|
Activity Number:
|
246
- Improved Disease Classification Through Extensions of ROC Curve Estimation and Biomarker Characterization
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 9, 2022 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Medical Devices and Diagnostics
|
|
Abstract #323503
|
|
|
Title:
|
Analyzing Partially Paired Data for Diagnostic Radiology Studies Using the Obuchowski-Rockette Method
|
|
Author(s):
|
Stephen L Hillis*
|
|
Companies:
|
ASA
|
|
Keywords:
|
Obuchowski-Rockette;
diagnostic radiology;
MRMC;
partially paired data;
reader performance
|
|
Abstract:
|
In multi-reader multi-case (MRMC) diagnostic radiology studies it is usually desirable to have conclusions apply to both the patient and reader populations. A typical MRMC study design involves several radiologists assigning disease ratings to cases that were imaged using several imaging modalities. The area under the receiver operating characteristic curve (AUC) is a commonly used reader-performance metric. A common method for analyzing such studies is the Obuchowski and Rockette (OR) method (1995). A shortcoming of the OR method has been that it could not be used with partially paired data where not all readers read the same cases. For example, readers may not all read the same study images because of a technical problem (such as an online transmission problem) or because the study design has each reader read only a random sample of the available truth-verified cases. In this talk I discuss how the OR method can be easily adapted for partially paired data and empirically validate the approach through simulations, as well as providing examples of applying the approach to real data sets.
|
Authors who are presenting talks have a * after their name.
