Two Carnegie Mellon researchers create digital database of poisonous plants
April 15, 2013 8:00 AM
Marios Savvides, director of the biometrics center at Carnegie Mellon University demonstrates the app on poison ivy. The app identifies the plant, explains if it is toxic, and recommends treatments if it is.
Marios Savvides, director of the biometrics center at CMU, uses a smartphone with an app his team has created with Cynthia Morton, curator of botany at the Carnegie Museum of Natural History.
By Doug Oster Pittsburgh Post-Gazette
The biometrics lab at Carnegie Mellon University is filled with young, brilliant minds intensely working on computers. It doesn't look like they ever leave the glow of the monitors as bags of chips and other snacks sit on their desks.
On one screen is a human face covered with hundreds, maybe thousands, of dots.
The place looks like most computer labs on campus except for a strange apparatus resembling a telescope. It's a cylinder probably 12 inches in diameter and 3 feet long with large square lights on each side -- it definitely looks like something military.
New software helps identify poisonous plants
Dr. Marios Savvides of Carnegie Mellon University demonstrates a new app designed to help identify poisonous plants. (Video by Doug Oster; 4/15/2013)
The team here is working on developing and refining facial recognition software. That high-tech cylinder is related to iris recognition research the lab does for the Department of Defense.
It's not the kind of place you would expect to find a plant expert, until now. Cynthia Morton, curator of botany at the Carnegie Museum of Natural History has partnered with Marios Savvides, director of the CyLab Biometrics Center, to create an innovative new system that can identify poisonous plants.
The genesis of their relationship was when the two kept bumping into each other on campus. They became friends as Mr. Savvides always had a plant question or two for Ms. Morton. Both committed to their respected fields, they wondered if there was a project to collaborate on. When one of Ms. Morton's colleagues was researching poisonous plants, they found a match for their skills.
They began researching how poisonous plants are handled by the medical community. Both were dumbfounded to discover there was no definitive electronic database for identifying and treating the plants in question.
When a patient comes to the emergency room after eating a plant, doctors have a couple of options if they can't figure out what was ingested. They can search through volumes of medical toxicology books, or go online to a site called Micromedics. Ms. Morton's research found much of the information regarding identification, symptoms and treatment was missing. "I was shocked, I was really shocked," she said. "After talking to ER doctors I realized we've got something here."
They looked at national poison control records and found a list of the most reported plants. It included backyard cultivars, plants in the wild and houseplants. Some are poisonous and others aren't. On the top of the list of ingested plants is poison ivy; No. 10 is dandelion, which is not toxic. In between are plants such as dieffenbachia and oleander, both poisonous.
The two researchers combined forces, figuring the facial recognition software Mr. Savvides and his team developed could be used to identify plants, too. "That's the most awesome feeling," Mr. Savvides said. "I believe the true home runs left are cross-disciplinary research, that's when we fill in the technology gaps."
They pared the poison control list of 32 plants to 10 and presented the idea to UPMC for funding. It wasn't long before UPMC gave them $100,000 to create the initial database and app (software for a smartphone and tablet). Ms. Morton and her team would collect the specimens and scan them. Each plant listing needs 50 to 100 two-dimensional images. The pictures are mainly of the foliage in different stages of development. Mr. Savvides team uses different mathematical representations of the shapes, textures and other features to analyze an image. The results are fascinating. Using a smartphone with a beta version of the software installed, Mr. Savvides photographed a plant. In an instant, all the information about the variety was displayed on the tiny screen -- technical name, toxicity and treatment.
The software can also identify a plant through a methodical process in which a patient can rule out choices of leaf shape and other features until the right one is discovered. Patients can also scroll through photos until they recognize what they ate.
Without the database, if the plant can't be identified, doctors need to treat the patient accordingly, which could mean unnecessary treatment. "This could really help people," Ms. Morton said, "whether it's reducing the invasiveness of medical procedures or saving someone's life."
The methodology for adding plants to the system has been fine-tuned. The researchers said more work is needed to improve the functionality of the software. They hope to make it available to the medical community after the research team is satisfied with the system's performance.
The pair has taken their research to doctors and patients, looking for suggestions. They hope for more funding to increase the number of plants in the database. Ms. Morton thinks the applications for the project are limitless and could expand greatly as more plants are added.
"You'd have a poisonous control database for the United States, Canada and then Europe," she said, "so you can look up almost anything wherever you are in the world and see if it's poisonous."