David Hachuel wants pictures of your poop — for science.
The computer scientist-turned-entrepreneur is working to build the world’s largest database of human stool photos — up to 100,000 in all. The images will be used to teach an artificial intelligence to tell the difference between stool that’s consistent with good health and stool that could be evidence of gastrointestinal ailments like irritable bowel syndrome or Crohn’s disease.
The color, shape and consistency of stool hold important clues that help doctors make diagnoses. Hachuel thinks the photos can form the basis of an app that nonphysicians can use to obtain such information on their own. We’re already tracking our steps and calories and monitoring our heart rates; he hopes to build a poop tracker reliable enough to spot gastrointestinal problems quickly, easily and with minimal embarrassment.
But first, he needs oodles of photos.
“Algorithms need a lot of data to understand what they’re looking at,” said Hachuel, CEO of Auggi, a Brooklyn, New York-based startup that he hopes can commercialize stool-reading technology. “We don’t care or want to know who is contributing to the images — but we are very thankful for the data.”
High-tech medical care
Advances in computer vision have already transformed medical care. Computer vision is now augmenting, and in some cases replacing, the work of radiologists, lab technicians and other medical professionals. Similar technology is available to laypeople, via A.I.-powered smartphone apps. Existing apps can help identify skin cancer, offer insights about emotional well being and suggest personalized fitness plans.
Bringing a powerful AI tool into the bathroom could help people learn if they are among the 70 million Americans living with gastrointestinal disease, according to Hachuel.
What do experts think of the idea? Douglas Bradford Mogul, an assistant professor of pediatrics at Johns Hopkins University School of Medicine who isn’t involved with the stool-tracking initiative, thinks such an app could be helpful — especially for teenagers and others who tend to be reluctant to bring their concerns about unusual bowel movements to their doctors.
But he cautioned that — as with any do-it-yourself healthcare app — at-home poop analysis might cause problems. “One wouldn’t want hundreds of thousands of people going to their doctors because they received information from some algorithm that said that ‘this might be irritable bowel syndrome,'” he said. “I think, on the one hand, you could have some potential for some value, but I also think it could lead to unnecessary use of health care resources and anxiety.”
Eric Topol, founder of the Scripps Research Translational Institute in La Jolla, California, and an expert on the use of computers in health care, offered another reason for skepticism. He said many medical researchers believe that the quickest path to improving patient health involves analyzing the genetic makeup of all the bacteria inside poop — not by using A.I. to analyze photos of poop. “If you just take pictures and you get A.I. feedback and it doesn’t change anything,” he said, “what good is it?”
But Hauchel said that rather than offering a definitive diagnosis, the technology will simply help users make incremental lifestyle changes — in much in the same way that people use fitness trackers and other “quantified-self” tools.
Each year, millions seek medical care for gastrointestinal distress, and some bring pictures to their doctor visits. Hachuel stumbled upon the weird world of poop photos while pursuing a master’s degree in computer science at Cornell University, where he met a gastroenterologist who said his inbox was full of stool images sent by patients desperate to understand their digestive trouble.
Realizing there might be an opportunity to use A.I. to interpret such photos, Hachuel teamed up with MIT researchers to develop a computer vision tool that relies on so-called deep learning — a sophisticated form of pattern recognition that, once trained on a trove of images, is able to identify and label images just as well as any human.
The goal was to teach their AI to categorize human stool using the Bristol stool scale, a medical tool that identifies seven distinct types of feces: from sausage-shaped forms that suggest normal bowel function to small, hard lumps that suggest constipation and watery feces, which of course indicates diarrhea.
Lacking a database of poop photos, Hachuel and his collaborators used Play-Doh to create thousands of examples of fake stool, dropped them into a toilet, snapped photos and then fed them into a computer. Once trained, the AI was able to label the fake poop pictures with 100 percent accuracy, according to Hachuel.
But Hachuel and his team want to hone their technology with images of real poop, so they’re inviting everyone to upload their snaps to a website, which he said would strip away identifying information before adding it to a database that would form the basis of an app that helps people categorize and track their stool over time.
“Once the A.I. is trained, individuals will be able to use it to track bowel movements objectively,” Hachuel said, adding that the idea is to track patterns over time and use them to inform a medical provider’s assessment.
“We’re essentially using technology to help those who suffer from a chronic condition to better understand how their lifestyle affects their condition,” Hachuel. “Everybody should be curious about what their gut health is about.”