Data analysis for detection and localization of multiple abnormalities with application to mammography

Area of Science:

• Medical Imaging Analysis
• Diagnostic Accuracy Studies
• Reader Performance Evaluation

Background:

• Assessing diagnostic accuracy requires evaluating both detection and localization of multiple abnormalities.
• Existing methods for reader performance studies have limitations in handling multiple errors per patient.

Purpose of the Study:

• To develop and compare a novel regions of interest (ROI)-based approach for assessing reader diagnostic accuracy.
• To evaluate the ability of the new method to detect and correctly locate multiple abnormalities.

Main Methods:

• The new approach partitions images into ROIs, with readers assigning confidence scores to each.
• Statistical methods for clustered data are employed to assess and compare reader accuracy.
• The method was applied to a study of detecting malignant breast cancer lesions in film and digitized images.

Main Results:

• The ROI-based approach, free-response receiver operating characteristic (FROC), and patient-based methods differ in false-positive rate (FPR) estimation.
• The patient-based approach inadequately handles multiple errors within a single patient.
• The FROC approach presents challenges for statistical evaluation.

Conclusions:

• The ROI-based approach effectively captures both detection and localization tasks.
• ROI-based accuracy measures are simple to interpret and clinically relevant.
• Statistical methods exist for estimating and comparing ROI-based accuracy.