I would like to better understand the "Mask Count" variable in the ROI Statistics tool.

I'm working with hyperspectral images to which I personally haven't applied any specific masking. Nevertheless, when I create some ROIs and inspect them with the ROI statistics tool (either in ENVI or IDL) statistics are computed on a subset of the ROI pixels (e.g.: the total number of pixels in the ROI is 11 of which only 1 is used to compute the statistics, and therefore the standard deviation is 0, and 10 are masked). In the aforementioned example the ROI statistics panel would read, under "ROI Summary", Pixel Count = 1 and Mask Count = 10.

This behaviour disappears if I open the images with QGIS and save them back to ENVI format. After this step the statistics are computed on the full amount of pixels in the ROI (i.e.: Pixel Count = 11 and no Mask Count variable is present).

My guess is that there is some kind of bad pixels (?) metadata within the image itself that are stripped away when processed by QGIS. If it is so, can somebody please explain to me how to expose this metadata layer, mark the masked pixels and eventually instruct ENVI to ignore it altogether when computing statistics?

If it is not so, and the behaviour is anomalous, does somebody have any indication about what can be causing this masking and how to eliminate it?

Thank you!