I have already considered procedural generation, and I haven't ruled it out.
I made a PG dungeon algorithm a year ago for a friend, but his project didn't go any further than asking me to make that algorithm, which annoyed me. It's quite bit different, but I could reuse the ideas I had then as a starting point here.
However, I'm concerned that the end result wouldn't be worth the effort. Levels designed by a human can be clever, interesting, and with balanced difficulty. I think these qualities are important for arcade games with fixed level sequences. Procgen all too often produces data that seems to lack those qualities.
So I'm undecided. It's something I can look at towards the end.