In this episode, I talk with David Golan, co-founder and CTO of Viz.ai, about diagnosis of acute and emergent diseases. Viz.ai increases the speed of diagnosis and care for a variety of conditions to improve the lives of patients.

Highlights:

  • Increasing access to lifesaving treatments.
  • The importance of the full system, not just the machine learning component, in accelerating workflows.
  • Their clinical AI team includes med students, MDs, biomedical engineers, and neuropsychologists.
  • Bias can be created by a lower performance on a subset of the population in a way that is unknown to developers, users, and clinicians.
  • Careful monitoring of algorithms to identify subsets of data with poor performance.
  • Unbiased collection and stratification of data for FDA submission.
  • The importance of good annotation and monitoring infrastructure.
  • Relatively simple model architectures can take you a long way.

Quotes:

“So the vision of Viz.ai basically says, let’s turn this whole thing on its head from a serial process to a parallel process. Let’s connect all the hospitals to the cloud, have all the imaging, all the scans stream to the cloud in real time where a machine learning module can identify stroke. And now even more acute pathologies and alert all the relevant team, whether it’s the ED physician, the radiologist, the neurologist, the neurosurgeon, the anesthesiologist, the stroke coordinator, everybody. It could be 20 people. At the same time, they all get an alert on their phones. So whenever they are, they can get the data immediately. They get the scans, they have communication tools with them to, within a couple of minutes, communicate among themselves, circulate the information, make a decision in, inform everybody on the decision. Essentially, go, go, go.”

“Nowadays we have a clinical AI team. It’s actually the largest team, uh, at Viz.ai. It has med students, it has qualified MDs, it has people from biomechanical engineering backgrounds or neuropsychology or newer anatomy backgrounds. So all sorts of backgrounds. The people who understand the clinical side, understood enough of the AI process to work well with the AI engineers. And we form these very holistic teams for every new project.”

“I think the biggest concern is that a specific ML model might have lower performance or a different performance profile on a subset of the population in a way that is unknown to the developers, us, and to the users, the clinicians. If, for example, a certain age group or a certain ethnicity group do not get the same accuracy and therefore miss out on the opportunity to benefit from a device, at the very least, this should be known.”

“I would invest in a very good annotation and monitoring infrastructure from a very, very early point in time.”

“We have found that deep understanding of the data, good inspection of the data, false positives, false negatives, different training regimes of the same architectures with different data sets has a bigger impact on performance than fancier or bigger or deeper architecture. And this is somewhat counterintuitive and this would be my advice to leaders in startups as you are recruiting people. Make sure that you’re not recruiting those people who just want to implement the newest, fanciest, flashiest architecture that was just published because it’s cool, but those people who are willing to roll out the sleeves, get dirty with the data, even if the architecture at the end of the day is, is simple.”

Links:


Resources for Computer Vision Teams:

LinkedIn – Connect with Heather.
Computer Vision Insights Newsletter – A biweekly newsletter to help bring the latest machine learning and computer vision research to applications in people and planetary health.
Computer Vision Strategy Session – Not sure how to advance your computer vision project? Get unstuck with a clear set of next steps. Schedule a 1 hour strategy session now to advance your project.
Foundation Model Assessment – Foundation models are popping up everywhere – do you need one for your proprietary image dataset? Get a clear perspective on whether you can benefit from a domain-specific foundation model.