Particular studies differentiating the type of tissue
ii<-F==0; X0=X[ii,];Y0=Y[ii]
```
14 Will-be-set-by-IN-TECH

scheme is competent to complement the radiologists' efforts in their daily clinical practice, to help them as second readers in the interpretation of mammograms, and also to improve

Future work will address the issue of reducing the number of false positives, without decreasing the sensitivity, by applying different statistical models to our dataset. Another way to improve the results yielded by the CAD system would be to deal with a higher quality of the digital images, because subtle clusters can be more easily identified in the detection process if the image is digitized at a greater resolution. Thus, to further improve the sensitivity, high

This work has been partially supported by Ministerio de Ciencia e Innovación, MTM2008-01603/MTM. Roca-Pardiñas' research was supported by grant MTM2008-03010 from the Spanish Ministry of Education & Science, and by grants PGIDIT07PXIB300191PR and PGIDIT10PXIB300068 PR from the Galician Regional Authority (Xunta de Galicia).

The R code developed for calculating probability values for the GLMs and GAMs, as well as for obtaining the corresponding AUCs and plots is now given and explained. The starting point of this routines considers data covariates stored on the **X** covariates, while *Y* indicates if the data corresponds either to a true detection (*Y* = 1) or to a false positive cluster (*Y* = 0). The factor considered in the study is *F* = 0 for fatty tissue, and *F* = 1 for dense tissue.

The steps followed in the present work were the calculation of the correlation values, and the study of covariates, initially without taking into consideration the type of tissue, calculating in this situation the linear and additive models, employing the functions stepGLM2 and stepGAM2 respectively (described in this Appendix), and obtaining for both cases the corresponding areas uner the ROC curve, using the empiricROC function. Plots and tables

The same analysis was next calculated, but separating the data by the type of tissue where they were embedded in the breast. Graphs, tables and Auc were also obtained. The R code is

##################################################################

################################################################## # Selection of covariates employing the AUC criterion for the GLMs

##################################################################

GLMresults=stepGAM(X,Y,"glm"); GAMresults=stepGAM(X,Y)
