...

Dan Ruderman talks about the potential of machine learning in cancer research and explains what career lessons we can take from Mister Rogers.


Machine learning and computer vision have the potential to bring huge improvements to the medical sector. Cancer research, in particular, can benefit from new imaging and analysis techniques. But it’s not only the disruptive technology that is interesting; the people behind it are at least as fascinating.

I had the opportunity to talk to Dr. Dan Ruderman, Assistant Professor and Senior Director of Analytics at the Ellison Institute for Transformative Medicine at USC. Dan is not only working on the forefront of cancer research today, he is also living proof that a career path that seems random at first can lead to very meaningful work.


“I’ve always loved the smell of solder”

Dan always had a curious mind. When he was a kid, the huge mainframes at Lawrence Berkeley Labs, where his father worked, fascinated Dan. He and his dad spent long weekends in the basement of their house, building microcomputers and programming them.

“It was the early 80s, so we had 6502- and Z80-based machines”, Dan says. “I’ve always loved the smell of solder,” he adds with a smile.

His physicist father didn’t only teach Dan about computers: “I was very curious and my father could almost always answer my questions. And if he couldn’t, he came up with a good way to think about the problem.” This strategy of problem-solving would follow Dan throughout his career, opening doors that others wouldn’t even notice were there.


Wall Street meets Disney

Although Dan studied physics in college, his knowledge of programming and computers was what drove him forward, paired with his hunger to solve any problem that was set before him.

As a graduate student, he went on to work on neural networks and visual neuroscience during the 1990s. Then the dot-com boom persuaded him to leave academia behind. Being interested in a wide range of topics, Dan changed industries like other people the brand of their breakfast cereal: “I did consulting for trading platforms for Wall Street, I did some tuberculosis tracking software in California. I did facial motion capture for Disney Feature Animation.”


Living like Mister Rogers

mr-rogers-neighborhood

For each job, Dan’s knowledge of computer science was his key to success. While each position focused on a different area, computers were what they all had in common. Dan mentions Mister Rogers’ Neighborhood, the American children’s TV show. Every episode, Mister Rogers would go on a trip and visit somebody different around his neighborhood. Dan explains: “I always wanted my life to be like that. I didn’t want the same experience every day. No, I wanted that [Mister Rogers’ life].”

In each new field, Dan was eager to learn. “They all have their own way of solving problems. And so I paid attention to that,” he says. At first, this knowledge seemed kind of scattered and random, Dan admits: “People looked at my resume 10 years ago and said ‘That guy is nuts!’”

But finally, the pieces started to fall into place. Because cancer had taken his father, Dan took the first opportunity he had to work in cancer research. “[This work] showed me that cancer was becoming a very data-intensive challenge,” he tells me. After a few years in pharmaceutical research and at Danny Hillis’ Applied Minds, Dan found his way back to academia. He now is a faculty member at the Ellison Institute at USC.

Today, all the seemingly random knowledge acquired throughout the past decades isn’t so random anymore. The early neural network research, the 3-D motion-tracking, the algorithmic thinking of Wall Street trading — it all helps Dan in his work today. “Now [people] look at [my resume] and say ‘Wow, it’s like these things all make sense actually’,” Dan says and laughs.


Letting computers do what they’re good at

“[We’re] investigating (…) the 3-dimensional microstructure of tumors,” Dan says, explaining his research work. His team is analyzing images of tumors on a cellular level to identify certain characteristics of a cancer. A 3-D representation of a tumor may allow doctors to better predict the progression of a tumor and to identify the cancer subtype, among other functions.

Machine learning is playing a leading role in this research. “Until now, pathologists have analyzed these images by eye,” Dan tells me. “But that’s exactly the kind of thing that computer vision is really good at, picking up patterns. We’re leveraging those great successes to digital pathology.”

He continues: “Pathologists would have to look at every cell and see which ones look like they’re dividing (…). It’s very laborious.” A computer vision algorithm, on the other hand, can analyze such an image much faster. It can pay close attention to details, while at the same time looking for larger structures. “We just can’t see those things. A computer can see everything at once.”


AI should assist humans, not replace them

ai-assist-humans

Dan emphasizes that his research is not intended to replace human pathologists with machines. Instead, computers should assist humans by pre-filtering the massive amounts of information and help doctors and clinicians focus on what’s important. “The human eye isn’t the best way to interpret everything that nature throws at us. (…) For example, we’re very bad a looking at an image and counting things,” Dan explains. “Computers can pick out parts from the ever-expanding data that the clinician should pay special attention to.”

In the near future, Dan hopes to see his research applied in clinics around the world. Already today, new technologies allow for the collection of huge amounts of very precise data from the human body. AI-based systems can help to interpret this vast amount of information and use it to help doctors make the right decisions.

Dan gives a concrete example for the future of digital pathology: for breast cancer biopsy, a pathologist would analyze microscopy images using software that automatically highlights suspicious areas. “You’ll have a system that says ‘Hey, this looks like that category of breast cancer, you may want to check that out’. Diagnostic accuracy is crucial for getting the treatment right.”


Cloud-based cancer screening

This analysis could even happen in the cloud, significantly reducing the costs for such services. Dan emphasizes how this development could help poorer countries to improve their healthcare systems. It could enable hospitals around the world to offer the highest-quality diagnostics, as long as they have a slide scanner and an internet connection. “I think this telemedicine aspect can really change a lot of lives,” Dan hopes.

From the days in his parents’ basement back in the ’80s to cutting-edge cancer research today, Dan Ruderman has zig-zagged his way through various industries, from Wall Street to Disney. His toolbox filled with computer science and problem-solving skills, he always went where it seemed the most exciting and where he could make a difference.

This seemingly chaotic career path has eventually led Dan to one of the most meaningful and critical research fields of today. No doubt Mr. Rogers would approve of that.