AI Assist for Ultrasound

Project performed while I was a Senior Staff Engineer at Siemens Healthineers

AI Assist is an ultrasound application that allows users to automatically replace their color reference of interest and pulse wave doppler gate with the push of a button. This is a benefit for diagnostic cardiac sonographers who are doing 8-10 ultrasound exams a day. They are repetitive in nature and generally follow the same pattern as required by the hospital or clinic. This creates a lot of repetitive stress injuries resulting in users having to leave the job within three years of service.

By adding the one button push, we reduced the interactions significantly and sped up the exam times by 10 minutes.

Technical benefits

AI Assist leverages live view recognition and AI-enabled placement to automomously perform AI-enabled placement of color Doppler ROI and spectral Doppler gates in real-time. By analyzing the ultrasound image and recognizing the specific cardiac views among the 12 standard transthoracic ultrasound views, it accurately places the Doppler regions of interest on the appropriate anatomical hemodynamic targets.

Color box placement 98% accurate (no adjustment or minor adjustment)
PW Gate placement 95% accurate (no adjustment or minor adjustment)

Role

UX design and research
Managed stakeholder feedback
Strategy
Project manager
Usability studies

Team

Clinical stakeholders
Developer Engineers
Ultrasound Engineers
Product Management
System Architects
Risk

Process

Took over for another designer

Another designer started this work but I ended up taking it over. I met with the advanced development team to learn what had been done and where I needed to pick up and fill in the gaps.

Information architecture

There were some figma designs done, but I took those and blew them out to all of the states and scenarios that the feature would touch.

Created definition, gathered requirements and use cases

The figma designs were a concept, but there are dozens of rules and states that need to be documented and fleshed out. There are many technical rules that are required to be tracked. I worked with our feature lead and technical team to check be sure that I captured everything for engineering as well as FDA documentation. We iterated on this several times.

State changes

There were iterations on the feature from a technical perspective that required designs and specification tracking. I managed those details and provided the technical documentation updating.

System Configuration

We allow users to turn this on or off within the system dependent upon their preferences.

Prototyping

Routinely tested the prototype for defects and stayed in clear communication with the feature team.

Testing and shipping

I checked in on the feature as needed to track its progress, test as necessary and managed spec changes. Feature was shipped and is on the system today.

Challenges

Taking over for the original designer was hard. First, he had moved on before arrived so I couldn’t get his perspective on the project. Second, I had to get familiar with an entirely new team of people and untangle his documentation. He did a good job, but it was hard to understand.
It was hard to test the feature as it is transducer dependent. We didn’t always have access to transducers.
I was still new and the learning curve was steep.

Learnings

Learned how to work with a really technical team of scientists — it was amazing.
All users have their own protocol. We had to build this to one protocol. In the future we hope to add more functionality.

My deliverables:

Design documentation
Figma prototypes
Information architecture
Definition delivery
Pixel perfect developer ready designs in Figma
Localization documents