The Result
Created a database for online clinic chatbot to support automatic filling anamnesis from the clients. The bot is recognizing the client’s answer and filled the json document (it is very simple by using AWS service). If the quality of the json document is not sufficient (not enough data – TensorFlow) some additional questions are generated and result json documents are merged.
The Challenge
The client approached Mellivora experts to create an automatic bot for working with patients online. NLP technologies had to be used for the purpose of gathering knowledge base for the anamnesis build-up. The team was provided with the links to websites of online chats with the doctors and text records of the dialogues with the patients. Mellivora team had to find the relations to create a precise anamnesis form.
The Solution
To reach the set goal, we used Medical Named Entity and Relationship Extraction (NERe) by Amazon Comprehend which helps to return the medical information such as medication, medical condition, test, treatment and procedures (TTP), anatomy, and Protected Health Information (PHI). It also helps to identify relationships between extracted sub-types associated to Medications and TTP. There is also contextual information provided as entity “traits” (negation, or if a diagnosis is a sign or symptom).
Technology Stack
Technology stack was based on AWS and is as follows:
- Storing data: AWS S3 – storing data (input data, files)
- Data stream process engine: AWS Lambda – https://docs.aws.amazon.com/lambda/index.html#lang/en_us – to run (aws comprehend-medical automatically when file appears at the S3)
- NLP extractor: Amazon Comprehend Medical Tool – https://aws.amazon.com/ru/comprehend/medical/ (generated JSON output files)
- Medical Named Entity and Relationship Extraction (NERe);
- Medical Protected Health Information Data Extraction and Identification (PHId)
- AWS TensorFlow – https://aws.amazon.com/tensorflow/?nc1=h_ls (CNN results classifier, analytics)