Analysis Featured

How Machine Learning Turned Pathological

3scan machine learning for histopathology

For 150 years, biologists have been analyzing disease tissue in the lab much the same way. They cut it, put it on glass, and then shine a light on the glass to determine whether the tissue is healthy or not. Thanks to computer vision and machine learning, researchers can examine tissue 100 times faster. Todd Huffman at 3Scan has developed a process that is much more efficient, accurate, and time saving.

“We can now generate data so much faster than before,” Huffman said. “There’s a lot more to see. In fact, there’s more to see than humans can look at with the naked eye.”

Using distributed computation, Huffman’s method generates up to 200 terabytes of imagery data off one object. With that much bandwidth, cloud-based storage is a necessity. However, only certain bits of that information is stored in the cloud. If his crew needs immediate access to the results, “we’re talking in the second and millisecond range,” that data is stored at the local level. If the computations take place over a longer period of time, then that’s when the information is moved to the cloud.

3Scan uses a patented knife-edge scanning microscope (KESM) that uses a diamond for the cutting edge. This works extraordinarily well because diamonds are strong and transparent. Pointing their optical equipment at the robotic microscope on the edge of the blade, his crew can image the blade as it cuts the tissue being analyzed. The camera used is from the Teledyne DALSA Piranha series. Employing Apache Spark to distribute computations, 3Scan has migrated from convolutional networks to their own in-house machine learning algorithm. Their aim is to turn tissue biology and histopathology into a data science.

How 3Scan is Changing the Chase for a Cancer Cure

Using a business model known as contractor research organization, or CRO, 3Scan manufactures its own equipment to test tissue sent to their labs by their customers.

“We build our own robots and analysis tools because no one else manufactures this stuff,” Huffman said. “Our workflow consists of a research group taking the disease tissue and fixing it in formaldehyde. We then stain it, image it, and analyze that tissue then send the analysis back to the customer.”

Huffman started working on the concept for 3Scan in 2005, picking up the mantle from another scientist who passed away.

“At the time, robotics and computations weren’t advanced enough,” Huffman said. “His work came to an end, but I thought it was a good idea, so I picked it up and headed forward with it.”

Initially, Huffman worked on the connectivity of neurons in the brain, but that posed a few challenges.

“Neurons are quite small and complicated,” he said, “and brains are huge on a microscopic scale. I had a convoluted way of studying the architecture at scale to the throughput that is needed.” That’s when he started working with another scientist who was using a diamond blade as a light source in optics. “Because diamonds are the strongest substance we know about, and transparent, it works very well.”

In 2011, he and his co-founders–Megan Klimen, Matt Goodman, and Cody Daniel–kicked off 3Scan. Since then they’ve raised $20.67 million in two rounds of funding with Lux Capital and Data Collective leading the way. Prior to that, they raised over $400,000 in seed capital with a grant from the Thiel Foundation giving their initial juice.

3Scan’s primary customers are universities and pharmaceutical companies.

“Universities try to understand diseases and pharmaceutical companies try to cure them,” Huffman said.

With about 20,000 pathologists in the U.S. and another 80,000 scientists who work with disease tissue, Huffman estimates there are 100,000 professionals who can use 3Scan’s services. They’ve barely touched the surface of that.

The Future of Pathology is Robotics

Huffman believes there will be a bigger push toward robotics in biology and medical research because of the necessity for repeating results.

“When work is done by hand, scientists can miss things,” he said. “That’s a problem. With robotics, everything is specific because you have to have it programmed.”

For that reason, he sees the use of robots, sensors, computer vision, and machine learning increasing. 3Scan itself is working on developing an entire family of robots to join KESM.

“We’re working on modernizing biological investigations,” he said. “But to make that happen, we need computer vision engineers. We need people who can work with moving and processing large amounts of data.”

The problem for smaller companies like his, however, is they have to compete with larger companies like Google for talent. Google has an entire division devoted to deep learning. Facebook, Microsoft, and Apple are also looking for talent from the same pool, and have much deeper pockets.

Headquartered in San Francisco, 3Scan directly competes with Google in Mountain View for suitable talent, but computer vision and machine learning engineers with a social conscience and a heart for curing disease could find a sense of satisfaction using their talents building pathological robots to analyze tissue in search for evidence of disease.


Allen Taylor

About the author

Allen Taylor

An award-winning journalist and former newspaper editor, I currently work as a freelance writer/editor through Taylored Content. In addition to editing VisionAR, I edit the daily news digest at Lending-Times. I also serve the FinTech and AR/AI industries with authoritative content in the form of white papers, case studies, blog posts, and other content designed to position innovators as experts in their niches. Also, a published poet and fiction writer.

Add Comment

Click here to post a comment

Your email address will not be published. Required fields are marked *