MedTech Speed to Data

Pictorlabs is a California-based startup developing a cloud-based platform that uses artificial intelligence to improve tissue sample analysis through virtual histological staining.

In Episode #34 of the Speed to Data Podcast, Key Tech’s Andy Rogers speaks with Pictorlabs Chief Product Officer Raymond Kozikowski about his company’s all-digital approach to tissue sample testing.

Need to know

·       Histopathology — The visual analysis of stained tissue samples to diagnose cancer and other conditions.

·       Tests have long turnaround times — Selection, preparation, and imaging can take as long as a day to return one test’s results to the physician.

·       Tests are requested sequentially — The results of one test determine the next test in the decision tree, so physicians can’t order all the tests simultaneously.

·       Cancer patients must wait — On average, there is a forty-day gap between biopsy and first treatment.

The nitty-gritty

As Dr. Kozikowski explains, “Histopathology has traditionally been a chemistry-based testing paradigm. Every cancer case starts with a biopsy, and those tissues are transformed into data that inform the diagnosis and therapeutic options.”

Pictorlabs’ solution uses one tissue sample to create a virtual stain that simultaneously generates results for dozens of tests within minutes. 

“What we’re doing is teaching AI algorithms the relationship between validated test results and the underlying signature from that unstained piece of tissue,” Dr. Kozikowski said. “From a single patient sample, you’re no longer limited to running one chemical-based test. You can run ten, twenty, thirty AI-based tests.”

Although the company thought it faced a long march toward the clinical market, Pictorlabs found an opportunity in a different market.

“There’s a really robust cancer research market, both the academic medical centers and the pharma companies. Where we really got traction wasn’t necessarily as a replacement [technology] but a complement to other kinds of tests.”

Dr. Kozikowski cites spatial biology as an example. Cells express their genes and RNA differently depending on their location in tissue. Understanding this spatial relationship could yield new, more targeted therapies.

“A challenge with interrogating RNA targets,” Dr. Kozikowski explains, “is that you often can’t also run traditional staining-based tests. With virtual staining, we’re actually able to complement those RNA-based tests with a pseudo-staining result. This is perfectly fit for purpose in those workflows.”

Data that made the difference:

The importance of data to AI development isn’t surprising, but Pictorlabs needs more than quantity.

“There’s also a lot of nuance in the design of that dataset and making sure it’s fit for purpose,” Dr. Kozikowski says. “Has it seen the diversity of human disease in that training dataset to really make sure that it generalizes accurately and robustly?”

Partnerships with the research community have helped refine Pictorlab’s technology. One of these relationships is with Dr. Michael Kallen, a pathologist at the University of Maryland’s School of Medicine.

“Diagnosing lymphoma or leukemia can be very, very complex. You have the challenge of managing a complex workflow in the lab and the complexity of making sense of all those test results spread over weeks or maybe even a month.”

“[Dr. Kallen] saw the opportunity. We’ve been partnered with that department for a while now, exchanging data to help train algorithms and get feedback from pathologists. We’ve just received an innovation grant to deploy our technology side-by-side with their existing workflows to look at the value.”

Watch the full video below to learn more about Pictorlabs’ virtual staining solution and to hear Dr. Kozikowski’s advice to product managers and entrepreneurs.

What is MedTech Speed to Data?

Speed-to-data determines go-to-market success for medical devices. You need to inform critical decisions with user data, technical demonstration data, and clinical data. We interview med tech leaders about the critical data-driven decisions they make during their product development projects.