NEWS.md

title
NEWS

version 2.1.3

---

CRAN submission v2.1.3 - postponed until Jan 01, 2024

• Bug fix, Issue 90
• Bug fix, Issue 89
• Fixed user data-entry errors in Excel file (check column names and data type)
• bootstrap and DeLong in StORAnalysis disabled for XMod analysis
• Renamed StSignificanceTesting to St
• Renamed StSignificanceTestingCadVsRad to StCadVsRad
• Renamed UtilVarComponentOR to UtilOrVarCov
• Renamed UtilVarComponentsDBM to UtilDBMVarComp
• Removed UtilOutputReport.R as St output has been updated so this function is no longer needed.
• SPLIT-PLOT (A & C) analyses is now in one function: StSP, implemented Dec 1, 2023

Changes 05/19/2023

• Fixed FROC simulator, easier to read code; both simulators are NOT limited to I = 2.
• This affected goodValues; reinitialized affected ones.
• truthTableStr is used more consistently (except for LROC and ROI paradigms).
• TODO: see below

Possible errors in RSM predicted TPF and other cpp implementations

• dataset11 has unusual lesion distribution: there are 14 elements but they range from 1 to 20 with some lesionIDs missing
• Added test_cpp_vs_R where I compare cpp to R implementations

Changed function names to RSM_FPF and RSM_TPF

• Formerly RSM_xROC and RSM_yROC
• This renaming was missed in the earlier submission
• Further cleanup of documentation entries in rsmFormulae.R

Started new version

• 2.1.2.9000

CRAN submission

• Submitted 11/8/22 1:30 PM
• Accepted 11/8/22 2 PM
• Version: 2.1.2
• Date: 2022-11-08 18:38:38 UTC
• SHA: c3e0a0ac2eed9101ce3a0d4130103c9652930fba

After ggplot2 update to 3.4.0

• Standardized RSM exported function names: RSM_LLF, RSM_wLLF, RSM_FPF, RSM_TPF, RSM_NLF. These now match the usage in RJafrocFrocBook
• 11/7/22 took out dealing with zeroes in lesDistr vector, too complex
• Wherever geom_line() occurs, check that size aesthetic is replaced by linewidth

After Peter's fix to issue #85 due to changes in ggplot object structure 10/25/22

• The fix just checks if object is a ggplot object, not details of the object
• I checked that this works with the new version of ggplot2 installed by the following line
• devtools::install_github('tidyverse/[email protected]')
• Ran R CMD check success and merged developer into master branch

Checking all AUC related Cpp code vs. my R code 10/7/22

• Discovered no need to replace error function with Phi function implementation; this is already done in Cpp code
• Added UtilAnalyticalAucsRSM_R function which does not use Cpp code
• Took out the tempTest flag in UtilAnalyticalAucsRSM; this uses Cpp
• R CMD CHK

❯ checking dependencies in R code ... WARNING
  '::' or ':::' import not declared from: ‘R’

• Why is y_ROC_FPF_R generating a warning? This function is not called anywhere, so I commented it out; the warning disappears

Work post acceptance of v2.1.1

• universal change: lambdaP to lambda and lambda to lambda_i 9/19/22
• this corresponds with revised notation in book
• PlotRsmOperatingCharacteristics depends on RSM parameters, not intrinsic parameters 9/22/22
• Removed tests in test-LROCFomCheck.R could not resolve in time 9/22/22
• Set default lesDistr = 1 in PlotRsmOperatingCharacteristics as we don't want user to have to input this value when it is not needed for the requested operating characteristic (e.g., FROC or AFROC).

version 2.1.1

---

CRAN resubmission 8/12/22

• Version 2.1.1, accepted 8/12/22
• On cran211 branch.
• Eliminates two html rendering notes occurring on 2.1.0.
• Corrected 2 URL formatting errors, in DESCRIPTION and RJafroc-package.Rd.

version 2.1.0

---

CRAN accepted 7/24/22

• Version 2.1.0
• On cran210 branch.
• Steps to reduce file size to less than 5 Mb:
  • Removed RoiData.xlsx.
• Otherwise identical to developer and master as of 07/7/22.

Extensive changes to handling of lesDist and relWeights 7/19/22

• Removed unnecessary dimension on lesDist, it is now always 1D
• Affected files are UtilLesDistrVector.R and UtilLesionWeightsMatrix.R
• lesDistr and relWeight must have same lengths and sum to unity
• relWeight = 0 imposes equal weights
• can remove a lesion by setting the corresponding lesDist entry to zero
• need further testing on above capability

added Ch19Vig1FrocSampleSize.Rmd 6/7/22

• Added added Ch19Vig2FrocSampleSize.Rmd and Ch19Vig2FrocSampleSize.Rmd
• These vignettes got accidentally removed, perhaps because it was failing tests
• R CMD CHK works
• Found lots of instances of four backticks, instead of 3; this is probably why some vignettes were not loading on website

created DfWriteExcelDataFile 3/15/22

• needed to write Nico simulated FROC data to JAFROC format excel file
• removed JAFROC format writes capability from DfSaveDataFile; it is now in DfWriteExcelDataFile

fixed wAFROC1_dpc 3/11/22

• needed to divide final value by (K * K2)
• cpp code unaffected; this only affected R version used to debug
• checked R code vs cpp; see JT_R_Py_Foms.xlsx in PyJafrocscraps directory

added example to ChisqrGoodnessOfFit code 3/8/22

argument of St functions 1/24/22

• analysisOption must be DBM or OR

Clarified weights matrix 1/7/22

• See test-RSM-formulae.R
• Search in code for rsm-pred-wafroc-curve: rsmFormulae.R and UtilAnalyticalAucsRSM.R
• See RJafrocFrocBook, search for rsm-pred-wafroc-curve
• Push to developer and master

TBDIF 12-26-21

• To be done in future
• Created much confusion in RJafrocFrocBook, chapter on 3-fits
• RsmFormulae.R: This file is a mess.
• Remove AUCs in PlotRsmOperatingCharacteristics? - these are done in UtilAnalyticalAucsRsm
• Add to tests?
• Remove redundant column in weights matrix?

Changed RSM_yROC to accept physical parameters 12-26-21

• So as to be consistent with RSM_xROC
• Created much confusion in RJafrocFrocBook, chapter on 3-fits
• Added checks for valid RSM parameters in RsmFormulae.R .
• Changed all such functions in RsmFormulae.R to accept physical parameters.

Fixed PlotRsmOperatingCharacteristics not working for zeta1 = -Inf

• Global search string "bug fix 12/7/21" to locate all changes.
• Starting value of for-loop cannot be -Inf; detect it and set to -3

Fixed PlotRsmOperatingCharacteristics returning correct plots but incorrect AUCs

• Global search string "bug fix 11/24/21" to locate all changes.
• Basic issue was that I was using zeta1 = -20 instead of the supplied value.

Fixed Issue 73 and deprecated the MRMC file format 10-28-21 - 10-29-21

• Global search string "T1-RRRC for ROC data #73"" to locate all changes.
• Basic issue was missing truthTableStr in any file *.imrc when read by DfReadDataFile.
• See under tests: StSignificanceTestingCadVsRad: Issue T1-RRRC for ROC data #73 for recreation of this issue.
• Updated R and RStudio.
• File DfExtracDataset.R was also affected: the change fixes an error that did not get caught before.
• Updated documentation and links in StSignificanceTestingCadVsRad.R.
• Note that For non-JAFROC data file formats, the readerID and modalityID fields must be unique integers, as indicated in documentation of DfReadDataFile().
• Giving thought to removing support for all non-JAFROC formatted files; otherwise I need to maintain support for four file extensions (*.lrc, *.txt, *.csv and *.imrmc) for the simplest data structure (one rating for each modality-reader-case). This is unnecessarily complicating the code. Final resolution: I will support only *.imrmc. Other formats can still be read by DfReadDataFile() and then saved to a JAFROC format file for analysis within the RJafroc package.

Added ability to read Excel format LROC datasets 6/11/21 - 6/14/21

• Extended DfReadDataFile to accommodate LROC data; added flag lrocForcedMark
• Must use newExcelFileFormat = T for this capability
• Added toy LROC files: see inst/extdata/toyFiles/LROC/lroc*.xlsx
• See ReadJAFROCNewFormat.R, just before final return, for added code
• Added tests in test-DfReadDataFile().

Corrected sample size vignettes 4/12/21 and 4/14/21

• Ch19Vig1FrocSampleSize.Rmd and Ch19Vig2FrocSampleSize.Rmd
• Fixed SsFrocNhRsmModel.R to not return lesion distribution and weights
• Fixed vignettes that were using the old structure returned by sig. testing function
• Fixed 2 FROC SS vignettes; fixed SsFrocNhRsmModel.R to do binning internal to the function
• Added a test for SsFrocNhRsmModel().
• Updated Rcpp to 1.0.6. NOTE: version 1.0.6.6 created horrendous errors - R aborts.

Intrinsic vs. physical RSM parameters 4/2/21

• All C++ functions take physical parameters
• Rest take intrinsic parameters (2 exceptions, like RSM_xROC and RSM_pdfN)
• Cleanup:
  • PlotRsmOperatingCharacteristics.R,
  • UtilAnallyticalAucsRSM.R,
  • rsmFormulae.R
  • affected related test files: test-RSM-formulae.R and test-model-aucs.R
  • Used goodValues to check that nothing has changed

Moved to RJafrocBook 1/3/21

• Vignette Ch10Vig1QuickStart
• Vignette Ch10Vig2QuickStart
• Function Compare3ProperRocFits.R
• Associated files in inst: MRMCRuns and ANALYZED

Added functions RSM_pdfN and RSM_pdfD

• Needed for Swets predictions in book; but of general utility.
• Other new functions added of type RSM_*()
• Need to vectorize all Cpp functions; no need to carry both scalar and vector types.
• Add tests for new functions RSM_*()

version 2.0.1

---

CRAN submission process

• This is on cran3 branch.
• Steps to reduce file size to less than 5 Mb:
  • Removed tests and vignettes (this needs to be done on all computers I am using).
  • Removed all files from inst/MRMCRuns except Tony, the one that is used in an example.
  • Removed CrossedModalities.xlsx and references to it.
  • Removed DfReadLrocDataFile.R and findings.txt. Ran devtools::document() to fix NAMESPACE.
  • Removed RoiData.xlsx.
• Otherwise identical to developer and master as of 12/8/20.
• testthat failure on Ubuntu developer is resolved, see master branch: checkEnvironment = FALSE in expect_equal() on ggplot2 comparisons to goodValues.
• On CRAN

Simplify handling of lesion distribution and lesion weights

• Motivation: basically only need to specify two 1D arrays: lesDistr and relWeights.
• Since these are involved in the C++ calls, cannot totally eliminate the 2D arrays; and they are also useful for printouts.
• Eliminated two functions and much simplification:
• Eliminated UtilSpecifyLesionWeightsDistr
• Eliminated Convert2lesDistr
• Simplified system for specifying lesion distribution lesDistr
• Simplified system for specifying lesion weights distribution relWeights
• Affected many functions
• Basic idea is to keep the complexity of weights etc. concealed from the user
• Passed tests; 12/3/20

Working on analytical AUCs from RSM

• Motivation: could obviate the long simulations in CAD optimization chapter in book
• Added UtilSpecifyLesionWeightsDistr() which is distinct from UtilLesionWeightsDistr(), as the latter works on datasets.
• Observed that for equal weights AFROC and wAFROC analytical AUCs are identical. Need to think this out.
• Changed function name from UtilAucRSM() to UtilAnalyticalAucsRSM() to distinguish from UtilFigureOfMerit() which works on datasets.
• Added function Convert2lesDistr() to save me time converting from 1D lesion distribution to 2D version which is the standard in the rest of the package. It is currently not called anywhere.
• Added zeta1 dependence to UtilAnalyticalAucsRSM().
• Nov 30, 2020

Update StSignificanceTestingCadVsRad()

• Shortened the name of the function as shown above
• Needed to standardized output to avoid klutzy code in RJafrocBook
• Fixed variance components returns in DualModalityRRRC()
• Fixed CI returned object - difference was incorrect, changed sign

Modified seed behaviour, no need for SimulateFrocDatasetNoSeed()

• Needed for compatibility with plots in 14-froc-meanings-xx.Rmd chapter.
• Otherwise different random numbers are generated and it throws all the plots off.
• If seed is not supplied, then SimulateFrocDataset() behaves as before the Nov 17 change
• Nov 20. 2020

Added seed to SimulateFrocDataset()

• Ability to specify seed in order to reproduce FROC datasets.
• In book chapter 13- on effect of zeta1 on FOM and finding the zeta1 that maximizes wFROC AUC.
• Had to fix several test files.
• Consider adding seed to other simulation functions.
• Nov 17, 2020.

Fixed errors reading FROC file with no non-diseased cases

• Toy file with no non-diseased cases: frocLocatClass.xlsx.

• Symptom: UtilFigureOfMerit, with "wAFROC1" FOM failed in C++ code in function double wAFROC1() with message Not compatible with requested type: [type=character; target=double]

• The problem was traced to ReadJAFROCOldFormat.R (I was using OldExcelFileFormat) which was returning NL and LL as characters, not numerics.

• Fix: convert NL and LL from character using as.numeric.

NLRating <- as.numeric(NLTable[[4]])
LLRating <- as.numeric(LLTable[[5]])

• Then I tried NewExcelFileFormat, which gave following error: stop("Error in reading LL/TP table") Replaced code in ReadJAFROCNewFormat.R as follows:

if (is.na(tt2)) next else { # this is the change
  if (tt2 != 1)  stop("Error in reading LL/TP table") else
  # the is.na() check ensures that an already recorded mark is not overwritten
  if (is.na( LL[i, j, k, el])) LL[i, j, k, el] <- LLRatingCol[l]
}
# if (is.na(tt2)) stop("Error in reading LL/TP table") else {
#   if (tt2 != 1)  stop("Error in reading LL/TP table") else
#     # the is.na() check ensures that an already recorded mark is not overwritten
#     if (is.na( LL[i, j, k, el])) LL[i, j, k, el] <- LLRatingCol[l]
# }

Also replaced

el <- which(unique(truthTableSort$LesionID) == LLLesionIDCol[l]) - 1

with

if (K1 != 0) {
  # this gives 0,1,2,..,max num of lesions
  # which includes zero, hence the minus 1
  el <- which(unique(truthTableSort$LesionID) == LLLesionIDCol[l]) - 1
} else {
  # this gives 1,2,..,max num of lesions
  # which does not include zero, hence no minus 1
  el <- which(unique(truthTableSort$LesionID) == LLLesionIDCol[l])
}

• Added a test comparing FOMs with new and old formats for wAFROC and K1 = 0 file
• Added a test comparing FOMs with new and old formats for wAFROC and K1 != 0 file
• Added to tests: file test-UtilFigureOfMerit.R
• Ran R CMD check

Fixing non-character input error

• Finished October 23, 2020.
• Fixed error in DfReadDataFile.R in function checkTruthTable(); this bug discovered in working with HUGE one reader dataset.
• For single reader or single modality dataset, if input is not converted to character, then error results in I <- length(strsplit(modalityIDCol[1], split = ",")[[1]]) and similar expression for J.
• Fix is to use as.character as in:

  readerIDCol <- as.character(truthTable$ReaderID) # bug fix to avoid non-character input error below
  modalityIDCol <- as.character(truthTable$ModalityID) # do:

Simplified plotting routines

• Finished October 2, 2020.
• Necessitated by work on CAD vs. RAD plots in RJafrocBook.
• PlotEmpiricalOperatingCharacteristics.R.
• Especially function gpfPlotGenericEmpiricalOperatingCharacteristic.
• The revisions are tested via R files in inst/fixPlots.
• I am following the help page of ggplot2
• Did away with with function usage in this function: hard to tell what is going on and the help page on this function seems to discourage this type of usage in packages
• But, get silly NOTE about undefined global variables: genAbscissa, genOrdinate, Reader, Modality.
• These are members of a dataframe, so I dont see why they are visible at global level.
• Does not happen with the other dataframes in this packages.
• So I assume it is a ggplot2 related issue.
• I solved it by initializing these at the very beginning of the gpfPlotGenericEmpiricalOperatingCharacteristic function to NULLs.

Added extensive comments in StORSummaryFRRC.R

• Added extensive comments in StORSummaryFRRC.R on how I am calculating CI for individual treatments averaged over readers;
• The method was "reverse-engineered" from inst/Iowa/VanDyke.txt, as I cannot find a better reference for the equations used.
• This created debugging problems as break points did not work; added # at the end of each inserted commented line, 86-103; this seemed to solve problem;
• passes R CMD check

Updated package documentation and Ch00 Vignettes

• Commented out examples in DfSaveDataFile.R as it creates non-standard files in doc directory; this did not happen before the R update.
• Also, the ggplot output structure appears to have changed; had to regenerate goodvalues in test-PlotEmpiricalOperatingCharacteristics.R.
• Need to update documentation and DESCRIPTION.
• Need to use artificial intelligence instead of CAD as this is the new thing.
• Moved split-plot vignettes to Dropbox; cleaned up Ch00 vignettes - RocFctrl, FrocFctrl, RocSpA and FrocSpA
• Moved description of other's data formats to Dropbox.

Added stats to ORAnalysisSplitPlotA

• ORAnalysisSplitPlotA returns list containing FOMs, ANOVA and RRRC
• Implements formulae in Hillis 20-14 paper on page following Table VII.
• Tested with toy file and collaborator dataset
• Merged to developer branch and deleted SplitPlotA branch
• ran pkgdown::build_site()
• Updated R and RStudio

Handling of FOMs that depend on single-truth-state cases

• Some confusion in my mind about handlling of normal case only or abnormal case only FOMs, like MaxNLF, MaxLLF, HrSe, HrSp, etc. Resolved after studyin XZ code, StOldCode.R. The handling is shown in UtilPseudoValues.R. I believe it is now correct. For MaxNLF, HrSp, and ExpTrnsfmSp the relevant number of cases is K1, for MaxLLF and HrSe it is K2 and for all the rest it is K.

Revised UtilPseudoValues 8/7/20

• More compact code handles all FOMs - no exceptions, as before, for MaxLLF, etc. Extensive simplification to accomplish handling of different FOMs.
• Modified FOMijk2VarCovSpA and FOMijk2VarCov to accept a varInflFactor logical argument, allowing jackknife, bootstrap and DeLong - based estimates to be more compactly handled.
• UtilPseudoValues handles all designs without resorting to separate functions for FCTRL, SPLIT-PLOT-A and SPLIT-PLOT-C
• Modified frocSpA toy dataset to stress code (unequal numbers of abnormal and normal cases, multiple marks on NL and LL worksheets, etc)
• Checked dataset05 MaxLLF MaxNLF vs. JAFROC
• See below, note added today relating to handling of descriptions$fileName. This fixed problem with expect_equal() failing depending on how the goodValues were generated - from R command line vs. Run Tests. Also, in creating a dataset object, where appropriate, fileName <- "NA" instead of fileName <- NA; the latter generates a character expected error when an attempt is made to strip path name and extension in convert2dataset and convert2Xdataset.
• Updated and reorganized tests
• Implemented SPLIT-PLOT-A analysis for unequal numbers of readers in the two groups. The formulae (from Hillis 2014) are modified to use modality-specific components, i.e. Var_i, Cov2_i and Cov3_i. The modified formulae reduce to Hillis' formulae when the number of readers in each group are identical. Communicated results to collaborator.
• Corrected error in handling of MaxNLFAllCases FOM; see comments in UtilMeanSquares(); regenerated one goodValue file.

Read real SPLIT-PLOT-A dataset

• Did not find any data entry errors in /toyFiles/FROC/1T3Rvs4R.xlsx
• Simplified rdrArr handling: this is done in checkTruthTable, where SPLIT-PLOT-A is handled separately;
• Added source fileName to descriptions$fileName field of DfReadDataFile() return; this will keep a record of how the dataset was generated
• Note added 8/7/20: above change created problems passing tests in R CMD check, as long file names may not agree; simplified to remove all but the file base name; regenerated many goodValues files

Update for reading SPLIT-PLOT-A data files

• After emails with collaborator, need for this type of analysis.
• Need to comment DfReadDataFile.R and ReadJAFROCNewFormat.R and add more checks in the code for illegal values.
• Indiscriminate sorting introduced problems; now, sorted caseID column is used now in only 3 places to find the correct case indices, where normal cases are ordered first, regardless of how they are entered in the Truth worksheet:

k <- which(unique(truthTableSort$CaseID) == truthTable$CaseID[l])
k <- which(unique(truthTableSort$CaseID) == NLCaseIDCol[l])
k <- which(unique(truthTableSort$CaseID) == LLCaseIDCol[l]) - K1

Tests for UtilOutputReport

• Included tests for UtilOutputReport() for text output only
• Could not get version that compared actual outputs to work in R CMD check.
• Only works at command line (running tests one at a time)
• Failing code is commented out

Renamed SP datasets 7/9/20

• Renamed the split plot dataset to datasetFROCSpC
• The C stands for Table VII Part (C) in Hillis 2014
• Need to distinguish between different types of split-plot datasets
• What I was doing so far was split-plot-c
• The new analysis requires split-plot-a

Code for checking for non-sequential lesionIDs in TRUTH 7/8/20

• Updated all tests to display contexts more consistently.
• Moved unnecessary files from inst to Dropbox.
• Main work was on DfReadDataFile.R and testing code for non-sequential lesionIDs in TRUTH worksheet.
• Thinking about modifications to handle split-plot data along the lines of HYK datasets
• Also learning about grep.

Implemented extensive testing comparing RJafroc to Iowa software

• See test-StCompare2Iowa.R in testthat directory.
• Does line by line comparison of RJafroc to results of OR-DBM MRMC 2.51 Build 20181028 for VanDyke and Franken datasets.
• These, and many other tests, are run automatically every time the RJafroc software is checked using R CMD check.

Note on discrepancy vis-a-vis Iowa software

• Noted was a discrepancy between Var(R) and Var(TR) values reported by OR-DBM MRMC 2.51 Build 20181028 and RJafroc for Franken dataset.
• Their code does not implement the required max(Cov2 - Cov3, 0) constraint while RJafroc does.
• RJafroc reports VarTR = -0.00068389146 while their code reports VarTR = -0.00071276.
• Specifically, msTR - Var + Cov1 + max(Cov2 - Cov3, 0) = -0.00068389146 and msTR - Var + Cov1 + Cov2 - Cov3 = -0.00071276.
• This also affects the VarR values (see block of comments in UtilORVarComponentsFactorial near line 161). Cov1, Cov2, Cov3 and Var are the same between both codes.
• I am aware that these discrepancies do not affect sample size estimates, but can cause confusion for the code maintainer and the end user.

Updated sample size routines

• 3 Papers by Hillis et al on SS estimation for ROC studies: 2004, 2011 and 2018
  • Hillis SL, Berbaum KS (2004). Power Estimation for the Dorfman-Berbaum-Metz Method. Acad Radiol, 11, 1260--1273.
  • Hillis SL, Obuchowski NA, Berbaum KS (2011). Power Estimation for Multireader ROC Methods: An Updated and Unified Approach. Acad Radiol, 18, 129--142.
  • Hillis SL, Schartz KM (2018). Multireader sample size program for diagnostic studies: demonstration and methodology. Journal of Medical Imaging, 5(04).
• The sample size routines have been updated to reflect the most recent publication. The equations in the code now have documentation as to their source(s).
• The routines have also been checked using the Java calculator provided on the U of Iowa website (pss20190918.jar).
• Briefly, the procedure defaults to the OR method, even when DBM variance components are provided, in line with the recommendations in the 2011 paper. For continuity with prior work a LegacyCode flag is provided to force execution of the original DBM method (2004 paper).
• Added a few tests using the Franken dataset which yields negative Var(TR). Noticed a small difference in predicted power between forced DBM (0.78574588) and OR methods (0.8004469).
• Updated source of equations in SsPowerGivenJKDbmVarCom.
• The code rewrite was conducted on a new branch, UpdateSsRoutines off the developer branch.
• The changes were merged to the developer branch and then to the master branch.

Major reorganization of dataset structure

• Instead of varying lengths for ROC/FROC, LROC and SPLIT-PLOT, all datasets now are lists of length 3, with each member (ratings, lesions, descriptions) consisting of sub-lists:
  • ratings contain 3 elements: $NL, $LL and $LL_IL.
  • $lesions contains 3 elements: $perCase, $IDs and $weights.
  • $descriptions contains 7 elements: $fileName, $type, $name, $truthTableStr, $design, $modalityID and $readerID.
• This considerably simplified the handling of different types of datasets.
• The version number will be bumped to 2.0.1 on final submission to CRAN.
• Since there are no downstream dependencies, I feel this big change is justified at this time. It will make it easier for me to maintain the code.
• The code rewrite was conducted on a new branch, SimplifyDatasets off the developer branch.
• The changes were merged to the developer branch and then to the master branch.

Major simplifications to all significance testing St functions

• Separated RRRC branches etc to separate files; likewise for DBM and OR branches, now the files are much shorter and easier to maintain
• Changed returned data structure to a list of dataframes, see next comment; this makes for much cleaner and easier printing
• Consistent returned objects from all St functions: list with data frames FOM, ANOVA, RRRC, FRRC and RRFC.
• Output now closely follows that of Iowa software OR-DBM MRMC 2.51 which I ran using VmWare, Windows XP; Iowa software did not run on Windows 8 on two different machines (see below)
• Ran detailed comparison to OR-DBM MRMC 2.51 and coded the checks in tests/testthat/test-St-Compare2Iowa.R for VanDyke dataset
• Also visually compared RJafroc results against OR-DBM MRMC 2.51 for dataset04 converted to ROC (see Iowa code results in inst/Iowa/FedRoc.txt);
• Shortened UtilOutputReport considerably, by using print(dataframe) instead of reading values from list or dataframe variables and then using sprintf with unreadable C-style format codes
• Shortened SPLIT-PLOT analysis by returning Cov2 = Cov3 = 0 instead of NA
• Confirmed that this gives same results as the version in master branch
• Added vignettes back, rebuilt website
• Passes all OSX checks except for file size NOTE: checking installed package size ... NOTE installed size is 16.7Mb; sub-directories of 1Mb or more:doc 12.7Mb;extdata 1.3Mb
• Marked regions of code requiring further inspection by TBA - handle this next
• Will merge this code into master branch if it passes Travis

Replaced stringsAsFactors = FALSE everywhere data.frame is used except ...

• Except in PlotEmpiricalOperatingCharacteristics, where factors and levels are used
• Must specify this anytime the variable is used in a test as otherwise different versions will give factors or characters and not match those in goodValues folder
• After old versions are phased out, this can be safely removed, as all versions will have this as the default
• Setting options(stringsAsFactors = FALSE) at beginning of a function, e.g., StSignificanceTesting, passes this option onto called functions, e.g, StDBMHAnalyis
• Use getOption("stringsAsFactors") to determine state of this option; must restart R to obtain default value (TRUE on the release version); calling a function that sets it to FALSE keeps the new value after exiting from function; must restart R again to get proper default.

After repeated Travis failures

• same issue; have to specify stringsAsFactors explicitly for each data.frame call, due to different defaults in different verions of R
• Cannot set to TRUE at beginning of function, as in options("stringsAsFactors" = TRUE), as this is deprecated
• Basically undid all changes in next note (see below)
• passes R CMD check on OSX and Travis checks

Removed stringsAsFactors arguments in all calls except...

• In all calls to data.frame, except in plotting functions, PlotEmpiricalOperatingCharacteristics, where factors are used
• In current R3.6.3 option($stringsAsFactors) = TRUE (stringsAsFactors is case sensitive!)
• This means that functions that don't require factors, such as SsPowerTable() should set options(stringsAsFactors = FALSE) explicitly at the beginning of the code, in order to work in current and previous version of R (i.e., release and old-release)
• Functions that do use factors/levels such as plotting functions (PlotRsmOperatingCharacteristics and FitCorCbmRoc) should set options(stringsAsFactors = TRUE) in order to work in developer version of R, where default is options("stringsAsFactors") = FALSE
• This is all very confusing as I am having to dive into code written 6 years ago by someone else
• Going to test on Travis now

Return transpose for foms member of StSignificanceTesting return object

• For consistency with OR DBM MRMC 2.51, and also for cleaner output, as number of treatments is usually less than number of readers
• Note added 5/30/20: this was undone; transpose is no longer used

Moved official good value files to Dropbox

• They generate non-portable file name warning on R CMD check
• Moved VanDyke results to inst/IowaResults/VanDyke.txt

Compared to latest official code

• mrmc_setup_w10_July_2019.exe; VanDyke VanDyke.lrc dataset; Dropobox/IowaSoftware/VanDyke.lrc
• OR DBM MRMC 2.51 <beta> Build 20181028 </beta> miplmrmc
• Software only runs under Windows XP
• Tried Windows 8 on different machines (iMac and MacBookPro) under VmWare Fusion; no luck, even after following directions twice on website
• Need to compare OR ouputs - WIP
• Need to fix documentation on StSignificanceTesting - WIP

Discovered error

• For StSignificanceTesting(dataset02, method = "OR", option = "FRRC") - done
• Need to put in testthat all combinations of method and option - done
• Different objects returned by StSignificanceTesting depending on choice of option - almost done
• Need to standardize as otherwise RJafrocBook is klutzy - WIP

Added seed specification to UtilVarComponentOR

• Added seed specification to UtilVarComponentOR to allow comparison with RJafrocBook
• Added to tests: covEstMethod = jackknife, bootstrap and DeLong for two datasets
• dataset02 and dataset04 converted to ROC
• Will merge to master so that RJafrocBook code passes Travis

Fixing significance testing with independent calculations in RJafrocBook

• Need to modify RJafroc to eliminate code duplication and improve style in all significance testing functions - move this to issues
• I am only getting to understand it now (as I work on RJafrocBook)
• One reader case can now be handled by StSignificanceTesting(rocData1R, FOM = "Wilcoxon", method = "OR")
• May not need StSignificanceTestingSingleFixedFactor which currently only handles DBM method - add to issues
• Removed restriction of StSignificanceTesting to J > 1
• Will merge to master so that RJafrocBook code passes Travis

Fixed error with msTC

• Found another error in msTC calculation in UtilMeanSquares
• Was trying to be too cute for my own good (collapsing two for-loops into one)
• Discovered error while doing first principles calculation in RJafocBook, DBM chapter
• Changed covEstMethod argument to OR method to lower case ("jackknife" or "bootstrap")

Fixed issue with optim when flipping groups

• see issues #50 (closed) on RJafroc/master
• Allow a paramter of binormal model to be between -4 and 4
• Removed all vignettes; since these are now in RJafrocBook
• Removed all but one dataset (FZ_ALL.xlsx) from extdata/datasets so I dont get file size error (extdata was 2.5 MB, reduced to 1.3)

version 1.3.2

---

After work on cran2-update work

• Copied altered files from R directory (which commented out examples which were taking lots of CPU time)
• Includes a correction to Compare3ProperRocFits.R
• Pulled out all vignettes (these have been moved to repository RJafrocBook)
• This gets file size below 5 MB

Work post acceptance of v1.3.2, as of 3/7/20

• Going back to work interrupted by having to fix the errors on R-devel, see next section below.
• This is v1.3.2.9000.
• Got all tests working! Resulted in fix to StDBMHAnalysis.R that fixed test that I had to skip on mac for context("SignificanceTestingAllCombinations"). Need to get this fix (lines 45-51) over to cran2 branch as I am thinking of splitting the package up by separating the cran2 branch as the base package RJafroc and depending on RJafroc for new package RJafroc2. This would solve the file size problems that I am running into. Just an idea.
• Current file size is 18.4 Mb!
• Synced with developer branch on GitHub and merged with master.

After email from Kurt Hornik [email protected]

• Created new branch off cran2 1.3.1 called cran2-fix
• Bumped version to 1.3.2
• RJafroc failing on Linux
  • r-devel-linux-x86_64-debian-clang
  • r-devel-linux-x86_64-debian-gcc
  • r-devel-linux-x86_64-fedora-clang (this showed up post email)
  • r-devel-linux-x86_64-fedora-gcc (this showed up post email)
• Has to do with new default (R 4.0.0) for options(stringsAsFactors = FALSE)
• To recreate this problem in R CMD check I set options(stringsAsFactor=FALSE) near beginning of each plotting function (3 functions) using data.frame() and levels() to convert strings to factor levels
• To make problem go away I explicitly specified stringsAsFactor=TRUE in each call to data.frame() where necessary.
• Removed examples from FitCorCbmRoc() as they were generating excessive CPU time NOTES. Will need to add these to vignettes, later.
• Ran R CMD check successfully
• Ran all checks in ScriptsForCranSubmission.R
• Submitted to CRAN
• Accepted by CRAN
• See cran2-update/master branch for content relating to this version

Modified UtilPseudoValues.R to work with SPLIT-PLOT data

• Created simulated SP datafile inst/extdata/toyFiles/FROC/FrocDataSpVaryK1K2.xlsx.
• Created simulated SP dataset datasetFROCSp corresponding to modalities 4,5 of dataset04
• Update vignette Ch00Vig5SimulateSplitPlotDataset.Rmd.
• Modified StORHAnalysis.R and to work with SP-A dataset provided method = "OR" and covEstMethod = "jackknife" is used
• Corrected an error in analysis; see ~Dropbox/RJafrocChecks/StfrocSp.xlsx for details.
• Updated this file 2/19/20
• R CMD check successful ... except for file size NOTE (18.4Mb)

Created split plot dataset; update all datasets; confirm truthTableStr and DfReadDataFile()

• v1.3.1.9000
• created simulated split plot Excel dataset from Fed dataset: Ch00Vig5CreateSplitPlotDataset.Rmd; confirmed it is read without error!!
• updated datasets - see inst/FixRJafrocDatasets/ConvertDataset.R; worked on DfReadDataFile function
• Discoverd that .xls input does not work*; took it out as an allowed option; probably has to do with openxlsx
• checked truthTableStr with a data file that has only 1 and 3 lesions per case; was concerned about 4th dimension of truthTableStr; see Dropbox/RJafrocChecks/truthTableStr.xlsx for results of checks; note that fourth dimension will be 4, even though there are no cases with 2 lesions
• I think I need a separate vignette on truthTableStr - more for my sake
• added raw excel file datasets corresponding to included datasets to inst/extdata/datasets; found missing file SimulateFrocFromLrocDataset.R - not sure why I took it out;
• Added sheet AnnK to truthTableStr in Dropbox/RJafrocChecks
• Also tests that OldFormat file when read creates identical dataset to that created by NewFormat: basically two Excel fiies are identical except old format lacks the three extra columns; see checkDfReadDataFile.R
• Modified UtilFigureOfMerit to accomodate split plot dataset with varying number of cases for each reader
• Created a datafile inst/extdata/toyFiles/FROC/FrocDataSpVaryK1K2.xlsx that really exercises the DfReadDataFile function (case index is unsorted); resorted to data frames and sorting to successfully read it (it is used in three places - truthTableStr, NL and LL). See inst/extdata/testUtilFigureOfMerit/*.R for exercising files
• Need to include this file in tests
• Updated this file 2/10/20
• R CMD check successful

version 1.3.1

---

Extended dataset object structure

• Bumped version number to 1.3.1 after corrections to DESCRIPTION file
• Version on CRAN is 1.3.1

Extended dataset object structure

• Bumped version number to 1.3.0 as I am moving towards a CRAN submission
• Lost development branch while using GitHub; decided to do Git manually
• Why is .gitignore not working?
• Additional members added 12/27/2019 by DPC
• Ann discovered bug in code: does not handle single reader properly
• Ann uncovered another bug in code: did not handle diseased cases first in Excel Truth sheet
• Both bugs have been fixed
• Make it easier to correlate the NL and LL values with those in the Excel file and catch data entry errors in DfReadDataFile()
  • design = design,
  • normalCases = normalCases,
  • abnormalCases = abnormalCases,
  • truthTableStr = truthTableStr
• Need to update all datasets and check all occurences where DfReadDataFile() is used
• Included my own CPP coded wAFROC plot function. Learning a lot form Dirk's book website https://teuder.github.io/rcpp4everyone_en/.

Split plot dataset

• Modifications to DfReadDataFile() to allow for split plot datasets completed.
• Must use newExcelFileFormat = TRUE as otherwise the code defaults to the old Excel format.
• The new format includes more stringent tests, IMHO, to catch data entry errors:
• TruthTableStr is created in checkTruthTable() which is used in subsequent read NL and LL worksheets.
• Work to be done to include split plot in significance testing.
• Corrected dataset03 which had -Infs for 1-ratings; need to check other ROC data files.
• Added vignettes describing data format using toyFiles and use of DfReadDataFile().
• Corrected error in old DfReadDataFile function.
• Passes R CMD check with file size note.

Error in MS_TC corrected

• Noted by Erin Greco
• The missing "-1": UtilMeanSquares() line 88 msTC <- msTC * J/((I - 1) * (Ktemp - 1)) has been corrected
• Reset goodValues values in test-StSignificance-testing.R at line 128

Extended plotting function to LROC data

• PlotEmpiricalOperatingCharacteristics() now accepts ROC, FROC and LROC datasets.
• Simplified code.
• Included in unit tests.
• Added legend.position argument to allow better positioning of legend.

Added FROC sample size vignettes and functions

• Ch19Vig1FrocSampleSize.Rmd: Compares FROC power to ROC power.
• Ch19Vig2FrocSampleSize.Rmd: FROC power calculation for a number of situations.
• SsFrocNhRsmModel(): constructs an RSM-based model, which allows one to relate an ROC effect size to a wAFROC effect size, and returns parameters of model to allow FOM estimation for ROC and wAFROC. Following functions are used to calculate the lesion distribution and lesion weights arrays:
• UtilLesionDistribution: renamed to UtilLesionDistr
• UtilLesionWeightsDistr:

Significance testing functions

• StSignificanceTesting(): corrects errors affecting method = "OR" and covEstMethod = "Jackknife". I messed up while trying to simplify XZ code. It calls:
• StDBMHAnalysis():
• StORHAnalysis():
• Ran Windows JAFROC on virtual Windows 8 machine and saved results (inst/VarCompDiscrepancy/includedFrocData_Inferred_ROC.txt) to validate current significance testing functions. Included unit tests in tests/testthat.
• Ran first XZ CRAN upload (version 0.0.1) code (StOldCode.R) to compare against current significance testing code. Included unit tests in tests/testthat.
• test-St-Compare2JAFROC.R: compares current code to Windows JAFROC results.
• test-St-Compare2Org.R: compares current code to RJafroc 0.0.1.
• test-St-CompareDBM2OR.R: compares current code DBM to current code OR results, when appropriate.

CAD and LROC

• gpfMyFOM(): interpolation error in LROC PCL and ALROC FOMs. Hand calculations showed that the approx function did not work for small datasets. Wrote my own simple interpolation code. See LrocFoms() in gpfMyFOM.R. See ChkLrocFoms.xlsx in inst/StSigTesting for details on hand calculation of LROC FOMs.
• LROC FOMs now apply to UtilFigureOfMerit() and all significance testing functions. These changes only affected values at small FPFValue, 0.2 or less.
• Most FOM related functions now accept FPFValue to accommodate LROC datasets.
• StSignificanceTestingCadVsRadiologists(): CAD results updated (only values for FPFValue 0.2 or less were affected); see CadFunctionTests.R in inst/CadTesting. See CadTestingNicoData.xlsx in inst/CadTesting. Included unit tests in tests/testthat.
• StSignificanceTestingCadVsRadiologists(): cleaned up and now runs all FOMs.
• SimulateLrocDataset(): FROC to LROC simulator based on RSM. Could be used for NH testing. RSM can now predict all paradigm data.
• DfFroc2Lroc(): Simulates an "AUC-equivalent" LROC dataset from an FROC dataset. This is neat!
• DfLroc2Froc(): Simulates an "AUC-equivalent" FROC dataset from an LROC dataset.
• DfLroc2Roc(): convert LROC dataset to ROC dataset.
• An error in dataset2ratings() has been corrected.

Variance component input

• SignificanceTesting functions now accept variance components, without having to specify a dataset.

Other affected functions and new functions:

• UtilVarComponentsDBM():
• UtilORVarComponentsFactorial():
• SsPowerGivenJKDbmVarComp:
• SsPowerGivenJKOrVarComp:
• SsSampleSizeKGivenJ:
• SsPowerGivenJK:
• StSingleTreatmentRandomReader:
• Ensured that FPFValue argument immediately follows FOM, where applicable.

Needs further testing

• StSignificanceTestingSingleFixedFactor:

Extensions needed

• The addPlot routine in StSignificanceTestingCadVsRadiologists has been renamed to CadVsRadPlots(). It should be deprecated in future as PlotRsmOperatingCharacteristics() has more consistent visual output (and capabilities like handling lists of treatments and readers).
• Need a function that checks validity of FOM for dataset: isValidFom?
• Need to compare predicted curves for LROC and FROC data: does SimulateLrocDataset() predict both flattening out of LROC plot and wAFROC going to (1,1)?
• Split plot analysis

version 1.2.0

---

• Corrected all references to package name to RJafroc (note capitalization)
• Checked downstream dependencies - none as of July 23, 2019: revdep("RJafroc") yields character(0)
• Corrected error that was causing Solaris failure (Peter Philips)
• Corrected error in UtilPseudoValues.R that was caught by testthat
• Corrected StSignificanceTesting.R that was caught by testthat (Peter Philips)
• R CMD check generates testthat failure when run under RStudio, see following output, but not when run as devtools::test():

* checking tests ...
 Running ‘testthat.R’ [158s/160s]
 ERROR
 Running the tests in ‘tests/testthat.R’ failed.
 Last 13 lines of output:
   Component "Source": Attributes: < Component "levels": 3 string mismatches >
   List member = 2, Dataset = dataset02, FOM = Wilcoxon, method = DBM

   ── 2. Failure: SignificanceTestingAllCombinations (@test-significance-te
   CurrentValues[[listMem]] not equal to GoodValues[[listMem]].
   Component "Source": Attributes: < Component "levels": 3 string mismatches >
   List member = 2, Dataset = dataset05, FOM = HrAuc, method = DBM

version 1.1.0

---

• Added travis-ci testing after each push; and build passing badges, etc.
• Removed dependence on caTools package, which was not being supported; extracted function trapz() from it and inserted directly into gpfMyFOM.R - see comments in that file of what led to this
• Removed dependence on xlsx package, which requires rJava and JAVA, replaced with dependence on openxlsx package. Was having difficulty installing rJava correctly after each OSX or R update.
• Corrected errors in UtilOutputReport.R.
• Fixed bug in UtilOutputReport that was preventing overwriting of existing output file, even when the user keys "y" in response to prompt
• Added correlated contaminated binormal model, CORCBM, fitting and related functions to make package current with 2017 CORCBM publication.
• Fixed error in PlotEmpiricalCharacteristics.R that was giving incorrect plots for other than ROC and wAFROC plots
• Added ChisqrGoodnessOfFit function, replacing 3 functions
• Cleaned up plotting code; using one function genericPlotROC.R instead of 3 functions
• Updated results of CBM, PROPROC and RSM fitting after discovering error in df calculation in RSM chisquare statistic; book results are wrong; only 2/236 fits yield a valid chisquare statistic
• Renamed ExampleCompare3ProperRocFits() to Compare3ProperRocFits()
• Corrected overwriting error in value returned by Compare3ProperRocFits()
• Added two vignettes: QuickStartDBMH and QuickStartDBMHExcelOutput
• Checked downstream dependencies - none as of Nov 11, 2018: revdep("rjafroc") yields character(0)

version 1.0.2

---

• StSignificanceTestingCadVsRadiologists was not working for different numbers of readers. As noted by Alejandro, the number of readers was hard coded. Fixed this and extended DfExtractDataset to include LROC datasets.
• Removed function SsFROCPowerGivenJK: FROC power is implemented in Online Appendix Chapter 19 (see email exchange with Kota Aoyagi)
• This version installed on SOLARIS!

version 1.0.1

---

• Package was not installing on Solaris - overloading errors. Changed sqrt(2) in RsmFuncs.cpp to sqrt(2.0). However, Solaris is incompatible with ggplot2; so will recommend that Solaris version not be distributed on CRAN.
• Sorry, but I'm not sure what's different between the CRAN Solaris machine and R-hub's Solaris machine. You could prepare a new package submission for CRAN with the caveat that, since you do not have access to a Solaris machine, your fix is speculative and may yet fail to compile on the CRAN Solaris machine.
• The CRAN Repository Policy (https://cran.r-project.org/web/packages/policies.html) also states:
  Package authors should make all reasonable efforts to provide cross-platform portable code. Packages will not normally be accepted that do not run on at least two of the major R platforms. Cases for Windows-only packages will be considered, but CRAN may not be the most appropriate place to host them. So you could in theory argue your case that your package does not support Solaris, and request that CRAN not distribute your package on that platform. But given that the issue you're bumping to is (not) documented explicitly in the R manuals, I'm not sure how much success you would have.

version 1.0.0

---

• Renamed functions for better organization;
• Removed shiny GUI interface
• Support for LROC datasets and cross-modality datasets
• CAD vs. radiologist analysis, both single modality and dual modality

version 0.1.1

---

• An error in the p value calculation that gave incorrect p value (possibly exceeding one) when the first modality performed better than 2nd has been fixed. Thanks to Lucy D'Agostino McGowan for pointing out the error and the fix. This error, which does not occur in Windows version of JAFROC V 4.2.1, was not noticed as in all example files the 2nd modality performed better.

version 0.1.0

---

• A "shiny" based GUI has been added, accessed by the function RJafrocGui(). This allows a user only interested in analyzing a data file to access the underlying code in a "user friendly" way. The GUI is similar in functionality to that of Windows JAFROC 4.2.1 software.

• For the curve plotting functions, legend position and direction are automatically decided if they are not explicityly specified.

• The the output number of significant digits for statistical power in power table has been set to 3.

• Variance and covariance calculation error for ROI data has been fixed.

• A bug in the JAFROC data reading function that caused an error when encountering non-numeric values has been fixed.

• Floating point ratings are rounded to 6 significant digits when saving a dataset in JAFROC format.

• A bug in the plotting routine that affected plots for a single rating FROC dataset has been fixed.

• A bug in the plotting of AFROC curves for a dataset containing only non-diseased cases has been fixed.

version 0.0.1

---

• Original version posted to CRAN (by Xuetong Zhai)