As the Staff User Experience Designer at General Electric Healthcare, based in San Ramon, California, I took on the challenge of leading the design aspect of the Artificial Intelligence Clinical Pathways (AICP) project. The core idea behind AICP was to leverage artificial intelligence to expedite the diagnosis and treatment of critical conditions. To offer a more detailed understanding of the project, a supplementary video is available that discusses the design process and its impact.

Process

To tackle the design challenges head-on, I delved into existing research conducted collaboratively by GE's research team and UCSF. This was coupled with firsthand observational research where I shadowed radiologists in their work environments, such as dark rooms dedicated to diagnostic readings in on-site hospital settings. These immersive experiences revealed the nuanced roles, needs, and challenges faced by radiologists, who specialize in diagnostic imaging, and clinicians, who are directly responsible for patient care.

Preceding the actual design work, I coordinated with the product management team to define the use cases that would be the basis for engineering UX specifications. I then crafted a well-structured UX roadmap. This roadmap, updated and shared with the entire team through a weekly newsletter, ensured a synchronized approach across different departments—be it researchers, engineers, or visual designers.

A significant obstacle was designing a severity prioritization system for AI findings that would not only suit both radiologists and clinicians but also mitigate the risk of false negatives. I resolved this by establishing three levels of severity displayed within the system. To validate this approach, I crafted a medium-fidelity clickable prototype that was put through several rounds of usability testing. This allowed us to make necessary iterations based on stakeholder and expert feedback.

The final design product was a culmination of iterative feedback and usability tests. It featured two separate worklists—one patient-centric for the care team and another that was study-centric for radiologists. Moreover, an AI panel was implemented for radiologists to view and validate AI findings, all while a separate view for the care team displayed both validated and unvalidated AI findings.

To guarantee a seamless translation of design into function, I provided the engineering team with detailed functional interaction design specifications. Additionally, I conducted quality assurance walkthroughs after each feature was implemented, ensuring that the final product met the high standards we aimed for.

Conlusion

The new design has been well-received, successfully catering to the unique needs of radiologists and clinicians alike. It not only helps radiologists validate AI findings efficiently but also assists the clinical care team in making prompt treatment decisions. The design is adaptable for future modalities and conditions. During its presentation at RSNA in 2018, the team received positive feedback; Breast specialist Dr. Joe Russo even mentioned that AICP would help him reduce unnecessary biopsies. I want to extend my gratitude to everyone who was involved in this enriching and impactful design process. Thank you.