HealthTech NHS AI Prostate Cancer Screening -- MHRA Class IIa

MHRA Class IIa for prostate AI

• 1UK Notified Body technical file review (BSI Assurance UK),
• 2IEC 62304 Class B documentation (software lifecycle -- requirements, design, implementation, testing, maintenance),
• 3ISO 14971 risk management (hazard identification, risk estimation, risk control),
• 4clinical evaluation (literature review + prospective clinical validation -- ClickMasters supported MHRA clinical evaluation report),
• 5UKCA marking (post-Notified Body review),
• 6MHRA DESH registration,
• 7Post-Market Surveillance Plan.

NICE ESF

NHS procurement increasingly requires NICE Evidence Standards Framework (Tier 3b for clinical decision support AI).

NICE ESF evidence

ClickMasters supported economic evaluation (cost per QA-adjusted cancer detected -- compared to MRI-only pathway).

Prostate MRI AI

• 1mpMRI (multi-parametric MRI -- T2-weighted, diffusion-weighted, dynamic contrast-enhanced sequences),
• 2PI-RADS scoring AI (Prostate Imaging Reporting and Data System -- AI assigns PI-RADS score 1-5 per lesion),
• 3lesion detection (AI detects and highlights suspicious lesions on T2 + DWI sequences),
• 4biopsy targeting (AI recommends biopsy needle trajectory for detected lesions),
• 5confidence score (per-lesion AI confidence -- high confidence PI-RADS 4-5 = recommend biopsy, low confidence = re-read).

Training data

42,000 mpMRI studies from 8 NHS Trusts (multi-centre -- UK population representativeness).

DCB0129

AI cannot independently classify as benign -- all studies reviewed by urologist.

NICE ESF Evidence Framework

NICE Evidence Standards Framework for Digital Health Technologies (NICE ESF): (1) Tier 1 (self-care with no clinical claims -- not applicable), (2) Tier 2 (inform or prompt clinical discussion), (3) Tier 3a (clinical decision support using own data -- not applicable), (4) Tier 3b (AI clinical decision support using algorithmically-generated outputs -- applicable to prostate AI).

Tier 3b evidence requirements

(a) clinical validation (randomised controlled trial or well-designed prospective cohort), (b) economic evaluation (cost-effectiveness -- cost per cancer detected, QALYs), (c) real-world data plan (ongoing monitoring post-deployment), (d) equalities evidence (AI performance across demographic groups).

ClickMasters NICE ESF Tier 3b

supported 3-year prospective observational study design for NICE submission.

Prostate MRI workflow integration

• 1DICOM receiver (mpMRI study arrives from MRI scanner -- DICOM C-STORE to AI service),
• 2AI inference (3D CNN model -- inference 45-90 seconds per study),
• 3PI-RADS annotation (AI creates DICOM Secondary Capture with highlighted lesions, PI-RADS score per lesion),
• 4radiologist review (PACS reporting workstation -- AI annotation displayed, radiologist assigns final PI-RADS),
• 5MDT reporting (Multi-Disciplinary Team meeting -- AI and radiologist scores combined for biopsy decision).

PACS integration

Sectra IDS7 via DICOM SR (Structured Report) for AI scores.

No AI autonomous decision

biopsy is always a clinical decision -- AI provides decision support.