MVP

The PIXL MVP forms part of The Tracer Bullet (TTB) MVP for FlowEHR, UCLH's cloud research & innovation infrastructure initiative.

Deliverables

Delivering the PIXL MVP requires that

0. We pre-specify the cohort of patients who received chest X-rays (estimated ~300k images)
1. We intelligently query the PACS/VNA for the chest X-rays from this cohort without causing operational systems to fall over
2. We automatically de-identify DICOM elements with a simple whitelisting approach and removal of PII overlays
3. We automatically push DICOM instances to a DICOM node in Azure via DICOMweb
4. We extract EHR and free-text radiology reports for the specified cohort
5. We de-identify free-text radiology reports with Presidio
6. We de-identify PII EHR attributes using a blacklisting approach
7. We link de-identified data securely
8. We automatically push radiology reports & EHR data into Delta Lake on Azure
9. We automatically ingest data from Delta Lake into the Feathr feature store
10. We provide controlled access to the DICOM node and the Feathr feature store to an Azure TRE workspace
11. We offer useful written guidance to the research team
12. We provide workable policies and SOPs for managing image extraction via the PIXL pipeline

Out of Scope

• Real-time data ingestion
• DICOM pixel data de-identification
• Identifiable patient data in Azure
• Perfection

Target Delivery Date

mid-November 2022
mid December 2022