Dear colleagues,
we have acquired fluorescence images (1024x1024 pixels) using two channels from 34 samples which belong to two groups (negative/positive). As these images show high heterogeneity and the samples differ in size, we segmented the tissue from the background and randomly choose 100 pixels within the segmented area. We now want to compare the two groups (positive/negative) as well as the correlation between the channels statistically.
Our null-hypothesis would f.e. be:
There is no difference is fluorescence intensity in channel 1 between the group positive and negative.
We thus have 100 observations per sample which are however not linked to an in-subject variable as normally would be the case for an mixed ANOVA. Performing a t-test directly on the 3400 values would lead to very low p-values and would be statiscally incorrect as the observations within one sample are most likely correlated.
How can we take this multiple observations into account to compute the p-value?
Hello Mikael,
If the 100 pixels were randomly chosen for each of the 34 cases, and your hypothesis concerns intensity of the pixels, then one approach that could be used is to average the intensity of the 100 data points for each case, then compare the two sets of cases for possible differences in (mean) via a t-test (with 32 df).
The alternate approach is to consider pixels as nested within cases, then handle the analysis as either a doubly nested anova or hierarchical linear model. The final test of group differences, however, will still have the same df (1, 32), and the test is functionally the same as the first approach.
Good luck with your work!
Thank you for the answer. I will perform the analysis by just averaging the 100 points as this seems more intuitive.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA