About

This is an API to serve disease diagnostics based on UMLS databases. Disease diagnostics engine is a part of Sparrow AI. The API has endpoints to:

1. Get disease names (with probabilities) given symptoms
2. Suggest symptoms (correct names, related symptoms)
3. Get information on diseases, medicines

This repo acts as core part of SparrowAI. The API is triggered by IBM Watson Assistant via Sparrow middleware.

Technical details

This project is a FLASK based API interfacing database (Currently SQLite). Due to large size of the database, we have not hosted the database on cloud, but have created the API in a way that changing sqlite.py in umls folder to custom database connector will easily help us migrate the project to cloud based database like IBM DB2.

Installation:

Dockerhub link - https://hub.docker.com/r/jaylohokare/diseases-predictor

Quick setup

#Command to directly pull image from dockerhub and run
docker run -d -p 5000:5000 jaylohokare/diseases-predictor:0.5

#Download the docker compose file and run
docker-compose up

The API will be live at 192.168.99.100:5000 (Windows) 127.0.0.1:5000 (Linux/Mac)

To build using dockerfile:

1. Download database umls.db and place it in umls/databases - https://drive.google.com/file/d/1BmnzCv2hjU49vXxBv_kI4FUae-lIAdz0/view?usp=sharing

2. Download 'data' folder and place it at root location - https://drive.google.com/file/d/1nwD3aEMwW5CF8CWOscOT3IMztuNBpaYq/view?usp=sharing

3. Docker build command -

docker build -m 25g -t medicalapi .

To run after building -

docker run -d -p 5000:5000 medicalapi

Other docker useful commands-

#Clear all docker images cache
docker system prune -a

#Get all docker instances
docker container ls

#List all containers (only IDs)
docker ps -aq

#Stop all running containers
docker stop $(docker ps -aq)

#Remove all containers
docker rm $(docker ps -aq)

#Remove all images
docker rmi $(docker images -q)

#To push to docker hub
docker images 
docker tag ID jaylohokare/diseases-predictor:0.3
docker push jaylohokare/diseases-predictor

API format

UMLS databases are based on CUIs for medical lexicons. We have built couple of APIs to convert plain text symptoms to CUIs and then use these CUIs to predict diseases.

/api/validateSymptom

Input:
Takes symptom name (Need not be perfect match to content in DB)
Example - {"name" : string_SymptomName}

Output:
Returns CUI if entered name is exact match to DB content, or returns list of symptoms that are closest match and their CUIs
Example - {"CUI1": string_Symptom1, "CUI2": string_Symptom2....}

/api/getDetails

Input:
Takes symptom CUI 
Example - {"code" : string_CUI}

Output:
Returns description from UMLS

/api/sySuggest

Input:
Takes symptom CUI 
Example - {"symptom" : string_CUI}

Output:
Returns list of relevant symptoms
Example - [{"label": string_SymptomName, "value": string_SymptomCUI}, {"label": string_SymptomName, "value": string_SymptomCUI}, ....]

/api/predict

Input:
Takes list of symptoms CUIs as input
Example - {"symptoms" :[string_SymptomCUI1, string_SymptomCUI2, ...]}

Output:
Returns list of diseases with their CUIs, probabilities, related symptoms
Example - [{"disease": string_DiseaseName, "disease_cui": string_DiseaseCUI, "prob": float_Probability, "sy": arrayOfStrings_relatedSymptoms}, {}...]

/api/predictFromString

Input:
Takes | delimited string of symptom names as input
Example - {"symptoms" :"symptom1|sympto2|"}

Output:
Returns list of diseases with their CUIs, probabilities, related symptoms
Example - [{"disease": string_DiseaseName, "disease_cui": string_DiseaseCUI, "prob": float_Probability, "sy": arrayOfStrings_relatedSymptoms}, {}...]