A contaminated binormal model for ROC data: Part II. A formal model.

RATIONALE AND OBJECTIVES A contaminated binormal receiver operating characteristic (ROC) model is proposed to account for ROC data that have very few false-positive reports even though many healthy subjects are sampled. The model assumes that no signal information is captured for a proportion of abnormalities, and that these abnormalities have the same distribution as noise along the latent decision axis. MATERIALS AND METHODS The authors developed a formal psychophysical model, presented here in detail. They have specified the psychophysical assumptions of the theory, and have provided proofs that include all essential details, from assumptions to implications. With the technical details that are provided, this theory can be implemented with computer programs to fit data. RESULTS The new model can fit ROC data in which some or all of the ROC points have false-positive fractions of 0 and true-positive fractions of less than 1, without implying that performance is perfect. The resulting ROC curves are always proper, never exhibiting inappropriate chance line crossings. The model predicts that, under certain conditions, a bimodal categorical rating histogram will be observed for the signal distribution. The model predicts a relationship between the mean and standard deviation of the signal distribution and holds that, for expert decision makers, there are situations in which the prevalence and utility matrix preclude operating points in some ROC regions. The model has a straightforward extension to the joint detection and localization ROC curve. CONCLUSION The contaminated binormal model accounts for ROC data with few or no false-positive reports.

[1]  P F Judy,et al.  Visualization and detection-localization on computed tomographic images. , 1991, Investigative radiology.

[2]  J. Swets.  Indices of discrimination or diagnostic accuracy: their ROCs and implied models. , 1986, Psychological bulletin.

[3]  H R BLACKWELL,et al.  Neural theories of simple visual discriminations. , 1963, Journal of the Optical Society of America.

[4]  John A. Swets,et al.  Evaluation of diagnostic systems : methods from signal detection theory , 1982 .

[5]  R. Swensson.  Unified measurement of observer performance in detecting and localizing target objects on images. , 1996, Medical physics.

[6]  D. Dorfman,et al.  Maximum-likelihood estimation of parameters of signal-detection theory and determination of confidence intervals—Rating-method data , 1969 .

[7]  D. M. Green,et al.  Signal detection theory and psychophysics , 1966 .

[8]  C. Metz,et al.  Visual detection and localization of radiographic images. , 1975, Radiology.

[9]  C. Clark,et al.  Orthopaedics: Essentials of Diagnosis and Treatment , 1994 .

[10]  K. Berbaum,et al.  Proper receiver operating characteristic analysis: the bigamma model. , 1997, Academic radiology.

[11]  K S Berbaum,et al.  Degeneracy and discrete receiver operating characteristic rating data. , 1995, Academic radiology.

[12]  D. Dorfman,et al.  Maximum likelihood estimation of parameters of signal detection theory—A direct solution , 1968, Psychometrika.

[13]  C E Metz,et al.  The "proper" binormal model: parametric receiver operating characteristic curve estimation with degenerate data. , 1997, Academic radiology.

[14]  B. Everitt,et al.  Finite Mixture Distributions , 1981 .

[15]  C. Metz,et al.  "Proper" Binormal ROC Curves: Theory and Maximum-Likelihood Estimation. , 1999, Journal of mathematical psychology.

[16]  Kevin S. Berbaum,et al.  A contaminated binormal model for ROC data , 2000 .

[17]  K S Berbaum,et al.  A contaminated binormal model for ROC data: Part III. Initial evaluation with detection ROC data. , 2000, Academic radiology.

[18]  J SWETS,et al.  Decision processes in perception. , 1961, Psychological review.