Good idea.
I looked at the source code but the images are converted to format A8R8G8B8 before saving to vtdb.
I use format A4R4G4B4 for transparent and R5G6B5 for non-transparent images.