Use of Equivalent Relative Utility (ERU) to Evaluate Artificial Intelligence-Enabled Rule-Out Devices

We investigated the use of equivalent relative utility (ERU) to evaluate the effectiveness of artificial intelligence (AI)-enabled rule-out devices that use AI to identify and autonomously remove non-cancer patient images from radiologist review in screening mammography.We reviewed two performance metrics that can be used to compare the diagnostic performance between the radiologist-with-rule-out-device and radiologist-without-device workflows: positive/negative predictive values (PPV/NPV) and equivalent relative utility (ERU). To demonstrate the use of the two evaluation metrics, we applied both methods to a recent US-based study that reported an improved performance of the radiologist-with-device workflow compared to the one without the device by retrospectively applying their AI algorithm to a large mammography dataset. We further applied the ERU method to a European study utilizing their reported recall rates and cancer detection rates at different thresholds of their AI algorithm to compare the potential utility among different thresholds. For the study using US data, neither the PPV/NPV nor the ERU method can conclude a significant improvement in diagnostic performance for any of the algorithm thresholds reported. For the study using European data, ERU values at lower AI thresholds are found to be higher than that at a higher threshold because more false-negative cases would be ruled-out at higher threshold, reducing the overall diagnostic performance. Both PPV/NPV and ERU methods can be used to compare the diagnostic performance between the radiologist-with-device workflow and that without. One limitation of the ERU method is the need to measure the baseline, standard-of-care relative utility (RU) value for mammography screening in the US. Once the baseline value is known, the ERU method can be applied to large US datasets without knowing the true prevalence of the dataset.