Intention Mining in Medical Process: A Case Study in Trauma Resuscitation

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Human-Computer Interaction

Background:

• Medical teams operate with implicit intentions during procedures.
• Understanding these intentions is crucial for process improvement and patient safety.
• Current methods lack the ability to uncover multi-level intentions in complex medical tasks.

Purpose of the Study:

• To develop and evaluate an unsupervised approach for mining multi-level intentions from observed medical treatment activities.
• To apply this method to the complex real-world scenario of trauma resuscitation.

Main Methods:

• Utilized a hierarchical hidden Markov model (H-HMM) to model multi-level intentions.
• Employed a state-splitting method with maximum a posteriori probability (MAP) for accurate H-HMM inference.
• Validated the approach using a dataset of 123 trauma resuscitation cases.

Main Results:

• The intention mining approach achieved a classification accuracy of 86.6% for medical teams' intentions.
• Demonstrated the capability to identify multi-level intentions within the trauma resuscitation process.
• Qualitative and quantitative evaluations confirmed the approach's effectiveness.

Conclusions:

• Unsupervised intention mining is feasible for complex medical processes.
• The developed H-HMM-based approach effectively uncovers hidden medical team intentions.
• This work provides a foundation for improving medical process analysis and decision support.