-
Notifications
You must be signed in to change notification settings - Fork 1
MVP
Nel Swanepoel edited this page Nov 18, 2022
·
5 revisions
The PIXL
MVP forms part of The Tracer Bullet (TTB) MVP for FlowEHR, UCLH's cloud research & innovation infrastructure initiative.
Delivering the PIXL
MVP requires that
- We pre-specify the cohort of patients who received chest X-rays (estimated ~300k images)
- We intelligently query the PACS/VNA for the chest X-rays from this cohort without causing operational systems to fall over
- We automatically de-identify DICOM elements with a simple whitelisting approach and removal of PII overlays
- We automatically push DICOM instances to a DICOM node in Azure via DICOMweb
- We extract EHR and free-text radiology reports for the specified cohort
- We de-identify free-text radiology reports with Presidio
- We de-identify PII EHR attributes using a blacklisting approach
- We link de-identified data securely
- We automatically push radiology reports & EHR data into Delta Lake on Azure
- We automatically ingest data from Delta Lake into the Feathr feature store
- We provide controlled access to the DICOM node and the Feathr feature store to an Azure TRE workspace
- We offer useful written guidance to the research team
- We provide workable policies and SOPs for managing image extraction via the PIXL pipeline
- Real-time data ingestion
- DICOM pixel data de-identification
- Identifiable patient data in Azure
- Perfection
mid-November 2022
mid December 2022