Bias Reduction in the Design and Analysis of Imaging Studies

• Activity Number: 450
• Type: Topic Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Philip Lavin and Scott Chasan-Taber*+
• Companies: Lavin Consulting and Consultant
• Keywords: Diagnostic Imaging ; Modeling ; POM ; MRMC

Abstract

The introduction of new imaging methodologies is challenging because of the need to use a MRMC design. A case study involving suspicious breast masses is presented where blinded readers must evaluate a new technology after evaluating a control methodology. The goal is to improve specificity while sustaining near 100% sensitivity. Five mass features were prospectively defined to help readers evaluate masses with the new technology. The use of rules and regression models can help guide readers who are asked to provide a probability of malignancy estimate. The logistics and consequences of using such assistance is discussed.