search

srikanthguduru / imrmc

Automatically exported from code.google.com/p/imrmc
0 stars 0 forks source link

Use 'color scales' instead of device as modality #27

Open GoogleCodeExporter opened 9 years ago

GoogleCodeExporter commented 9 years ago 
My experiment tests 3 color scales in 4 display devices.  We want to do the 
variance analysis using color as a 'modality'. Each color scale would be a 
modality. Do you think this approach would be appropriate? 

If we were to do it, we have to overcome the fact that the software does not 
allow more than one experiment result per reader/modality and we would then 
have up to 4 results from the same reader/modality. One idea was to consider 
the same set of cases as different when read with the different devices. Again, 
do you think this would be correct?

Original issue reported on code.google.com by silv...@gmail.com on 4 Mar 2015 at 12:28

GoogleCodeExporter commented 9 years ago 
I don't think there is any problem treating the color scales as a modality. 
What I hear you saying is that you want to pool the observations across 
devices. I don't have an opinion regarding whether or not this is appropriate 
or not. There are tests out there to determine if the data is poolable, but 
this is not my expertise. 

Pooling the data would yield multiple observations per reader/case/modality. 
iMRMC cannot treat multiple observations per reader/case/modality.

I understand what you are describing for analyzing multiple observations, give 
each replicate (each device specific observation) a new ID. This assumes the 
replicate observations are independent, when they clearly are not.

Your situation reminds me of one solution that probably won't be of immediate 
help to you. It is the work described in Nonparametric Analysis of Clustered 
{ROC} Curve Data, Obuchowski1997_Biometrics_v53p567.

There is surely a way to treat the data, but it's not easy.

Original comment by Brandon.Gallas on 7 Mar 2015 at 6:16

GoogleCodeExporter commented 9 years ago 
Removing priority so that answered questions go to the bottom of the issues 
list.

Original comment by Brandon.Gallas on 25 Mar 2015 at 5:47