if STRUCT marks the start of each record and occurs nowhere else you would be better splitting on STRUCT and just ignoring or shifting off the first element. Split is mostly (always ?) faster than regex.

Cheers,
R.