DatasetSummary for non-segmentation tasks

[NabJa]

Is your feature request related to a problem? Please describe. The DatasetSummary class computes the dataset statistics (e.g. mean pixel intensity, ...) based on the image and its label. However, there are many cases where this summary might be helpful without having a label. An example would be a simple image classification task. Here, I would want to use these statistics to normalize/process my images, without having a mask as a label.

Describe the solution you'd like The DatasetSummary should allow me to set label_key=None to always compute all statistics without considering the label. Additionally it would be helpful to allow to add a function to select some foreground based on the image (like in transform.CropForeground).

Describe alternatives you've considered It's possible to generate some arbitrary mask (e.g. where image > 0) and add it as label to the dataset. However, this is attached to unnecessary overhead.

Additional context None