Medical Imaging Ai
DICOM, PACS, AI-assisted diagnosis, image segmentation, and radiology workflows
Trigger
Activate this skill when the user mentions:
- "imaging", "radiology" workflows or issues
- Building, fixing, or optimizing medical imaging ai
- Questions about "imaging" best practices
Core Concepts
- Model selection and evaluation
- Feature engineering and data prep
- Training methodology
- Deployment and serving patterns
- Monitoring and drift detection
Step-by-Step Workflow
- Frame — Define problem, success metric, baseline
- Expected: Clear problem statement
- Explore — EDA, feature analysis
- Expected: Understanding of data relationships
- Build — Train models, track experiments
- Expected: Logged reproducible experiments
- Evaluate — Test on holdout, check bias
- Expected: Evaluation report with confidence
- Deploy — Serve with monitoring
- Expected: Production model with drift detection
Tools & Technologies
- Experiment tracking
- Model registry
- Feature store
- Model serving
Best Practices
- Document decisions and rationale (ADRs, design docs, runbooks)
- Version everything - code, configs, data, and documentation
- Test incrementally; never claim passing without verification
- Respect domain-specific regulations and ethical standards
- Measure outcomes with meaningful metrics
Common Pitfalls
- Data leakage → Overly optimistic metrics → Strict temporal splits
- No monitoring → Silent degradation → Monitor prediction distribution
Tags
"imaging", "radiology"
LoopyLuci/Skills - 2026-09-24