The process involves a few steps - some of them because I am using Silverfast SE, which only scans to 24-bit RGB.
First I scan after modding the levels using the Silverfast Histogram to truncate the flat ends of the histogram. I use 'medium' USM when scanning.
Then after imported into photoshop, I convert to 48 bit, then export each channel as a 16 bit greyscale image.
I run each greyscale through Neat Image to minimise noise - this has the added effect of 'rebuilding' some tonal detail by averaging pixels interpreted as noise and resampling that average back into the image.
After noise reduction I open each of the processed greyscales in photoshop and run another tone adjustment using the levels tool, slicing off the low end until the blacks are black, and the whites are bright.
I then run a Smart Sharpen on each of the greyscales, using
Lens Blur.
After that, I recombine the image, copying each greyscale image into its appropriate channel in the RGB image.
I then run more curves / levels adjustment, colour enhancements, and a manual clone touchup to remove any dust marks I find.
Finally, I run an appropriate USM on the final image, and resize for output. I may run an additional minor USM on the resulting resize as well, depending on whether it needs it.