Statistical issues in ROC curve analysis

Howard E. Rockette; Nancy A. Obuchowski; Charles E. Metz; David Gur

doi:10.1117/12.18951

1 August 1990 Statistical issues in ROC curve analysis

Howard E. Rockette, Nancy A. Obuchowski, Charles E. Metz, David Gur

Proceedings Volume 1234, Medical Imaging IV: PACS Systems Design and Evaluation; (1990) https://doi.org/10.1117/12.18951
Event: Medical Imaging '90, 1990, Newport Beach, CA, United States

Abstract

Using simulation studies and the data obtained from our clinical investigations, we are comparing some of the techniques commonly employed in ROC analysis. The general areas of our investigation include estimation of the area under the ROC curve and its associated standard error, comparison of the areas under two ROC curves, and simultaneous comparison of the areas under more than two ROC curves. For single ROC curves, maximum likelihood estimation based on the binormal model is being compared to the approximate area obtained using the Wilcoxon Statistic. Standard errors obtained from classical asymptotic theory are being evaluated and compared to standard errors based on jackknifing. With regard to the task of assessing differences in the diagnostic accuracies of two systems, our investigations have compared the maximum likelihood estimation method to a nonparametric method based on the Wilcoxon Statistic, focusing primarily on correlated observations, e.g., from readings of different modalities' images of the same patients. Extensions of existing methodology to the problem of comparing ROC areas from more than two modalities are being addressed as well as problems associated with the incorporation of multiple readers.

Citation Download Citation

Howard E. Rockette, Nancy A. Obuchowski, Charles E. Metz, and David Gur "Statistical issues in ROC curve analysis", Proc. SPIE 1234, Medical Imaging IV: PACS Systems Design and Evaluation, (1 August 1990); https://doi.org/10.1117/12.18951

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available