The Mitotic Figure Counts repository contains the data, functions, and markdown files for a study comparing mitotic figure counting performance based on whole slide images (WSI images) from four scanners to the counts from a microscope.
Please cite the following article if data from this repository are used for your research. The article contains the details about this study.
All materials may be downloaded from this repository, the release page, or as an R package.
5 modalities : 4 whole slide imaging (WSI) scanners and 1 miscroscope.
5 observers : 5 pathologists.
4 slides==WSI images : HE-stained slides prepared from canine oral melanoma tissues that were scanned by the WSI scanners.
40 ROIs (Regions of Interest) : Ten (10) ROIs per slide. Each ROI was 200 x 200 um^2 (0.04 mm^2).
155 “candidate” mitotic figures (MF): 155 target cells from 40 ROIs.
Study is fully crossed.
A single data frame of the study data. Each row corresponds to a candidate mitotic figure and modality (155 candidates x 5 modalities = 775 rows). There is a column for each observer and the truth. This data is also included as an R object (data frame) in the R package. See Documentation for more details.
A single data frame of the mitotic figure counts per ROI and modality (40 ROIs x 5 modalities = 200 rows). There is a column for each observer and the truth. This data is also included as an R object (data frame) in the R package. See Documentation for more details.
A single data frame of the mitotic figure counts per WSI and modality (4 WSIs x 5 modalities = 20 rows). There is a column for each observer and the truth. This data is also included as an R object (data frame) in the R package. See Documentation for more details.
In this script we do an multi-reader multi-case (MRMC) analysis of the auc for each scanner. The MRMC analysis is accomplished by the OR method (Obuchowski and Rockette: Obuchowski1995_Commun-Stat-Simulat_v24p285). The function used is mrmcAnalysisOR [Documentation]. Since the data is binary, auc is the average of sensitivity and specificity or half of (Youden’s index + 1). Sensitivity is defined as the number of MFs detected by an observer divided by the number of true MFs. Specificity is defined as one minus the false-positive fraction, where the false-positive fraction is the number of false MFs that were positively marked, divided by the total number of false MFs. Furthermore, we account for the fact that there are multiple observations per case (multiple ROIs per WSI) when calculating the reader by modality covariances that are used in the OR method (clustered data: Obuchowski1997_Biometrics_v53p567). The function used is doAUCcluster [Documentation].
The ‘'’aucMRMCcluster’’’ analysis results are used to produce:
This R-markdown file combines text and R statistical analysis code for the Bland-Altman analysis and accuracy assessment performed in the paper. Both of these are MRMC analyses. The Bland-Altman analsysis makes use of U-statistics of degree 1,1 (One reader and one case). The U-statistics tools are found in the iMRMC package (CRAN package, GitHub repository)