Sudhen Desai, M.D., FSIR, interventional radiologist at Texas Children's Hospital, editor of IR Quarterly for the Society of Interventional Radiology (SIR) and on the Board of Directors for the Society of Physician Entrepreneurs, explained how artificial intelligence (AI) can assist in pediatric imaging and the pitfalls of training AI systems. He spoke at the 2019 Radiology AIMed conference.
Deep learning algorithms require large amounts of patient case data to train the systems to read medical images automatically without human intervention. However, in pediatrics, there are often much lower numbers of normal and abnormal scans that can be used compared to vast amounts of adult exams available. This makes it difficult to train systems, so AI developers are coming up with innovative new ways to train their software. Compounding issues with training pediatric imaging AI is that the normal ranges change very quickly for young children due to their rapid development. He explained what is normal for a 2-year-old may not be normal for a 5-year-old.
Desai and other pediatric physicians who spoke at the conference said AI could have a big impact on pediatric imaging where there are not enough specialists for the increasing image volumes.
Related content:
VIDEO: Implementation of Artificial Intelligence Tools in Radiology Practice — Interview with Lawrence Tanenbaum, M.D.
VIDEO: AI That Second Reads Radiology Reports and Deals With Incidental Findings — Interview with Nina Kottler, M.D.
Technology Report: Artificial Intelligence at RSNA 2018
VIDEO: Implementation of Artificial Intelligence Tools in Radiology Practice