Radiologist says specialty should harness AI to prepare for demand spikes
A noted radiologist is urging the specialty to prepare for future demand surges with the help of machine learning.
Harvard Medical School’s James V. Rawson, MD, shared his thoughts in a guest editorial published Wednesday by Academic Radiology. Imaging workloads continue to rise while the number of rads available is failing to keep up, leading to burnout in both academia and private practice, he wrote.
Rawson highlighted a March study published by the same journal, detailing how an AI model can forecast daily radiology workloads and aid in practice management.
“What if you could proactively predict your clinical volume for tomorrow and had an opportunity to plan/staff differently? Would your practice be able to respond?” Rawson—a senior lecturer at Harvard, RSNA Honored Educator, and Association of Academic Radiologists Gold Medal winner—wrote Aug. 6.
The original study, also written by Harvard radiologists, compiled a year’s worth of imaging demand data from two academic medical centers. Experts used these figures to produce an explainable prediction of the next weekday’s clinical workload, with continuous learning helping to maintain the AI model’s quality over time. They landed on a final model based on three features—the current count of unread images, number of exams scheduled to be performed after 5 p.m., and those slated for the next day.
When tested across various radiology divisions, Leslie K. Lee, MD, and co-authors found the model “significantly outperformed” trivial demand estimates. AI was able to provide an accurate daily prediction pattern, with the solution successfully implemented in an online dashboard. Retraining this model on a weekly basis using live data has resulted in “high, and sometimes increased, model quality.”
In his editorial, Rawson noted that, in the past, radiologists experienced ebbs and flows in demand, with slower days allowing them the opportunity to catch up. However, today’s reality means many radiologists are slammed every day. Various factors are to blame, including a growing and aging population and radiology residencies failing to keep up with retirements.
“Whether it is a workforce shortage or a volume problem, there is an imbalance between the two,” Rawson wrote. “As radiology transitions from a mom-and-pop industry to larger scale, small problems of variance can become amplified.”
He hopes that greater understanding of variation can minimize the number of days radiologists are overwhelmed, while also reducing moral injury and burnout. Forecasting will not solve the serious manpower shortage radiology faces, but it’s “half of the battle,” he believes, and could additionally reduce errors and boost patient safety.
“Radiology work schedules are often made as shifts without considering variation or changes in volume or complexity. Using AI to predict large variation in volume prospectively provides an early warning and an opportunity to plan,” Rawson concluded. “Processes such as internal moonlighting—possibly with surge-like incentives or sending cases to external services to help catch up—would need to be set up in advance. Practices would need to build adequate overflow values and mechanisms to support the periodic flexing of the radiologist workforce, which can be deployed on short notice. This would allow the practices to respond with an incremental workforce on short notice to meet the increased volume.”
Read much more in Academic Radiology.
Marty Stempniak has covered healthcare since 2012, with his byline appearing in the American Hospital Association's member magazine, Modern Healthcare and McKnight's. Prior to that, he wrote about village government and local business for his hometown newspaper in Oak Park, Illinois. He won a Peter Lisagor and Gold EXCEL awards in 2017 for his coverage of the opioid epidemic.