MediAssist

• 

  MEDI ASSIST - Logo

• 

  General pipeline

• 

  Feature matching pipeline

• 

  Feature extraction demo

• 

  Matching pipeline

• 

  Example feature matching

Inspiration

MediAssist is inspired by the work of the medical researcher community in Spain, who have worked relentlessly on medical term identification.

What it does

MediAssist automatically extracts keywords from clinical cases. It also can automatically compare two clinical cases, highlighting common features of both. It is a fully interpretable natural language processing system, that can help medical practioners navigate huge databases of clinical cases to make evidence-based decisions.

How we built it

The code for Medi Assist is written in Python and powered by the Natural Language Processing library spacy. It leverages lexical tagging to extract keywords from cases. It also uses semantical and lexical features to compare keywords from different cases. See the image gallery for more details and examples!

Challenges we ran into

Working with unstructured data is really really hard. We had to compromise by building a pipeline for unstructured data and another one for structured clinical cases.

Accomplishments that we are proud of

We took the idea from concept to working prototype in less than 48h!

What we learned

A lot about state-of-the-art NLP: POS tagging, semantical embeddings, syntactical matching and much more!

What's next for MediAssist

Study in depth the performance of the current system to undertstand better its limitations and improve it incrementally. Improve the recognizition of medical terms with custom semantical embeddings.

About us

We are Early Stage Researchers of the Natural Language for Explaninable Artificial Intelligence Project (NL4XAI).

Built With

• python
• spacy