Medical Workflow Modeling Using Alignment-Guided State-Splitting HMM.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Informatics

Background:

• Process mining is crucial for analyzing complex workflows across various domains.
• Existing methods often overlook the potential of hidden Markov models (HMMs) for workflow discovery.
• There is a need for more accurate and efficient HMM-based workflow discovery techniques.

Purpose of the Study:

• To introduce a novel algorithm, alignment-guided state-splitting HMM inference (AGSS), for workflow discovery.
• To evaluate the performance of AGSS against existing methods using real-world medical workflow data.
• To demonstrate the enhanced accuracy, efficiency, and readability of workflow models generated by AGSS.

Main Methods:

• Developed the alignment-guided state-splitting HMM inference (AGSS) algorithm.
• Applied AGSS to four real-world medical workflow datasets.
• Conducted a detailed case study on one medical workflow dataset.
• Compared AGSS with existing HMM inference algorithms and a hand-made expert model.

Main Results:

• AGSS generates more accurate and representative workflow models compared to existing methods.
• AGSS achieves significant efficiency gains (O(n)) due to trace alignment guiding state splitting.
• The AGSS-generated model was more readable and accurate than those from other algorithms.
• Identified three discrepancies between the AGSS model and the expert model, leading to expert model enhancement.

Conclusions:

• AGSS offers a superior approach for discovering workflow models using HMMs.
• The algorithm enhances both the accuracy and efficiency of process mining.
• The AGSS approach facilitates the refinement of existing expert-created workflow models in healthcare and other fields.