By Moichor, Principal Computer Vision Scientist, Matthew Guay
In the past decade, artificial intelligence (AI) has exploded out of academia to form the backbone of a new generation of commercial enterprise. Led by advances in neural network design, there are enormous opportunities for applying computer vision to power more human-like understanding of image data. Now, this technology is coming to veterinary pathology.
In a recent article we addressed the question: Why is this happening now? In this article, I’ll address the question: What’s so exciting about veterinary pathology for an AI scientist?
In a sentence, the field is having a “Goldilocks moment” for AI. It's not too early and not too late to transform cool AI research and development (R&D) into diagnostics tools that have real-world impact today.
To explain what I mean, I’ll share my own story of how I found my way from the National Institutes of Health (NIH) to a Silicon Valley startup. February 10th marks my nine-month anniversary of joining Moichor as a principal computer vision scientist.
Since May 2021, I’ve worked with Chief Technology Officer Thanh Le and Director of Pathology Dr. Kyle Webb, to research and develop smarter pathology AI tools while getting a crash course in veterinary pathology.
Signing on with an early-stage startup can be a risky bet, particularly after starting my career in a government lab, but I'm glad I made it. My work is immensely rewarding to me because I’m able to tackle challenging computer vision projects, play a part in making life easier for veterinarians and pets, and help shape a burgeoning tech business.
In February 2021 I was preparing to re-enter the job market after a five-year stint at the NIH, where I had been working on computer vision problems in biological electron microscopy. I had something specific in mind from the moment I started my job search.
I was picky in what I was looking for. I wanted to find R&D work on interesting computer vision problems that would have a significant impact on an important biomedical application.
In a field as fast-moving as AI, this can be tricky to do — today’s research challenge is tomorrow’s solved commodity software. Moreover, the problems AI researchers find interesting don’t always line up with the solutions practitioners need in the field.
Within biomedicine (and beyond), there are still many low-hanging opportunities for making practitioners’ lives better with information technology that don’t require any fancy AI. This is important work, but I was looking for a problem where AI could be central to the solution.
On the other side of the spectrum from the low hanging fruit, there are applications we can agree need an AI solution but are more challenging than anticipated — see self-driving cars, which have famously been “about a year away”, since at least 2016.
We will likely see solutions eventually, but producing AI applications that can be successful and useful today, right now, requires a careful alignment of human needs, data availability, and computational and organizational processes, which takes us to veterinary pathology.
Like Goldilocks, I was looking for a mission that was just right — not too small, or too late in the R&D cycle, not too complex or too early, but relevant and achievable now. As I assessed which fields might furnish this ideal opportunity, applied AI for pathology stood out.
What makes pathology stand out? There’s a lot of valuable work to be done in this field that is perfectly suited to AI. Modern microscopes can generate huge multi-gigapixel images. So, creating tools to automate cellular microscopy imaging offers new capabilities that are not feasible for humans to do routinely.
The breadth of species encountered in veterinary pathology is a unique feature of the field, not seen in human healthcare, that is also attractive to an applied AI scientist. The biggest questions in AI center on generalization — how can we create systems that are smart enough to handle the species-to-species variability that presents in veterinary testing? It's an added challenge, but a challenge ripe for innovation.
As I surveyed my prospects nine months ago, within the standout AI pathology space, Moichor stood out again. A good applied AI job prospect strikes a balance — you want a company where some AI applications already show commercial viability, but the need for further applications guarantees a steady stream of R&D projects for years to come.
Moichor was already finding success delivering automated visual CBCs for exotic species pathology, and after meeting with the founders I recognized that we all saw the same opportunities for modernizing and advancing veterinary diagnostics. Check out this earlier article for more about our CBC automation technology.
Looking to the future, what excites me most about Moichor is the opportunity to extend our CBC successes to solve more visual diagnostics problems. Between the rollout of our point-of-care device and the expansion of our automated test offerings, there’s a lot of room for technical innovation. Success requires clever computer vision R&D guided by interdisciplinary collaboration with our expert veterinary pathologists.
Thus, the stage is set for veterinary pathology AI to flourish, and at Moichor we’re pushing to make that happen. We’re growing our data science team to tackle bigger, more ambitious projects.
I invite anyone who’d like to play a part in solving interesting AI problems in microscopy — who’s intrigued by the unique challenges of veterinary data, or just loves animals and wants to help them — to get in touch. If you’re curious or have any questions about AI at Moichor, send me a message at email@example.com. I’d love to hear from you.