Aim: Reducing the false positives device alarms in ICUs.
- ICUs == Intensive Care Unit
- Average ICU takes care of 18 patients
- For every ICU patient, about 100 alarms are raised per day
REST API: Representational state Transfer, communication over the web using standard HTTP methods (GET, POST, PUT/PATCH (Update data), DELETE)
- Api documentation done with FastAPI
- Language is most likely python
- Appspot == Google App Engine (supports python apps), API is running on Google servers
Get request:
curl -X 'GET' \
'https://idalab-icu.ew.r.appspot.com/history_vital_signs' \
-H 'accept: application/json'
curl: Tool for making HTTP requests from the command line
-X 'GET': Request type
'http: ....': URL of the API endpoint
-H 'accept: application/json': Please send the response in json format
- Library for python: requests
- Headers == is a label or tag added to an HTTP request or response (metadata)
- Pooling == Repeatedly sending GET requests every few seconds to check if new data is available
- Connect to API of the hospital
- Fetch the raw real-time vital signs continuosly coming from the API
- Transform the data
- Store data in appropriate format
Environment:
conda activate env_menssana
| medical_emergency | monitor_alarm | Expected result | category |
|---|---|---|---|
| F | F | low probability | true negative |
| F | T | low probability | false positive |
| T | F | high probability (Aim?) | false negative |
| T | T | high probability | true positive |
- Special case: There is a medical emergency but no monitore alarm.
- Problem here: 12 cases - but not really the aim
- The aim is to detect false positive ones but not to be better than the monitor, right?
- Should be a criterion in the evaluation