An AI-Driven Clinical Decision Support System for Lung Disease Physicians
Leading ongoing UX research and foundational AI-driven design within a complex multi-partner clinical environment
Role
UX Research
Design lead
Stage / Scope
Early UX direction & evaluable concept definition
Methods / Tools
Requirements Analysis
Interaction & workflow structuring
Interface Design
Cliant / Framework
EU-funded clinical AI consortium / Reichman University
Background
Project Motivation
Joining this EU-funded clinical AI project meant entering a space where much of the early work focused on understanding physicians’ attitudes toward AI rather than shaping a product. Valuable psychological insights existed, but they hadn’t yet been translated into workflows or an initial user experience.
My first responsibility was to build that bridge: define an early direction for the system and create initial interface concepts that physicians could meaningfully respond to. Working within a multi-partner environment that combined clinical, academic, and industry perspectives, I focused on turning existing research into a practical foundation for the product while leaving room for ongoing discovery and refinement.
From a UX perspective, the challenge was to support physicians’ decision-making under time pressure, fragmented data, and limited transparency in AI reasoning.
Background
Project Motivation
Joining this EU-funded clinical AI project meant entering a space where much of the early work focused on understanding physicians’ attitudes toward AI rather than shaping a product. Valuable psychological insights existed, but they hadn’t yet been translated into workflows or an initial user experience.
My first responsibility was to build that bridge: define an early direction for the system and create initial interface concepts that physicians could meaningfully respond to. Working within a multi-partner environment that combined clinical, academic, and industry perspectives, I focused on turning existing research into a practical foundation for the product while leaving room for ongoing discovery and refinement.
From a UX perspective, the challenge was to support physicians’ decision-making under time pressure, fragmented data, and limited transparency in AI reasoning.
Approach & Process
A non-linear, research-informed and evaluation-driven workflow
This project evolved through overlapping phases shaped by existing research, tight timelines, and limited access to physicians. My focus was to create structure within these conditions while keeping the process flexible and responsive to new insights.
Approach & Process
A non-linear, research-informed and evaluation-driven workflow
This project evolved through overlapping phases shaped by existing research, tight timelines, and limited access to physicians. My focus was to create structure within these conditions while keeping the process flexible and responsive to new insights.
Requirements Analysis
01
Extracted relevant insights from existing psychological research to define initial product needs and trust-related requirements.
Concept exploration (AI-assisted)
02
Used AI tools (Base44, Lovable, LLMs) to rapidly explore structural and interaction directions, helping surface early patterns worth developing.
Interaction & workflow structuring
03
Shaped the system’s core logic; how information appears, unfolds, and supports different physician decision-making styles.
Interface Design
04
Created initial interface concepts in Figma to make the system tangible and ready for physician feedback.
User evaluations (remotely)
04
Reviewed early screens with physicians in remote sessions to validate direction, surface friction points, and guide iteration.
Challenges & Constraints
Defining the core issues shaping early design decisions
Within this context, my role was to build on existing psychological insights while extending them through UX research, translating both into interaction flows and interface concepts physicians could meaningfully engage with. Although prior work examined attitudes toward AI, it had not yet been expressed as workflows or interface structures.
At this stage, the work progressed from research insights to an initial design foundation, with interface concepts defined to enable evaluation and discussion with physicians and consortium partners.
Supporting Different Decision-Making Styles
How might we support different clinical decision styles without overwhelming or underserving any physician?
“Some cases only need a quick summary, but others require digging deeper into the details before I can decide.”
User Pain Point
Physicians differ in how much information they need before trusting an AI-generated suggestion. Some prefer a concise overview, while others require deeper evidence before making a decision. A single pathway risks overwhelming one group or underserving the other.
UX Solution
I structured the experience around two complementary modes. Quick Mode presents a high-level view for fast assessment, while Deep Analysis provides a detailed breakdown of the AI’s reasoning, evidence, and supporting data. This dual-mode structure ensures both decision-making styles are fully supported without compromising clarity.
Maintaining Physician Autonomy
How might we provide AI guidance while keeping clinical judgment clearly in the physician’s hands?
“I believe AI can be helpful, but it shouldn’t make the decision for me; the final judgment has to stay mine.
User Pain Point
Physicians are wary of AI systems that appear authoritative or prescriptive. Since they do not directly input data into the system, the interface must make clear that final judgment remains with them and that the AI is an assistive tool, not a decision-maker.
UX Solution
To support autonomy, the system presents multiple treatment suggestions rather than a single “correct” path, each labeled with a “High / Medium / Low Match” indicator. This reinforces that the clinician—not the AI—selects the appropriate treatment. When the system detects outdated or insufficient data, it flags this limitation and recommends additional testing, signaling uncertainty rather than authority. UI phrasing such as “AI Suggested Diagnosis” and “AI Suggested Treatment” further underscores that the system offers support, not directives.
Building Trust Through Explainability
How might we help physicians validate AI reasoning without increasing cognitive load?
“If AI suggests something, I need to understand what it’s based on before I can trust it.”
User Pain Point
Physicians cannot rely on AI-generated recommendations without understanding the reasoning behind them. A suggestion without justification is clinically unusable and undermines trust, especially in complex decision pathways.
UX Solution
I designed a structured “AI Reasoning” section within Deep Analysis that surfaces contributing factors, interpretation layers, and evidence the AI used to reach a suggestion. In Quick Mode, contextual panels such as “Targeted Therapy” provide a lighter-weight rationale. Together, these elements give physicians the transparency needed to validate the system’s output without overwhelming them.
Ethical Transparency and Appropriate Reliance
How might we make AI transparency clear enough for clinicians to rely on?
“How can I feel confident in a result, or how well it fits my specific patient, if I can’t tell whether the underlying data or evidence is solid?”
User Pain Point
Physicians need visibility into how confident the AI is, how well the patient aligns with specific models or treatments, and where the AI’s information originates. Without clear transparency cues, appropriate reliance becomes impossible and clinical risk increases.
UX Solution
I incorporated transparency signals throughout the interface, including an “AI Confidence” bar to communicate uncertainty, a “Patient Matching” score showing relevance to specific treatment models, and links to external clinical resources to enable traceability. These cues promote responsible use and help physicians understand when the system’s output is strong, uncertain, or requires further validation.
Early Evaluation
Evaluating early concepts with lung disease physicians
Early interface concepts were reviewed remotely with lung disease physicians to assess clarity, clinical relevance, and alignment with real decision-making needs. The goal was not formal usability testing, but early directional feedback to validate the approach and guide ongoing refinement.
What was evaluated
-
Overall structure and information flow
-
Clarity of reasoning and evidence presentation
-
Suitability of Quick vs. Deep modes for different decision styles
-
Communication of uncertainty and trust cues
-
Alignment with clinical autonomy and decision ownership
What we learned
-
Physicians appreciated having both high-level summaries and deeper evidence pathways
-
Reasoning structures helped clinicians understand and validate suggestions
-
Autonomy-supportive content and tone resonated strongly
-
Some phrasing and information density areas require refinement
-
Feedback confirmed the direction as a strong foundation for ongoing UX development
These early evaluations validated the core design direction and informed the next stage of UX research, where workflow-level testing and iterative refinement will play a central role.
Key Contributions
Establishing the UX groundwork
My work translated early research into a usable UX foundation that physicians and partners could meaningfully engage with.
Created the project’s first evaluable UX direction
Provided a structured foundation that replaced ambiguity and gave the consortium a clear starting point for design and discussion.
Facilitated early physician feedback
Created evaluable interface concepts that allowed the consortium to gather meaningful clinical feedback much earlier than planned.
Helped bridge research and product thinking
Translated psychological insights into actionable product considerations, ensuring the project stayed clinically relevant and grounded.
Takeaways and What next
A Work In Progress
This project has highlighted how deeply clinical decision-making varies, and how design must flex to support both fast, high-level orientation and deeper evidence-driven reasoning. Working at the intersection of psychological research, clinical expertise, and industry expectations reinforced the importance of translating abstract insights into structures clinicians can immediately evaluate. It also emphasized the value of transparency, both in how the system reasons and in the reliability and completeness of its underlying data.
Moving forward, the next phase focuses on conducting more structured evaluations with physicians to refine workflows, phrasing, and information density, and to test how the dual-mode experience supports real diagnostic practices. As technical integration across partners progresses, the UX foundation will continue to evolve through iterative feedback, deeper usability testing, and closer alignment with clinical workflows and model capabilities.