Project Overview
An NHS Trust cervical screening programme processing 84,000 cervical samples per year needed an AI-assisted pr...
Technology Stack
Compliance & Standards
The Challenge
An NHS Trust cervical screening programme processing 84,000 cervical samples per year needed an AI-assisted primary screening platform -- using AI to triage liquid-based cytology (LBC) samples before cytologist review. MHRA Class IIb SaMD (AI performing primary screening triage -- Class IIb risk classification, higher than Class IIa), UK Notified Body (BSI Assurance UK), IEC 62304 Class C, ISO 14971, DCB0129, NHS DTAC all 5 domains, NHS Cervical Screening Programme (NHSCSP) standards, NHS AI Lab 14 Standards, UK GDPR Article 9. Budget GBP160,000.
Our Approach
MHRA Class IIb SaMD
AI that performs primary screening (first-pass analysis before cytologist) is Class IIb (higher clinical risk than decision support Class IIa -- AI is performing the screening step).
Class IIb requirements above Class IIa
- 1UK Notified Body full Quality Management System audit (ISO 13485 QMS audit by BSI Assurance UK -- not just technical file review),
- 2IEC 62304 Class C (highest software class -- all functions with clinical safety implications require full traceability),
- 3clinical investigation (prospective clinical investigation -- not just validation study -- MHRA clinical investigation authorisation required),
- 4post-market clinical follow-up (PMCF -- ongoing clinical evidence collection post-launch).
Clinical investigation
ClickMasters supported MHRA CIV1 clinical investigation authorisation application -- 8-week MHRA review before investigation start.
LBC AI model
deep learning (ResNet-50 fine-tuned on NHSCSP LBC training data) performing primary screening of cervical cytology slides.
Clinical workflow integration
- 1digital slide scanner (Hamamatsu NanoZoomer -- digitises LBC slide at 40x magnification),
- 2AI primary screening (AI analyses digital slide -- confidence score per region, overall slide classification: normal/abnormal/inadequate),
- 3normal triage (AI high-confidence normal slides set aside -- still reviewed by cytologist but deprioritised),
- 4abnormal escalation (AI flags abnormal slides -- immediately reviewed by cytologist),
- 5inadequate classification (AI flags inadequate samples -- laboratory recall process).
NHSCSP sensitivity requirement
AI sensitivity for CIN2+ (pre-cancerous change Grade 2+) must be above 95% -- clinical validation required.
Training data
280,000 NHSCSP LBC slides (10-year dataset).
DCB0129 for Class IIb AI screening
higher clinical risk requires more extensive clinical safety case.
Primary hazard
AI misclassifying a CIN2+ sample as normal (false negative -- patient not recalled).
Mitigation
- 1AI cannot classify a slide as normal with 100% confidence -- all slides reviewed by cytologist (AI triage, not AI diagnosis),
- 2double-read protocol (AI abnormal + cytologist abnormal: immediate referral; AI normal + cytologist abnormal: immediate referral; AI abnormal + cytologist normal: second cytologist review -- discordance protocol),
- 3AI sensitivity monitoring (weekly -- AI sensitivity must stay above 95% CIN2+ -- automated alert if sensitivity drops below threshold),
- 4cytologist override (cytologist can override AI classification -- all overrides reviewed monthly by Lead Cytologist).
Clinical Safety Officer
Consultant Cytopathologist.
NHSCSP Programme Integration
NHSCSP (NHS Cervical Screening Programme): national programme run by NHS England.
Programme integration
- 1NCSP-is (Cervical Screening call-recall system -- GP patient lists, invitation management, result notification),
- 2HL7 v2.x result reporting (LBC result reported to GP via HL7 ADT + ORU messages),
- 3SNOMED CT coding (cervical cytology results coded in SNOMED CT -- Bethesda System terminology UK equivalent),
- 4national statistics (annual return to NHS England -- sample volumes, abnormality rates, CIN2+ detection rate).
PHE target
CIN2+ detection rate above national average -- AI primary screening expected to improve detection rate by 12-18%.
The Results
MHRA Class IIb UKCA obtained (BSI Assurance UK, ISO 13485 QMS audit passed).
Platform live at 40 weeks, GBP152,000. 84,000 samples/year.
AI sensitivity CIN2+: 97.2% (NHSCSP target: 95%).
AI specificity: 84.8%.
AI normal triage: 52.4% of slides classified normal (cytologist workload reduction).
Double-read triggers: 3.8% of slides.
Discordance review: all actioned within 24 hours.
NHS AI Lab all 14 standards documented.
DCB0129: zero clinical safety incidents.
UKCA Class IIb obtained.
AI sensitivity 97.2% vs 95% NHSCSP target.
Cytologist workload 52.4% triage reduction.
Discordance 3.8%.
AI Lab all 14.
DCB0129 zero.
The double-read protocol -- AI abnormal plus cytologist normal triggers second cytologist review -- was the DCB0129 design that made this clinically safe.
AI triage means the cytologist reviews every slide.
But the cytologist who reviews an AI-abnormal slide that they personally classify normal must have that classification reviewed by a colleague.
That discordance protocol adds one review per 26 slides and catches the cases where the AI was right and the cytologist was wrong. -- Lead Cytopathologist, NHS Trust (name withheld)
Project Details
Related Case Studies
View AllNHS-Connected Mental Health App β DTAC Approved
NHS-Connected Mental Health App β DTAC Approved. A UK mental health technology company needed to build a patient-facing digital therapy support app connected to NHS syst...
Workforce Management SaaS β UK Healthcare Staffing Agency
A UK healthcare staffing agency placing 2,800 NHS-registered nurses, doctors, and allied health professionals ...
NHS Outpatient Referral System β FHIR R4 Integrated
An NHS Foundation Trust needed to replace a 2007 paper-based outpatient referral system. GPs referred patients...
Ready to Transform Your Business?
Let's discuss how our technical expertise can help you achieve remarkable results.