Bart Steverink & Laurens Schinkelshoek
summary
Healtphlus.ai builds PERISCOPE©, a clinical decision support that predicts post‑surgical infection probability at 7 and 30 days using hospital EHR structured data. The pipeline uses hospital‑specific ETL into a uniform feature matrix, tfold cross‑validation with a temporal 80/20 split, and reports an incident rate around 10% incidence with good AU and calibrated probabilities. Explainability uses SHAP explainability (top‑10 predictors shown but not prescriptive). Certification (MDR/CE) and per‑site mapping and monitoring are required; models do not auto‑learn in production.
AI-written summary · about
chapters (21)
- 0:00Healthcare AI challenges
- 1:07Company background
- 2:52Surgical site infections
- 4:21Medical decision support
- 5:46Data and certification challenges
- 7:30Development strategy
- 8:12Ground truth labeling
- 11:04Data standardization
- 13:08Model training approach
- 14:36Model performance results
- 16:03Cross-hospital transferability
- 17:27Patient outcomes impact
- 18:30Expansion plans
- 19:31Explainability and trust
- 20:38Data privacy compliance
- 21:47CE certification plans
- 22:31Adoption challenges
- 24:39Unstructured clinical data
- 25:24Regulatory hurdles
- 26:50Model stability
- 28:26Q&A
related talks
32mSHAP and Beyond
18mPipelines for Clinical AI: Real-Time Sepsis Detection
24mDeveloping Innovative Systems Through Responsible AI In Healthcare
11mEnsuring Reliability in Healthcare Data Engineering
25mLifesaving AI and JavaScript
33mData Science in Healthcare: Beyond the Hype
22mAI or Not AI? That's the Answer!
11mDesigning Governed AI Pipelines for Healthcare Impact