I am writing a review article on common failure modes for a certain technology and to illustrate the different failures, I want to include a range of images from other publications. I have either obtained the permission to use the images or am using images from open access articles with a creative commons license. Some of the images I would like to use are (in the original source) part of a part of a single figure consistining of several subfigures (e.g. I would like to put only subfigure 2e from figure 2 into my paper). Does cropping a subfigure out of a composite figure already constitute a change or an alteration? If so, how should I attribute the changes in the image caption in my paper? Maybe this is really trivial, I nevertheless did not really find an answer anywhere.
I don't see anything special. A caption fitting the journal style but saying something like "taken from fig. 2e in ref #" seems perfect to me. You are not cropping the subfigure.
But even in that case, and even if the figure would be data, you just add "spectrum, plot, whatever cropped" from....
Obviously what you let out should not contradict or shouldn't be much related to the point that you actually discuss.
APA guidelines state adding the phrase "Adapted from". The relevant example shows:
General Format 1 (Figure from a Book):
Note: Descriptive phrase that serves as title and description. Reprinted [or adapted] from Book Title (page number), by Author First Initial. Second Initial. Surname, Year, Publisher. Copyright [Year] by the Name of Copyright Holder. Reprinted [or adapted] with permission.
And for example IEEE states the following:
Always place the Image title/caption below the image
If you adapt, amend, or otherwise modify the image, indicate this at the end of your caption
So adding that it is adapted seems like the correct thing to do here.
Image manipulation becomes problematic when one can judge them as fabrication (that is, the data visualised were actually never obtained through the methods you describe) or as falsification (whereby crucial information are lost) .
It follows that 'beautifying' an image to make them more reader-friendly (e.g. by changing colors) or to add valuable hints (e.g. arrows), or cropping an image to its essential parts while leaving out those sections that are irrelevant to your research would not be deemed fabrications or falsifications.
If you can safely assume that cropping the figure does not lead to an omission of crucial information, then I think you are fine. I would, however, disclose the approach by adding a note that you cropped the image, and by giving a reasons as to why you did so.
 See also this presentation, starting at p. 19: https://www.elsevier.com/__data/assets/pdf_file/0006/1028085/Researcher-Academy-Ethics-30-June-2020.pdf