Clinical Assistant Diagnosis for Electronic Medical Record Based on Convolutional Neural Network

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Natural Language Processing

Background:

• Current clinical decision support (CDS) and natural language processing (NLP) systems often rely on manually created knowledge bases and rule-based matching for disease diagnosis.
• This approach can be labor-intensive and may not capture the full complexity of clinical data.

Purpose of the Study:

• To develop and evaluate an automated clinical intelligent decision approach using Convolutional Neural Networks (CNNs).
• To enable automatic disease diagnosis from electronic medical records (EMRs) without requiring artificial rule or knowledge base construction.

Main Methods:

• A CNN model was developed to automatically extract high-level semantic information from EMRs.
• The model was trained and tested on a dataset of 18,590 real-world clinical EMRs.

Main Results:

• The proposed CNN model achieved a high accuracy of 98.67%.
• The model demonstrated a recall rate of 96.02% in disease diagnosis.
• These results indicate the feasibility and effectiveness of using CNNs for automated EMR analysis and diagnosis.

Conclusions:

• Convolutional Neural Networks are effective for automatically learning semantic features from EMRs.
• This automated approach offers a viable and efficient alternative to traditional rule-based systems for clinical decision support and disease diagnosis.