What if you found out your infant had eye cancer? That news would rock anyone’s world. But what if you had a tool that helped you catch it early enough that your baby didn’t have to lose his or her eye and didn’t have to go through chemo? You’d probably do almost anything to get it. Bryan Shaw has dedicated his time to helping parents detect this cancer sooner so their children don’t have to go through what his son went through—and he’s doing it for free. With computer scientists from Baylor University, he’s harnessed the power of a machine learning algorithm to detect cancer that no human eye can detect.
Below is a partial transcript. For the full interview, listen to the podcast episode by selecting the Play button above or by selecting this link or you can also listen to the podcast through iTunes, Google Play, Stitcher, and Overcast.
Bryan Shaw: “The very first person who ever contacted me because our app helped them was a gentleman in Washington State, and his little girl had myelin retinal nerve fiber layer, which is an abnormal myelination of the retina, and it can cause blindness, but it presents with white eye. And his little girl was five years old, and he kept seeing white-eye pics. He heard our story. He downloaded our app. Our app detected the white-eye pics. That emboldened him enough to grill the child’s doctor. You know, ‘My camera’s telling me this. Look, this app. I heard this story . . .’ The doctor takes a close look. The girl had been 75 percent blind in one of her eyes for years, and nobody had ever caught it.”
Ginette: “I’m Ginette.”
Curtis: “And I’m Curtis.”
Ginette: “And you are listening to Data Crunch.”
Curtis: “A podcast about how data and prediction shape our world.”
Ginette: “A Vault Analytics production.”
Curtis: “Data Crunch is again brought to you by data.world, the social network for data people. Discover and share cool data, connect with interesting people, and work together to solve problems faster at data.world. Did you know that you can add files via URL to your data sets on data.world? Data.world APIs allow you to pull live survey data into your data set, enable automatic file updates, and more. Get the full details on data.world APIs at docs.data.world, or search ‘Austin Cycling Survey’ on data.world to see live survey sync in action in Rafael Pereira’s data set!”
Ginette: “One quick reminder that our data competition is currently up on data.world. Be sure to post your submissions by May 5.
“Okay, now back to the story. If you know someone who’s about to have a child, has a child five or under, or plans to have children, you need to send them this episode, and you’re about to find out why from this man, Bryan Shaw.”
Bryan: “When Noah was three-months-old, we started noticing that a lot of his pictures had white pupillary reflections, what doctors call leukocoria, white core, white pupil, and that can be a symptom of a lot of different eye diseases.”
Ginette: “You probably put this together, but Noah is Bryan’s son. And to add in Noah’s mom’s perspective here, when she started noticing this strange white reflection in Noah’s eyes, like most moms today, she aggressively searched the Internet for answers. Like Bryan said, leukocoria could indicate a disease, or it could indicate nothing, but the Shaws decided they needed to tell their pediatrician about what they’d found.”
Bryan: “Noah passed all his red reflex tests, until we told his pediatrician that we noticed leukocoria, and he had a very good pediatrician—Pearl Riney, Cambridge, Massachusetts. And then she really looked really, really closely. And on that test, she noticed a white pupillary reflection and immediately sent us that afternoon to an opthamologist.”
Ginette: “At this point, Bryan’s wife, Elizabeth, was freaking out because she’d done all the research about leukocoria, or white eye, and she knew what white eye might mean for their four-month-old son.”
Bryan: “In Noah’s case, it was caused by tumors in the back of his eye, and they were reflecting light, and this is what was causing his pupils to be white. ”
Curtis: “Noah was diagnosed with retinoblastoma, or RB, a very dangerous cancer.”
Bryan: “The sad thing about RB is it’s a pediatric cancer. It only affects the retina when it’s developing, which is the first five years of life.”
Ginette: “It’s location is partly what makes it so dangerous.”
Bryan: “The tumors metastasize straight down the optic nerve to the brain, and then it’s tough to treat.
“We quickly started therapy. He underwent Chemoreduction. This is where you give the kid chemotherapy systemically throughout the entire body to try and shrink the tumors before you then go in to photoblate them with a laser or eradicate them with a radiation beam. And it didn’t work, but we were able to keep the tumors at bay through blasting the tiny ones that would pop up with lasers and freezing some of the larger ones, but eventually the tumor started to break up in his right eye, and that’s really bad. You get these seeds going everywhere. And so, when he was nine months old, we removed the right eye, and we treated the left eye with radiation therapy, which is an ordeal in and of itself because a kid, a nine month old, won’t sit still in a radiation beam. So you have to put him asleep. So every day, we had to do general anesthesia. Every day. And that’s considered a surgery, so we had 30 cycles of that. Thirty cycles of the radiation, and 30 times we had to put him under general anesthesia.
“Radiation, radiation sucks. It’s better than nothing, but it’s just not something you want to do on an infant under one year old. When you give a little infant radiation, you increase the risk of secondary cancers later in life, but we had to do it. That saved his left eye.”
Ginette: “During this process, Bryan had an idea, and it turns out he was in a unique position to do something about it.”
Bryan: “I’m a protein biochemistry professor. I got my Ph.D. from UCLA, and then I postdoced at Harvard.”
Ginette: “This is where he was when he found out Noah had RB.”
Bryan: “So about a year later, I left Harvard. I got a faculty position at Baylor and all this startup money and a lab and students who wanted to work with me and colleagues over in computer science, and I became curious, interested in trying to figure out how early had the white-eye pictures started showing up in our own family photos? So I got all the pictures of our family that my wife and I to some degree took. There were about 10,000 of them, and I looked through them all—every single one—and I found out that leukocoria started showing up at 12 days old.
“I went to his ophthalmologist, Shizuo Mukai, and I asked during one of his check ups for a prognosis if I had brought Noah in at 12 days old instead of 4 months old when he was diagnosed, and he predicted that Noah probably would have been able to keep his right eye. The tumors would have been small enough that we would have been able to salvage it, and it would have been worth salvaging ‘cause the tumor would have been smaller. He also said that in addition to keeping his right eye, Noah probably wouldn’t have needed proton beam radiation to his left eye, which is huge. And so, when he said that, I sort of freaked out. You couldn’t tell I was freaking out, but I was freaking out on the inside. And that’s when I realized that we needed to develop software that can search through parents’ photo libraries and alert them to the presence of leukocoria.”
Ginette: “This realization was a turning point for Bryan. If they would have detected RB earlier, Bryan and Elizabeth could have likely saved Noah from chemotherapy and everything that came with it, like potential future cancer for their son. So he wanted other mothers and fathers to have this option though a software that could detect it.
“It was during this same time period that Bryan was searching all the academic literature to see what research had been done on RB. And he found a lot of interesting information, some of which ended up shaping the software.”
Bryan: “David Abramson at Memorial Sloan Kettering published a paper on this, and he looked at the percent of RB children in his local hospitals, kids with RB, who had their symptoms initially detected by a parent or relative, a pediatrician, or an ophthalmologist, and in about somewhere around 80 percent of the cases, a parent detected the symptom first, and the symptom could be a lazy eye. It could be leukocoria. It could be any of the symptoms that are associated with retinoblastoma. So pediatricians only detected the symptoms ab initio, that means first among whomever, in less than 10 percent of the cases. Ophthalmologists detected it in the remaining fraction, ab initio. And this gets back to a well-accepted fact that the standard test that the pediatrician uses to look for retinoblastoma, the red-reflex test, is just not very good. So the red reflex test is used to look for a red reflex. They shine a light in the eye, and you look for a red reflection. It’s tough to perform because the kid is always, they’re noncompliant. They’re an infant. They’re moving around. It’s just tough. It’s tough to do.”
Ginette: “In addition to verifying that parents are usually the first to notice leukocoria and that the red reflex test isn’t very good, he also noticed that no one had published a paper on what he had found in his photo album—that he saw leukocoria in an infant at 12 days old.”
Bryan: “So I wanted to do the app, but I had to publish papers on it too and try to get grants, so the next step was publishing the first paper. I wanted to publish a paper in a scientific, peer-reviewed journal on me analyzing all my photos because nobody had ever done that before, and when you read in the literature about the correlation between leukocoria and the prognosis of the child, almost across the board the research articles suggested that leukocoria was a late-stage symptom, and that by the time you noticed leukocoria in a child, you weren’t going to be able to save the eye, but I knew that I was observing leukocoria at 12 days old. It was just what I call low-frequency leukocoria. It was showing up in like one in every 20 pictures or something, but as the tumors got bigger, and they multiplied and started showing up in more and more and more pictures, and that’s when we noticed it—high frequency leukocoria, so I would say high frequency leukocoria is a symptom of late stage or later stage RB, but the low-frequency leukocoria could be a symptom of early stage RB.
“So it was so funny trying to get it published. I sent it to Pediatrics, the premier pediatrics journal. They actually loved it. The editor of Pediatrics, Lewis First, is such a nice guy. I spoke to him on the phone. He asked me how my son was doing. He really got what we were trying to do. The pediatricians get it. The ophthalmologists get it.
“But he said, ‘Hey, look. This is a case study on one patient. If you want to publish it here, you’ll have to do more patients.’ So I sent it off to some optics journal. And the reviews were so awful. They were so bad. I’ll never forget this one peer reviewer. Their first sentence said, ‘there are so many problems with this paper, I don’t even know where to begin,’ and what they didn’t like was that I was using quote-unquote random photography in a nonclinical, recreational environment to study the pupillary reflexes of kids, in this case retinoblastoma.
“And I would retort. I would send these editors the nastiest emails. I know they have them framed ‘cause who can send a nastier email than a parent/scientist working on something that could have saved their kid, and then it gets shot down. I mean I know they probably framed this. But the paper was published in PLOS One, and it’s been viewed 10,000 times, and I’m sure the other journal wished they’d probably got that paper.”
Curtis: “After Bryan published this paper, his next step was to see if he could get a computer scientist at Baylor to help him build software that could accurately detect leukocoria. That person turned out to be Professor Greg Hamerly.”
. . . Above is a partial transcript. For the full interview, listen to the podcast episode by selecting the Play button above or by selecting this link or you can also listen to the podcast through iTunes, Google Play, Stitcher, and Overcast . . .
Ginette: “Thanks again for listening, and special thanks to Bryan Shaw, Greg Hamerly, and Ryan Henning! If you have any questions for us, feel free to email us at Curtis@vaultanalytics.com. We read all the responses we get, and we respond.”
“Is That You or Are You You” Chris Zabriskie,