Abstract:
|
While tasks in medical imaging and other diagnostic areas are often assisted by AI, most decisions require interpretations by trained human readers. Readers’ performance studies of technologies and devices gain additional complexity in unstructured tasks of detection and localization of multiple targets per image/subject. Several approaches are available for analyzing detection-localization studies in both structured (e.g., ROI) and unstructured settings (e.g., FROC). Recent developments unify both settings within the same framework widening the spectrum of analyses for unstructured settings. Yet questions remain on whether it is possible, or how, to borrow some statistical efficiency of the structured setting for analyzing readers’ performance in unstructured detection-localization tasks. In this talk we will discuss several approaches for analyzing reader’s performance in unstructured detection-localization tasks, illustrate properties of the statistical analysis, and offer practical recommendations.
|