University of Dayton researchers are helping Cincinnati Children's Hospital Medical Center by developing artificial intelligence to better detect lung nodules in CT scans of children. To date, no pediatric-specific systems have been developed or tested.
"In children, the presence of lung nodules as small as three millimeters in diameter can signal cancer elsewhere in the body has spread to the lungs," said Russ Hardie, UD professor of electrical and computer engineering, who received a two-year, $150,000 grant from the National Institutes of Health with Barath Narayanan, a senior research scientist at the University of Dayton Research Institute. "Early detection of such metastatic cancer, or cancer that spreads, is critical for therapy planning to improve outcomes for children and AI can assist radiologists in detection.
“The American College of Radiology promotes the principle: ‘AI used in pediatric patients should be designed for and shown to work in pediatric patients.’ This is exactly what we are aiming to do. It’s our guiding light. When we realized lung nodule detection for pediatric patients was a pressing issue for Cincinnati Children's Hospital Medical Center, it was a perfect fit."
Lung nodule detection by radiologists is challenging because of the large amount of data to review and the size of the nodules in pediatric cases is relatively small. The researchers hope to develop "another set of eyes" to aid radiologists in an otherwise time-consuming and tiring task subject to human error.
"When radiologists review a CT scan, there are hundreds, sometimes thousands of images,” Hardie said. “They scroll back and forth through the images along one of the three dimensions, looking for the telltale spherical shape of a nodule. However, there are vessels throughout the lungs. In a single cross-section image, it is difficult to distinguish vessels from a nodule. Everything just looks round.”
Hardie and Narayanan had previously created an efficient and accurate lung nodule detection system for adults called FlyerScan. A recent study by the researchers showed that the adult version of FlyerScan did not perform as well on pediatric data from Cincinnati Children's Hospital Medical Center.
"The vascular structure in children is more compact and nodules tend to be much smaller in pediatric cases than in adults, with many as small as one to two millimeters," Hardie explained.
That's when the team determined a need for a pediatric-specific lung nodule detection system, leading to the current NIH-funded project.
“The project would not be possible without the support and collaboration of the team of radiologists at CCHMC,” Hardie said. “The radiologists provide clinical expertise to guide the system development and evaluate its efficacy. The team at CCHMC is also providing CT scans and expertly annotated data with which to train the AI system.”
For more information on the project, contact Hardie at rhardie1@udayton.edu or Narayanan at narayananb1@udayton.edu.
For interviews, contact Shawn Robinson, associate director of news and communications, at 937-229-3391 or srobinson1@udayton.edu. Meredith Hopkins, director of strategic engagement, originally reported the story, portions of which are excerpted here.