How FROC data arise

What is FROC task from which data arise

There are three actors for FROC trial:

In this package, we assume that the each image specified true lesion location which are evaluated base on the gold standard and reader does not know such locations.

Flow of FROC task

Truth

Truth

Summarizing

Confidence Level No. of Hits No. of False alarms
5 = definitely present \(H_{5}=1\) \(F_{5}=0\)
4 = probably present \(H_{4}=0\) \(F_{4}=0\)
3 = equivocal \(H_{3}=0\) \(F_{3}=1\)
2 = probably absent \(H_{2}=0\) \(F_{2}=0\)
1 = questionable \(H_{1}=0\) \(F_{1}=2\)

Note that this is done only for the one image, and actual trial, there are many images, so number of hits and false alarms are more large.

The R script for the above data

 dat <- list(
#Confidence level.
c = c(5,4,3,2,1), 

#Number of hits for each confidence level.
h = c(1,0,0,0,0),

#Number of false alarms for each confidence level.
f = c(0,0,1,0,2),  

#Number of lesions
NL= 2,   

#Number of images
NI= 1,  

#Number of confidence level
 C= 3
)        

Should not run the function with this data. This data is too small to fit the model.

For multiple reader and multiple case, the notations are little troublesome, so I remain it.

FROC as Piecewise ROC;

By divide each images to the voxel of interest (VOI) or region of interest (ROI), we can obtain such FROC data. These design is very similar to the ordinary ROC analysis, since each ROI or VOI we only answer the dichotomous answer whether lesion exists or not.

Appendix ( Not required to read)

Note that I am not doctor, so the red circles never the actual lesions, it is only need to explain and I marked the red circle blindly. So, the image is probably healthy case. I do not know.

Now my health is …

Very very very bad :’-D, I have prurigo nodularis at all my body from head to leg, am I die ?? I want to found job, I had quited the previous job for my bad health, so please give me a job, help me..