Intro & scope
37 use cases across four patterns: IMAGING (radiology, pathology, lab, endoscopy, signals read as diagnostics), CLINICAL (documentation, decision support, reports, in-basket), MONITOR (remote/continuous monitoring, early warning, population and epidemic analytics) and OPS (coding, interoperability, trials, pharmacovigilance, regulatory/GMP). Scope boundaries: HLTH-003 (physician ambient scribe) vs HLTH-025 (nursing/long-term-care voice documentation); HLTH-004 (point-of-care decision support) vs HLTH-018 (chart summarisation/search); HLTH-007 (continuous monitoring) vs HLTH-012 (in-hospital deterioration) vs HLTH-020 (population risk); HLTH-001/023/024 (read, reconstruct, acquire images); HLTH-010 (post-market safety) vs HLTH-029 (GMP quality). Most clinical-use cases are medical devices (MDR) and high-risk under the EU AI Act (Art. 6(1) with Annex I); administrative cases are minimal risk; "depends" means the intended purpose decides. Health data is special-category under GDPR Art. 9, so a DPIA is required throughout.
Overview
| ID | Title | Pattern | Maturity | Path | HW tier | AI Act risk |
|---|---|---|---|---|---|---|
| HLTH-001 | Radiology image analysis (X-ray, CT, MRI) | HLTH.IMAGING | mainstream | A | HW3 | high |
| HLTH-002 | Digital pathology & slide analysis | HLTH.IMAGING | early-adopter | A | HW3 | high |
| HLTH-003 | Ambient clinical documentation (AI scribe) | HLTH.CLINICAL | early-adopter | A | HW3 | depends |
| HLTH-004 | Clinical decision support | HLTH.CLINICAL | early-adopter | A | HW3 | high |
| HLTH-005 | Medical coding & billing automation | HLTH.OPS | early-adopter | A | HW3 | minimal |
| HLTH-006 | Dermatology & ophthalmology screening | HLTH.IMAGING | early-adopter | A | HW2 | high |
| HLTH-007 | Remote patient monitoring & wearable analytics | HLTH.MONITOR | early-adopter | A | HW1 | high |
| HLTH-008 | Hospital capacity: bed, OR & staff planning | HLTH.OPS | early-adopter | A | HW1 | depends |
| HLTH-009 | Clinical trial patient matching & trial optimisation | HLTH.OPS | early-adopter | A | HW3 | depends |
| HLTH-010 | Pharmacovigilance & adverse event detection | HLTH.OPS | early-adopter | A | HW3 | minimal |
| HLTH-011 | Discharge letters & medical report generation | HLTH.CLINICAL | early-adopter | A | HW3 | depends |
| HLTH-012 | Sepsis & clinical deterioration early warning | HLTH.MONITOR | early-adopter | A | HW1 | high |
| HLTH-013 | ECG & cardiology signal analysis | HLTH.IMAGING | mainstream | A | HW1 | high |
| HLTH-014 | Dental imaging analysis | HLTH.IMAGING | mainstream | A | HW2 | high |
| HLTH-015 | Fall detection & ambient assisted living | HLTH.MONITOR | early-adopter | A | HW1 | depends |
| HLTH-016 | Veterinary diagnostics | HLTH.IMAGING | early-adopter | A | HW2 | minimal |
| HLTH-017 | Regulatory & medical writing for pharma and medtech submissions (CSR, CTD/eCTD, MDR technical documentation) | HLTH.OPS | early-adopter | A | HW3 | minimal |
| HLTH-018 | Patient chart summarisation & longitudinal record search for clinicians | HLTH.CLINICAL | early-adopter | A | HW3 | depends |
| HLTH-019 | Vocal biomarkers: voice- and cough-based health screening | HLTH.MONITOR | emerging | A | HW1 | high |
| HLTH-020 | Population health risk stratification & readmission prediction | HLTH.MONITOR | early-adopter | B | HW1 | depends |
| HLTH-021 | Radiotherapy auto-contouring & AI treatment planning | HLTH.IMAGING | mainstream | A | HW3 | high |
| HLTH-022 | Real-time AI for endoscopy & surgical video | HLTH.IMAGING | early-adopter | A | HW2 | high |
| HLTH-023 | AI image reconstruction & accelerated scan acquisition (MRI, CT, PET) | HLTH.IMAGING | mainstream | A | HW3 | high |
| HLTH-024 | AI-guided ultrasound acquisition & point-of-care ultrasound interpretation | HLTH.IMAGING | early-adopter | A | HW1 | high |
| HLTH-025 | Voice-based nursing & long-term care documentation (Pflegedokumentation) | HLTH.CLINICAL | early-adopter | A | HW2 | depends |
| HLTH-026 | AI-assisted laboratory diagnostics (blood-cell morphology, microbiology plate reading, urine sediment) | HLTH.IMAGING | mainstream | A | HW1 | high |
| HLTH-027 | Chronic wound assessment & documentation from smartphone photos | HLTH.IMAGING | early-adopter | A | HW1 | high |
| HLTH-028 | Clinical terminology mapping & health-data interoperability (SNOMED CT, LOINC, FHIR, OMOP) | HLTH.OPS | early-adopter | B | HW2 | minimal |
| HLTH-029 | GMP batch record review & deviation/CAPA assistant | HLTH.OPS | early-adopter | A | HW3 | minimal |
| HLTH-030 | Patient message reply drafting & triage for clinicians (in-basket assistant) | HLTH.CLINICAL | early-adopter | A | HW3 | depends |
| HLTH-031 | Epidemic intelligence & infectious-disease outbreak forecasting (incl. wastewater surveillance) | HLTH.MONITOR | early-adopter | B | HW2 | minimal |
| HLTH-032 | Neural & myoelectric interface decoding (BCIs, intelligent prostheses) | HLTH.MONITOR | emerging | A | HW1 | high |
| HLTH-033 | Prognostic digital twins for clinical trials (covariate adjustment & smaller control arms) | HLTH.OPS | early-adopter | A | HW1 | minimal |
| HLTH-034 | Automated insulin delivery & glucose prediction (closed-loop artificial pancreas) | HLTH.MONITOR | mainstream | A | HW1 | high |
| HLTH-035 | EEG & sleep-study (polysomnography) analysis | HLTH.IMAGING | mainstream | A | HW2 | high |
| HLTH-036 | Embryo assessment & selection in IVF | HLTH.IMAGING | early-adopter | A | HW2 | high |
| HLTH-037 | AI-assisted robotic surgery (autonomous camera control & surgical sub-task automation) | HLTH.CLINICAL | emerging | A | HW3 | high |
Shared stack
Common to all tiers: EU-hosted or on-premise only, FHIR R4 / DICOM / HL7 as integration layer (HLTH-028 terminology service shared), pseudonymisation gateway, audit log and human-in-the-loop sign-off, MDR/AI Act technical documentation and post-market monitoring.
HW1 (CPU / edge, no GPU): signals, tabular, small vision. Used by HLTH-007, 012, 013, 015, 019, 020, 024, 026, 027, 032, 033, 034. - Runtime: ONNX Runtime / OpenVINO or TFLite on gateway or device; streaming via MQTT + Redpanda/Kafka. - Models: gradient boosting (XGBoost/LightGBM), 1D-CNN/transformers for ECG/EEG, small CNNs for images. - Storage: PostgreSQL + TimescaleDB; MLflow for registry and drift monitoring. - Shared with: PRED, ANOM, IOT families.
HW2 (single 24-48 GB GPU): vision and small LLM/ASR. Used by HLTH-006, 014, 016, 022, 025, 028, 031, 035, 036. - Serving: NVIDIA Triton or MONAI Deploy for imaging; vLLM with 7-14B model for text; faster-whisper for ASR. - Retrieval: pgvector or Qdrant plus a multilingual embedding model (e.g. bge-m3) and reranker. - Orchestration: Prefect/Airflow; Keycloak SSO. - Shared with: VIS, SPEECH, DOC families.
HW3 (multi-GPU server, 80 GB+ class): foundation-model-scale imaging and clinical LLMs. Used by HLTH-001, 002, 003, 004, 005, 009, 010, 011, 017, 018, 021, 023, 029, 030, 037. - Serving: vLLM/SGLang with 70B-class open-weight model (Llama/Qwen/Mistral) behind a guardrail layer (NeMo Guardrails or Llama Guard); MONAI/nnU-Net and pathology foundation models on a separate GPU pool. - Data: whole-slide and DICOM store (Orthanc/dcm4chee) with PACS integration; Qdrant/OpenSearch hybrid search over the record; Docling/OCR parsing. - Platform: Kubernetes, Vault, central prompt/eval harness with clinical golden sets. - Shared with: KNOW, DOC, LANG, SPEECH families (one clinical LLM gateway serves 003, 004, 011, 018, 030 and 005, 017).