Mine Clinical Concepts
The Wolfram Neural Net Repository contains a model that provides semantic embedding for about 110,000 clinical concepts learned from millions of publications, clinical notes and insurance claims. This example demonstrates how one can use such embeddings to define a distance between concepts and explore the space of clinical concepts.
Load the clinical-concept embedding model.
This model is trained on clinical-concept unique identifiers. Try the network on one of these identifiers.
Load a dataset of clinical concepts from the Wolfram Data Repository.
First, illustrate how the embeddings already separate virus-related concepts from bacteria-related concepts.
Find all virus-related concepts.
Extract identifiers of the virus-related concepts.
Perform the same operation for bacteria.
Visualize these concepts in a 3D representation.
Now use the embeddings to create a clinical concept search engine using FeatureNearest.
Find the 10 concepts most similar to metronidazole.
