Automatic Diagnosis With Efficient Medical Case Searching Based on Evolving Graphs

Abstract

The clinical data are often multimodal and consist of both structured data and unstructured data. The modeling of clinical data has become a very important and challenging problem in healthcare big data analytics. Most existing systems focus on only one type of data. In this paper, we propose a knowledge graph-based method to build the linkage between various types of multimodal data. First, we build a semantic-rich knowledge base using both medical dictionaries and practical clinical data collected from hospitals. Second, we propose a graph modeling method to bridge the gap between different types of data, and the multimodal clinical data of each patient are fused and modeled as one unified profile graph. To capture the temporal evolution of the patient's clinical case, the profile graph is represented as a sequence of evolving graphs. Third, we develop a lazy learning algorithm for automatic diagnosis based on graph similarity search. To evaluate our method, we conduct experimental studies on ICU patient diagnosis and Orthopaedics patient classification. The results show that our method could outperform the baseline algorithms. We also implement a real automatic diagnosis system for clinical use. The results obtained from the hospital demonstrate high precision.

Keywords

• Automatic diagnosis
• multimodal medical data
• graph similarity search
• evolving graphs