They are converted that way because they contain no background. After PDF v1.4 the resulting pixel on a page is the composition of all overlaying graphic objects, each of which may carry an alpha component. If there is no explicit opaque object (e.g. a white rectangle) overlaying the entire page, under all the other objects, then there is no background where there are no overlying objects. It could be that the "working" images are PDF v1.3 or have an explicit background layer.
You could:
adjust the PDFs at source, or paint the resulting image over a filled canvas image to obtain an image guaranteed to have a background