With the advancement of technology and the gradual maturity of the market, the application of artificial intelligence in fields such as healthcare is becoming increasingly widespread and in-depth. Knowledge graph technology, as a means of extracting structured knowledge from massive texts and images, is becoming one of the core driving forces behind the development of artificial intelligence.
◆ ◆ ◆
The knowledge graph describes the attributes of things and the relationships between them in a structured rather than purely textual way, consisting of nodes and edges. Nodes represent entities, concepts, or attribute values; any object, location, or person can be a node; edges represent the attributes of entities or the relationships between entities. For example, a node could be an organization, such as a cardiology department, or a disease, such as hypertension; edges describe the “disease-department” relationship between hypertension and the cardiology department.
Example of a Knowledge Graph
◆ ◆ ◆
OMAHA takes the common underlying needs for digital medical knowledge in the industry as the entry point, independently constructing a Chinese medical knowledge graph— the “Hui Zhi” medical knowledge graph (referred to as “Hui Zhi” graph). Focusing on four major areas: diseases, drugs, laboratory tests, and surgical procedures, it selects clinical guidelines, clinical pathways, medical textbooks, drug instructions, and the Chinese Pharmacopoeia as knowledge sources, and extracts basic medical knowledge to form the knowledge graph using a combination of machine and manual methods.
Currently, the “Hui Zhi” graph has released disease knowledge graphs and drug knowledge graphs, totaling approximately 120,000 entities and 960,000 triplets.
Example of the ‘Hui Zhi’ Graph
This article will introduce how to leverage the ‘Hui Zhi’ graph in health Q&A, rational medication, and other scenarios to achieve intelligent solutions.
Scenario 1: Intelligent Triage
The complexity of medical knowledge makes it difficult for many patients to easily and accurately obtain consultation information. In this context, AI intelligent triage has emerged. When using intelligent triage products, patients input their age, gender, and disease symptoms, and the AI engine understands the patient’s condition and the corresponding consultation department. Recommendations based on the ‘Hui Zhi’ graph can effectively enhance accuracy and interpretability:
The disease knowledge graph of the ‘Hui Zhi’ graph contains over 80,000 relationship data points from authoritative medical knowledge bases, such as “disease-department” and “clinical manifestations of diseases,” covering more than 90 departments, which can be used to establish a disease consultation department recommendation model, achieving predictions for over 30,000 diseases. Additionally, other disease-related relationships can be used to further expand query results. For example, providing corresponding prompts based on relationships such as “disease-symptom” and “disease-risk factors.”
The following image shows an example of intelligent triage based on the ‘Hui Zhi’ graph:
Scenario 2: Rational Medication
In traditional healthcare models, achieving rational medication faces challenges such as low efficiency of pharmaceutical services and inconsistent review capabilities and standards among pharmacists at different levels. Based on the relevant content and medication rules in the drug knowledge graph, a rational medication system can be developed to effectively warn against inappropriate prescriptions and errors related to selected drugs, assisting doctors in clinical medication decisions and improving prescription processing efficiency:
The drug knowledge graph of the ‘Hui Zhi’ graph is primarily constructed based on open data such as local drug insurance catalogs, drug instructions, and high-quality medical resources like the Chinese Pharmacopoeia and pharmacology. It includes relationships regarding the suitability or contraindications of drugs for different age groups and disease populations, as well as treatment relationships between drugs and diseases, which can serve as the knowledge base for rational medication models of over 20,000 drugs, providing support for doctors’ clinical medication decisions.(Note: The ‘Hui Zhi’ graph will release data on drug interaction relationships, which will provide more comprehensive knowledge support for rational medication scenarios. Please stay tuned.)
The following image shows an example of rational medication based on the ‘Hui Zhi’ graph:
Scenario 3: Auxiliary Diagnosis
AI-based auxiliary diagnosis technology has shown significant application value in multiple fields. The medical knowledge graph can serve as an important component of the auxiliary diagnosis knowledge base, providing interpretable basis for diagnostic recommendations.
The disease knowledge graph of the ‘Hui Zhi’ graph has established knowledge graphs for 17 subfields of diseases, including cardiovascular diseases, neurological diseases, and digestive system diseases, which can assist in achieving auxiliary diagnosis in different diagnostic processes:
1) Before the doctor makes a diagnosis, based on the knowledge graph, combined with symptoms, diseases, and clinical examination and test results from electronic medical records, quickly and intelligently analyze and interpret clinical knowledge and clinical data, recommending related diseases, symptoms, and signs;
2) After the doctor makes a diagnosis, during the consultation process, proactively alert the doctor if there are misdiagnoses, missed diagnoses, or insufficient evidence, and provide corrective suggestions.
For example, if a patient presents with symptoms of “vision loss” and “elevated blood sugar,” the system can query the ‘Hui Zhi’ graph to infer that the diseases associated with both symptoms are “diabetic retinopathy,” and recommend related examinations such as fundus fluorescein angiography. The system can prompt these results to the doctor both before and after the diagnosis to achieve the effect of auxiliary diagnosis.
Scenario 4: Literature Recommendation
The sheer volume of medical literature presents a challenge in quickly and accurately obtaining guideline information and establishing connections between literature. Applying the ‘Hui Zhi’ graph in literature recommendation can help find corresponding relationships based on search keywords and literature keywords, establishing connections between literature and accurately recommending relevant content.
In practice, OMAHA has built a guideline recommendation product based on the ‘Hui Zhi’ graph and the “Qiqiao Board” medical terminology set, using keywords as a bridge to establish relationships between guidelines, ensuring the efficiency of intelligent recommendations in clinical guideline searches and consultations, enhancing the breadth and value of applications.
For example, by searching for the keyword “COVID-19,” the ‘Hui Zhi’ graph can provide information on the pathogenic causes, clinical manifestations, and laboratory tests related to COVID-19, enabling corresponding literature recommendations. This helps users comprehensively understand the relevant knowledge system.
The complexity of medical knowledge is beyond imagination. In the future, OMAHA will continue to explore and open a new chapter in medical knowledge.
Learn More:
Get resource introduction: OMAHA (WeChat: omaha-phr)
Stay updated on the latest resources: HiTA Knowledge Service Platform hita.omaha.org.cn
Further Reading:
[August Release – Knowledge Graph] First release of knowledge graphs in the fields of endocrine and metabolic diseases, musculoskeletal system diseases
Pathway to Building the OMAHA Chinese Medical Knowledge Graph Model (OMAHA Schema)
Making Terminology ‘Speak’: Analysis of ‘Qiqiao Board’ Terminology Set Applications
OMAHA HiTA: Metadata | Terms | Knowledge Graph
To heal, we choose openness and sharing.
Click “Read Original” to experience more application examples!