Siemens Healthineers
Clinical UX / Workflow design
Clinical AI decision-support experience design
The problemModel output needed to fit clinicians’ diagnostic context, rather than arrive as an isolated results screen.
My contributionI mapped AI-assisted lung-detection output into clinical decision-support flows, informed by interviews and workflow benchmarking.
Clinical workflow integration
3
ML, DL and hybrid approaches mapped to clinical UX.
Clinical context
Interpret needs
Readable output
Integrate decision support
Concept review
Separate validation needs
- My role
- Project Intern — UX
- Team
- Healthcare professionals, engineers and stakeholders
- Timeline
- Jan–Apr 2022
- Tools
- Research interviews, Competitive analysis, ML/DL prototype exploration
Clinical UX internship: research and prototype exploration.
Process
The work behind the decisions.
- 01
Understand clinical work
Patient, clinician and engineering interviews and survey input framed workflow needs. Siemens workflows were benchmarked against Philips IntelliSpace and GE Thoracic Care Suite.
- 02
Map output to decisions
Explored ML, DL and hybrid approaches, then mapped readable model output into diagnostic decision-support flows.
- 03
Refine with stakeholders
Stakeholder feedback guided the concept; model evaluation and clinician experience remained distinct validation needs.
Key decisions
Design decisions that shaped the experience.
Keep output in clinical context
Place readable model output within the existing diagnostic workflow, rather than on a disconnected results-only screen.
Trade-off
Contextual integration needs workflow-specific information hierarchy, not simply more output.
Separate model and experience validation
Treat classification performance as supporting technical evidence, while identifying clinician usability as a separate evaluation need.
Trade-off
A strong model result does not establish diagnostic safety or clinician task success.
Design value & supporting evidence
Clinical AI decision-support experience design
Clinical workflow concepts connected readable AI output to decision-support interactions, alongside a reported prototype model evaluation.
- Decision-support approaches compared
- 3
- ML, DL and hybrid models assessed for readable clinical output.
- Prototype model accuracy
- 96%
- Technical model evaluation, not interface performance.
Reflection
The interface must help clinicians interpret output in context. A next iteration would test comprehension and workflow fit with clinicians, separately from model accuracy.
Next case study
Accessible care & payment journeys
I refined navigation, accessible forms and payment feedback for older and diverse users.
View case study