HEALTHCARE + AI / JAN–APR 2022 / Siemens - HealthineersMaking AI-assisted lung detection UI fit clinical work.
Researching diagnostic pain points and integrating machine-learning outputs into a more understandable medical workflow.
DESIGN ARTIFACT / WORKFLOW MODEL01
Interview patients, doctors and engineers
02
Compare workflows and prototype ML/DL models
03
Integrate results and refine the diagnostic flow
01 / CONTEXTThe project at a glance
MY ROLEProject Intern — UX · User research, competitive analysis, workflow design, ML/DL prototyping
COLLABORATORS & TOOLSHealthcare professionals, engineers and project stakeholders
PYTHON · FIGMA · TENSORFLOW · VS CODE · MICROSOFT OFFICE
02 / CHALLENGEThe problem behind the interface
THE PROBLEMLung-cancer diagnosis involves complex, time-sensitive imaging workflows. Additional tools can create more work unless their outputs fit existing clinical decisions.
WHY IT MATTERSMedical users need interpretable results and clear handoffs, not another opaque detection layer. Model performance and clinician usability must be evaluated separately.
03 / PROCESSHow I approached it
01
RESEARCH
Used interviews and surveys to identify pain points, then compared Siemens workflows with Philips IntelliSpace and GE Thoracic Care Suite.
02
DESIGN
Developed initial ML and DL approaches to and a hybrid model to examine different experiences around detection scenarios.
03
INTEGRATE
Explored how outputs could be incorporated into diagnostic workflows and refined the concept using stakeholder feedback.
04 / DESIGN JUDGMENTThe decisions and their rationale
DECISION 01Design around clinical interpretation
WHAT I DIDOrganized model outputs as decision-support information within an existing diagnostic process rather than a stand-alone prediction screen.
WHY IT HELPEDAims to lower workflow friction and help clinicians interpret information in context.
DECISION 02Separate detection quality from UX
WHAT I DIDCompared ML, DL and hybrid approach experiences while treating interface clarity as a distinct design challenge.
WHY IT HELPEDPrevents a strong technical metric from being mistaken for clinical adoption or usability success.
05 / OUTCOMEOutcome, evidence and limitations
96%
Reported model accuracy while maintaining easy readability of information
3 model approaches
ML, DL and hybrid experiences described in project materials
SCALING
Expand user testing across different geo-political scenarios to better understanding user responses and behavior 06 / REFLECTIONWhat I would validate next
How effectively clinicians interpret AI-assisted findings within their diagnostic workflow. Usability testing could help assess clarity, trust, and decision-making, while identifying opportunities to reduce cognitive load.