Refining Diagnosis Paths for Medical Diagnosis based on an Augmented Knowledge Graph

Niclas Heilig,Jan Kirchhoff,Florian Stumpe,Joan Plepi,Lucie Flek,Heiko Paulheim

DOI: https://doi.org/10.48550/arXiv.2204.13329

2022-04-28

Abstract:Medical diagnosis is the process of making a prediction of the disease a patient is likely to have, given a set of symptoms and observations. This requires extensive expert knowledge, in particular when covering a large variety of diseases. Such knowledge can be coded in a knowledge graph -- encompassing diseases, symptoms, and diagnosis paths. Since both the knowledge itself and its encoding can be incomplete, refining the knowledge graph with additional information helps physicians making better predictions. At the same time, for deployment in a hospital, the diagnosis must be explainable and transparent. In this paper, we present an approach using diagnosis paths in a medical knowledge graph. We show that those graphs can be refined using latent representations with RDF2vec, while the final diagnosis is still made in an explainable way. Using both an intrinsic as well as an expert-based evaluation, we show that the embedding-based prediction approach is beneficial for refining the graph with additional valid conditions.

Artificial Intelligence

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to improve the integrity and accuracy of the diagnostic path by enhancing the knowledge graph in the medical diagnosis process. Specifically, the paper focuses on:

1. **Transparency and Explainability in Medical Diagnosis**: The medical diagnosis system needs to be able to provide transparent and explainable diagnosis results in order to gain the trust and acceptance of doctors. Although the existing diagnosis systems based on the knowledge graph can provide explanations, their diagnostic paths may be incomplete and lack some rare conditions or alternatives.

2. **Incompleteness of the Knowledge Graph**: Although the existing medical knowledge graphs contain information on diseases, symptoms, and diagnostic paths, due to the limitations of knowledge itself and incomplete encoding, these graphs are often incomplete. This results in the fact that in practical applications, doctors may not be able to obtain comprehensive diagnostic support.

3. **Enhancing the Knowledge Graph with Patient Data**: The paper proposes a method to supplement and improve the existing diagnostic paths by training a machine - learning model with patient data (such as data from the MIMIC - IV database). This method uses knowledge graph embedding techniques (such as RDF2vec) to predict the missing edges in the knowledge graph, that is, new diagnostic relationships.

4. **Maintaining Diagnostic Explainability**: Although knowledge graph embedding techniques can be used directly for disease prediction, this will sacrifice the explainability of the diagnosis. Therefore, the method proposed in the paper focuses on enhancing the existing diagnostic paths rather than directly predicting diseases, thereby ensuring that the final diagnosis is still explainable.

In summary, the main objective of the paper is to improve the accuracy and transparency of medical diagnosis by enhancing the knowledge graph while ensuring the explainability of the diagnosis process.