RSNA 2022 Imaging Informatics Preview

Machine-learning model improves workload balance, fairness

By Jenny Jones, contributing writer

Monday, November 28 | 9:30 a.m.-10:30 a.m. | M3-SSIN02-5 | Room E350
A machine-learning model that estimates the difficulty of interpreting a study can improve fairness and workload balance, while helping to prevent cherry-picking of studies, according to this presentation.

Dustin Sargent, PhD, a data scientist with IBM Watson Imaging, will present how the team used simulations of real clinical sites to train their online regression model and discuss how the model works within the study distribution engine that assigns studies to radiologists.

The model produces what the group calls study difficulty estimation (SDE) using patient age, body mass index, study description, number of priors, and other data points. It also considers the reader's expected turnaround time for that exam type, independent of radiologist skill level.

The researchers studied the model's effectiveness by comparing results from their study distribution engine -- both and without SDE -- during 10-day simulations. They used the following metrics: the standard deviations of the total relative value units (RVUs), total difficulty units, number of exams read, percentage of x-ray exams, study preference and rejection rates per radiologist, and average turnaround time. They also evaluated an RVU-based distribution rule with and without the SDE.

The results showed that while turnaround times were faster without the SDE, it was at the expense of overworking efficient readers. The SDE, researchers said, resulted in a more equitable working environment that benefits radiologists and their patients.

Learn more by sitting in on this Monday talk.

Copyright © 2022

Last Updated mf 11/23/2022 4:41:01 PM