.
• Monitoring goal pursuit during resuscitation is cognitively demanding and error-prone.
• Accurate recognition of resuscitation goals can improve patient outcomes.

Purpose of the Study:

• To develop an interpretable deep learning model for automatically recognizing resuscitation goal pursuit.
• To aid clinical decision-making by inferring provider intentions from treatment activities.
• To enhance accuracy and interpretability in predicting medical events during trauma care.

Main Methods:

• A dual-GRU neural network model was trained on event logs from 381 pediatric trauma resuscitations.
• The model learned from time-level and activity-type-level features to predict goal pursuit.
• Attention weights were used to interpret model predictions, identifying critical activities and timestamps.

Main Results:

• The model achieved an AUC of 0.84 for airway stabilization and 0.83 for circulatory support.
• Key contributing activities and timestamps aligned with established clinical domain knowledge.
• The interpretable model accurately recognized provider intentions from limited treatment data.

Conclusions:

• The developed interpretable predictive model accurately recognizes provider intentions during trauma resuscitation.
• This model surpasses existing predictive models in accuracy and interpretability.
• Integration into decision-support systems can automate action tracking, optimize workflows, and ensure timely care.