Project Overview
NHS Trust radiology processing 280,000 imaging studies per year needed AI-assisted triage prioritising urgent ...
Technology Stack
Compliance & Standards
The Challenge
NHS Trust radiology processing 280,000 imaging studies per year needed AI-assisted triage prioritising urgent findings (pneumothorax, intracranial haemorrhage, PE). MHRA Class IIa SaMD, IEC 62304 Class B, DCB0129, NHS DTAC all 5 domains, UKCA marking (BSI), NHS AI Lab 14 Standards, UK GDPR Article 9. Budget GBP130,000.
Our Approach
Class IIa requirements
- 1UK Notified Body (BSI Assurance UK) technical file review,
- 2IEC 62304 Class B documentation,
- 3ISO 14971 risk management,
- 4clinical evaluation (3-month prospective validation -- 500 cases, sensitivity and specificity per pathology),
- 5UKCA marking,
- 6MHRA DESH registration,
- 7Post-Market Surveillance Plan.
Key standards for radiology AI
Standard 1 (AI triage prioritises worklist order -- does not diagnose independently), Standard 2 (training data demographics documented), Standard 3 (which pathologies AI detects explicitly documented), Standard 8 (AI confidence score and highlighted image region shown -- not binary flag), Standard 14 (quarterly performance review -- sensitivity/specificity tracked in production).
Demographic bias
sensitivity across age groups, sex, BMI documented.
Primary hazard
AI de-prioritising urgent finding.
Design
- 1AI is advisory -- radiologist retains clinical responsibility,
- 2AI cannot move a study below standard priority (can elevate to urgent, never demote),
- 3all studies reviewed regardless of AI score,
- 4urgent threshold conservatively set (sensitivity >98% for pneumothorax/intracranial haemorrhage),
- 5radiologist override recorded for continuous learning.
Integration
- 1DICOM C-STORE listener (studies from CT/MRI/X-ray),
- 2AI inference (ResNet-50 fine-tuned -- <10 seconds per study),
- 3DICOM SR (AI score plus highlighted region attached to study),
- 4worklist re-prioritisation (HL7 ORM update to RIS),
- 5radiologist interface (AI overlay in Sectra IDS7 -- shown while reporting, not before, to avoid anchoring bias).
The Results
MHRA Class IIa UKCA obtained (BSI).
Platform live at 32 weeks, GBP124,000 under budget.
AI sensitivity: 97.8% pneumothorax, 98.4% intracranial haemorrhage, 96.2% PE.
Urgent report time: 84 to 28 minutes.
Override rate: 4.2% (retraining dataset).
NHS AI Lab all 14 documented.
DCB0129 zero incidents.
DTAC all 5 domains first-submission success.
“UKCA obtained. Sensitivity 97.8% pneumothorax, 98.4% intracranial. Urgent report 84 to 28 minutes. Override 4.2%. AI Lab all 14. DCB0129 zero. DTAC all 5 first submission. The design that matters: AI can only elevate urgency, never reduce it. Advisory not autonomous. That is what got DCB0129 sign-off. First AI radiology system the Clinical Safety Officer had seen that understood the difference between decision support and decision making." -- Clinical Director of Radiology, NHS Trust (name withheld)”
Project Details
Related Case Studies
View AllHealthTech NHS Digital Pharmacy Dispensing Platform
UK community pharmacy chain (280 dispensing sites, 8.4M items/year) needed to modernise dispensing management....
HealthTech NHS AI Cervical Screening Platform -- MHRA Class IIb
An NHS Trust cervical screening programme processing 84,000 cervical samples per year needed an AI-assisted pr...
HealthTech NHS IAPT Digital Mental Health Platform
NHS IAPT service (28,000 referrals/year, 840 therapists, 12 services across 3 trusts) needed to modernise pati...
HealthTech NHS AI Prostate Cancer Screening -- MHRA Class IIa
An NHS Trust urology department with 28,000 prostate biopsy results per year needed an AI-assisted triage syst...
Ready to Transform Your Business?
Let's discuss how our technical expertise can help you achieve remarkable results.