Tablet Interface for Microtasking
Product Designer, User Researcher
Jan 2020 - August 2020
9 Months
Background
During my research sessions at PathAI, we had heard anedotes that users
Business goals
-
To decrease the turn around time for collecting annotations from pathologists.
-
To maintain the quality of annotations received for model training.
Design goals
-
To design tools for making region annotations quick and efficient.
-
Address some known pain-points with the annotation platform – bring consistency in navigation and layout.
Stakeholders
Machine Learning Engineers + Internal Pathologists + Annotator Community Manager
Thus, how might we make it easy for annotators to completely annotate a frame in a slide.
The older design
The annotation platform, when I took over the project consisted of –
-
Point tool – allows pathologists to place a point at the centre of each cell.
-
Pen/Pencil – allows pathologist to draw a boundary around a region of interest.
-
Magic wand – allows pathologists to select a cell's boundary.
Here are some of the high level pain-points (anecdotes from team members user studies) with the current experience of the annotation tool
-
Annotating smaller regions within a tissue is time consuming.
-
"It’s very difficult to annotate a donut like annotation with the current tools."
-
No provisions to edit existing annotations.
-
Lacks affordances of current labels causing user errors.
-
Annotating for long hours (with a mouse/trackpad) can cause cognitive, vision and muscular strain on palms/fingers.
Point Annotation
Research and Design
Empathy exercise + Surveys
Me, along with 20 internal employees at Path participated in a nuclei segmentation research experiment for another project. We made hundreds of annotations for multiple slides for ~3 hours. After this session I sent an experience survey to capture the experience of using the annotation platform in the context of making these nucleus boundary annotations.
I sent two surveys post our internal team used the annotation platform. The questions were open ended -
1. What was great?
2. What was not so great?
3. Suggest any ideas for improvement.
This exercise not only allowed the internal members to empathize with user problems, they had an avenue to voice their ideas to the product team. I also helped the team identify the gaps and opportunity areas in the product we have.
Brainstorming
I held a brainstorming session with the engineering squad along with ML stakeholders. As a pre session exercise I asked the members to use the annotation platform for a task and answer an experience survey. Then, I created this framework to help guide the brainstorming for the labelling tools-
-
Creating annotations
-
Editing/deleting annotations
-
Easily identify and remember which annotation is being made
Design Proposal
Reflection
This was a really impactful and interesting interaction design problem that came my way and here are some takeaways from this experience -
-
Involve engineering early in the process: For this work, I wanted to generate excitement in the engineering squad about the problem at hand. I felt like the workshop with the engineers went a long way in getting engineering inputs like - it is non trivial to edit annotation with the current framework and how bit map editing is different from the vector data that we store in the database today.
-
Making Quick prototypes: I think the hacky prototype that we put together on photoshop made it very easy for stakeholders to visualize the tool and it's interactions quickly and easily.
-
Building on top of each other's ideas: This work also inspired the engineering and ML squad to invest time in more ML powered tools to make making annotations even more quick and easy.
-
Thinking about the long term design and ripple effects on the system: Because the annotation platform is highly complex and caters to many different use case, the exercise to visualizing the vision for this work in all different scenarios helped all stakeholders see where we are headed in future.
Design responsibilities:
Research, Interaction Design, Visual Design
Tools used:
Design: Figma, Photoshop
Research: Zoom, Airtable, G Forms
