Package: BayesianFROC
Type: Package
Title: FROC Analysis by Bayesian Approaches
Version: 0.1.1
Authors@R: 
    person(given = "Issei",
           family = "Tsunoda",
           role = c("aut", "cre"),
           email = "tsunoda.issei1111@gmail.com")
Maintainer: Issei Tsunoda <tsunoda.issei1111@gmail.com>
Description: For details please see vignettes in this package. This package aims to provide new methods for the so-called Free-response Receiver Operating Characteristic (FROC) analysis. The ultimate aim of FROC analysis is to compare observer performances, which means comparing characteristics, such as area under the curve (AUC) or figure of merit (FOM). In this package, we only use the notion of AUC for modality comparison. In the radiological FROC context, by a word modality we mean imaging methods such as Magnetic Resonance Imaging (MRI), Computed Tomography (CT),Positron Emission Tomography (PET). So there is a problem that which imaging method is better to detect the lesions from shadows in radiographs. To solve the modality comparison issues, this package provides the new methods using hierarchical Bayesian models proposed by the author of this package. Using this package, one can obtain at least one conclusion that which imaging methods are better for finding lesions in radiographs with the case of your data. Fitting FROC statistical models is sometimes not so good, it can easily confirm by drawing FROC curves and comparing the curves and the points constructed by False Positive fractions (FPFs) and True Positive Fractions (TPFs), we can validate the goodness of fit intuitively. Such validation is also implemented by the Chi square goodness of fit statistics in the Bayesian context in which the parameter is not deterministic, thus by integrating it with the posterior predictive measure, we get a desired value. To compare each imaging methods, i.e., modalities,  we evaluate the AUC for each modality which gives us a comparison of modalities. FROC is developed by Dev Chakraborty, his FROC model in 1989 paper relies on the maximal likelihood methodology. In this package, I modified and provided the alternative Bayesian FROC model. Strictly speaking, his model does not coincide with models in this package. I hope that medical researchers use not only the frequentist method but also alternative Bayesian methods. In medical research, many problems are considered under only frequentist methods, such as the notion of p-values. But p-value is sometimes misunderstood. Bayesian methods provide very simple, direct, intuitive answer for research questions. To know how to use this package, please execute the following codes from the R (R studio) console, demo(demo_MRMC,package="BayesianFROC"); demo(demo_srsc, package="BayesianFROC"); demo(demo_stan, package = "BayesianFROC"); demo(demo_drawcurves_srsc, package="BayesianFROC"); demo_Bayesian_FROC(). References: Dev Chakraborty (1989) <doi:10.1118/1.596358> Maximum likelihood analysis of free - response receiver operating characteristic (FROC) data. Pre-print: Issei Tsunoda; Bayesian Models for free-response receiver operating characteristic analysis. Combining frequentist methods with Bayesian methods, we can obtain more reliable answer for research questions.
License: MIT + file LICENSE
Encoding: UTF-8
LazyData: true
RoxygenNote: 6.1.1
Imports: knitr, readxl, openxlsx, xlsx, stats, graphics, tcltk,
        grDevices, ggplot2, methods, rstantools, car, hexbin, MASS,
        crayon, DiagrammeR
Suggests: rmarkdown
Depends: rstan (>= 2.18.2), R (>= 3.4.0), Rcpp
NeedsCompilation: yes
VignetteBuilder: knitr
Packaged: 2019-05-04 09:53:36 UTC; 81909
Author: Issei Tsunoda [aut, cre]
Repository: CRAN
Date/Publication: 2019-05-09 14:00:02 UTC
