Related Experiment Video
Updated: Apr 30, 2026
07:02
Autologous Blood Injection to Model Spontaneous Intracerebral Hemorrhage in Mice
Published on: August 24, 2011
14.8K
Knowledge Graph-Enhanced Deep Learning Model (H-SYSTEM) for Hypertensive Intracerebral Hemorrhage: Model Development
Yulong Xia1, Jie Li2, Bo Deng1
1Department of Neurosurgery, The First Affiliated Hospital of Chongqing Medical University, No. 1 Youyi Road, Yuanjiagang, Yuzhong District, Chongqing, China, 86 13638354200.
Journal of Medical Internet Research
|June 12, 2025
Summary
The H-SYSTEM, an AI decision support tool, accurately assists neurosurgeons with hypertensive intracerebral hemorrhage diagnosis and treatment. This knowledge graph-enhanced system provides explainable plans, outperforming existing AI and showing high reliability compared to doctors.
Area of Science:
- Artificial Intelligence in Medicine
- Clinical Decision Support Systems
- Neuroscience
Background:
- AI development faces challenges in clinical practice, particularly for complex tasks like personalized treatment planning.
- Large language models (LLMs) show promise but struggle with the specialized knowledge and patient complexity in clinical settings.
Purpose of the Study:
- To develop H-SYSTEM, an explainable and efficient AI decision support system for neurosurgeons treating hypertensive intracerebral hemorrhage.
- To enhance AI decision-making accuracy and explainability by integrating a medical domain knowledge graph.
Main Methods:
- H-SYSTEM integrates named entity recognition (NER), semantic analysis, and reasoning modules.
- A medical knowledge graph (HKG) for hypertensive intracerebral hemorrhage was constructed to guide system modules and improve explainability.
- The system's performance was evaluated against neurosurgical doctors and large language models (LLMs).
Main Results:
- The H-SYSTEM achieved 94.87% accuracy compared to neurosurgical doctors.
- The BERT-IDCNN-BiLSTM-CRF model, used for NER, demonstrated superior performance (F1-score=91.11).
- H-SYSTEM showed high accuracy in diagnosis (88.18%), surgical therapy (98.53%), and rescue therapies (89.50%), with statistically significant consistency (P<.05).
- Compared to doctors and ChatGPT, H-SYSTEM achieved higher accuracy (95.26% vs 91.48%, P<.05).
- The system demonstrated strong generalization, achieving 92.22% accuracy across 605 patients from 6 medical centers.
Conclusions:
- H-SYSTEM offers significant efficiency and generalization in processing electronic medical records.
- The system provides explainable and detailed treatment plans, supporting rapid neurosurgical decision-making in emergencies.
- Knowledge graph-enhanced deep learning shows excellent potential for clinical practice tasks.
