Combining External Medical Knowledge for Improving Obstetric Intelligent Diagnosis: Model Development and Validation.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Obstetrics and Gynecology

Background:

• Data-driven processing of medical information is a growing trend in obstetrics.
• Electronic medical records (EMRs) are crucial for evidence-based medicine and intelligent diagnosis.
• Integrating clinical experience and external medical knowledge enhances diagnostic accuracy.

Purpose of the Study:

• To improve intelligent diagnosis performance in obstetrics using electronic medical records (EMRs).
• To effectively combine medical knowledge with EMR data for enhanced diagnostic capabilities.

Main Methods:

• The study treats intelligent diagnosis as a multilabel classification task.
• A novel neural network, the knowledge-aware hierarchical diagnosis model (KHDM), was developed.
• KHDM integrates EMRs and external medical knowledge using an attention mechanism within a deep learning framework.

Main Results:

• The KHDM model achieved an accuracy of 0.8929 on a Chinese obstetric EMR dataset.
• Performance surpassed existing advanced classification benchmark methods.
• The model demonstrated an advantage in interpretability.

Conclusions:

• A novel model, KHDM, was proposed to address the diversity of diagnostic results in Chinese EMRs.
• KHDM effectively integrates domain knowledge and attention mechanisms.
• The model significantly improves diagnostic accuracy by leveraging external medical knowledge.