M31 Mosaic processing help

I've just started making mosaics and this approach has been working so far.

In Siril, when I stack, I use the script that applies background extraction to every frame before stacking. Once I have the mosaic stitched I'll run a background extraction operation on the full mosaic. You might want to try that on your final image and see if it does anything for you.

Given that you've had 38 hours split on six panels that would give about 6 hours + in total integration, unless you meant that each panel has 38 hours into it, so it's a bit on the short end for my taste. But to start off I would give this a read: https://chaoticnebula.com/pixinsight-mosaic-workflow/ 

I'm not sure how it holds up in terms of processes/scripts available today as it's been a while since I've read it. Still, it does go through a lot of the basics and should be quite helpful. The first thing I'd do is drop GraXpert, if you depend on some sort of automated process I would rather suggest SetiAstro and the AutomaticDBE script. GraXpert does not treat the data very kindly and I would only use it in extreme cases. Second, don't do NoiseXterminator on your linear image. This is actually advise from the creator himself. If you still insist on it you should be sure to set the detail slider to zero. I would still wait until you've stretched it either way. DBE and some crop needs to be done in the mosaic phase but after you could try something like this; BXT Correct Only > SPCC (you need to calibrate your colors) > BXT deconvolution > SXT (StarXterminator) with large overlap.

Then you can process your stars and galaxy from there with whatever stretching method you prefer. I mostly prefer masks and HT transformation, GHS is a valid option too. After stretching you can start applying some noise reduction, or you could leave it till the very end of the process before re-screening the stars back into the image.

EDIT: Is the sub-frames properly calibrated? The image seems to have dust spots present above the galaxy to the left.

The gradient issue is very challenging.  Given your six-panel layout, the bright part of the galaxy always occupies one end of any panel, and this pretty much prevents good gradient removal in an individual panel no matter how you do it.  M31 will severely interfere with gradient removal.  If you try very hard, a gradient removal (using ABE, or DBE, or any other algorithms) may give you a single panel that looks fine, but after you stitch six of them you will still find complex residual gradients in the mosaic.

My suggestion is to use a shorter focal length scope/lens to image the whole area (better about 1.5x wider) of the mosaic and use this as a gradient reference in the photometric mosaic script.  The wide-field reference image doesn't need to be deep.  Just an hour or two of integration should be enough.  If it is wide enough, it should contain large enough area without the light from M31, so you can run ABE or DBE very well on it.  Such a wide-field image after ABE/DBE can serve as a very good model for photometric mosaic. You already spent 38 hr on this project.  One or two more hr of integration on a wide-field image can bring it to perfect.

One thing that may help slightly is to incorporate local normalization in the stacking process for an individual panel (and also the wide-field reference). This can make the gradient in each panel more manageable, especially for a panel that has very long integration like yours.