ACR AI-LAB

A first-of-it's-kind data science toolkit designed by the American College of Radiology's Data Science Institute to Democratize AI by empowering radiologists to develop AI algorithms to support the clinical needs and expand the adoption of AI technology for radiologists, facilities, and the health care community around the country.

PROJECT
Product Design

ROLE
UI/UX Lead
Information Architecture
Front-End Development Assist
Motion Graphics Art Direction
Marketing Visual Design

TEAM
PO, PM, BA, Data Scientists, Engineers, Back-end Developers, Front-end Developers

Artboard.png

Challenges

  • Limited development of AI due to lack of computer science expertise in health care institutions to build and customize AI models for investigative purposes.

  • Lack of diverse array of clinical data being shared for algorithm training due to patient privacy concerns.

  • The manual process of writing codes.

  • Lack of awareness and understanding of AI within the radiology community.

Meeting.png

Personas

  • Radiologist

    Jenny is an interventional radiologist, loves the wide breadth of treatments that she can offer to her patients. She enjoys working with colleagues in other disciplines to provide treatments within any organ system and a very wide range of conditions.

    FEELINGS
    Lack of confidence due to fear of losing her job with the advancement of AI technology.

    Frustration due to lack of computer science expert knowledge.

    GOALS
    Adapt and learn AI.

    Ability to apply to daily practice without depending on scientists and developers.

    Contribute to the radiology community.

  • Medical Student

    Sugi is a medical student at American College of Radiology with a strong interest in Interventional and Diagnostic Radiology. She has some experience from her volunteer work and technologically savvy.

    FEELINGS
    Excited to see where technology will take the radiology industry.

    Frustrated with the current systems that are outdated.

    Frustrated by lack of resources on technologies around radiology.

    GOALS
    Understand and learn new technology.

    Practice and participate.

    Contribute to the radiology community.

  • Data Scientist / Developer

    Mohammed is a researcher and a developer at the radiology facility. He has been experimenting machine learning for some time now and tested building algorithms. His passion is to offer innovative solutions by combining his skills and AI technology.

    FEELINGS
    Concerned about the security of the patient data.

    Concerned about the lack of awareness and knowledge around machine learning amongst his peers and radiologists.

    Exhausted from the amount of coding to do.

    GOALS
    Improve machine learning with more diverse data from other facilities

    Educate and promote the benefits of AI

    Automate the manual workflow.

infastructure.png

Objectives


Develop a cloud version of the platform to introduce the concept, educate, and provide a simple step-by-step, hands-on toolkit that the participants can rapidly build, test, and customize AI models and experience the entire AI development life-cycle.



Promote the initiative during the ACR annual meeting conference by demonstration and contest participation.



Develop an on-premise version of the platform for use at their own facilities, using their own data, to meet their own clinical needs.

UIUX-AILAB-Whiteboard1.png

Information Architecture

The cloud version, although it was launched successfully, needed structural reexamination. So I took it back to the drawing board and audited content and structure.

 

Strategy

Created visual animated presentation to show the current state and strategy to solve the problem.

With client approval, new wires were created.

 

UI Design

UI-AI-Breast.png

Darkroom Annotation

Darkroom UI implemented for image-related functionalities to offer the radiologists a seamless and familiar experience as their facility darkroom tools.

 
UI-AI-Create.png

Create

Data drive multi-select user dropdowns that changed dynamically to suit the needs of AI configuration, due to the dynamic nature of pre-population, a multistep process was designed to improve performance.

UI-AI-Evaluate.png

Evaluate Card View

With only few models available, card view has been applied as a start and FDA approved models have special indication.

 
UI-AI-Evaluate Detail.png

Evaluate Detail

Second column to find dataset to evaluate and each evaluation has progress indicator and expandable view for overview.

UI-AI-Mobile.png

Mobile App

Progressive mobile app designed for cloud version.

Takeaways

This was a challenging project with an aggressive deadline and unpredictable expectations, but the constraints fostered creativity and strengthened the team dynamic. Above all, it was rewarding to be part of a technological evolution that brought significant awareness and a game-changing impact to the radiology field.

I expanded my design skills by working on complex data visualizations, darkroom imaging tools, and innovative AI model interfaces.

Collaborating closely with Data Scientists to turn concepts into reality reinforced the importance of context-centric design and highlighted how much more this product could improve with additional meaningful data.

If I could change one thing, I would incorporate a technical screening test for front-end contractors during the interview process to boost productivity. People are the greatest asset, and assembling the right team for the right job is key to a successful project.