The fact that an allocation is a pitched allocation shouldn’t matter for ROI definition.
In most cases, you should not need an API to define an ROI.
If you want to process an entire image, the ROI is the same as the image dimensions. The image dimensions aren’t affected by whether the underlying allocation is pitched, or not. If your overall image is a 640x480 image, and you are processing the whole image, then your ROI is 640x480.
If you want to process a rectangular region within an image, the ROI is the dimensions of that rectangular region. Again, not a function of whether the underlying allocation is pitched, or not.
There are various CUDA sample codes that use NPP and ROIs, you can study any of those.