Exclude some characters from a Lex regex - html

Question

I am trying to build a regex for lex that match the bold text in mardown syntax. For example: __strong text__ I thought this:
__[A-Za-z0-9_ ]+__
And then replace the text by
<strong>Matched text</strong>
But in Lex, this rule causes the variable yytext to be __Matched Text__. How could I get rid of the underscores? It would be better to create a regex that does not match the underscores or proccess the variable yytext to remove it?
With capturing groups it would be easer, because I would only need the regex:
__([A-z0-9 ]+)__
And use \1. But Lex does not support capturing groups.
Answer
I finally take the first option offer by João Neto, but a little modified:
yytext[strlen(yytext)-len]='\0'; // exclude last len characters
yytext+=len; // exclude first len characters
I've tried with Start conditions as he mentioned as second option, but did not work.

Answer 1

You can process yytext by removing the first and last two characters.
yytext[strlen(yytext)-2]='\0'; // exclude last two characters
yylval.str = &yytext[2]; // exclude first two characters
Another option is to use stack
%option stack
%x bold
%%
"__" { yy_push_state(bold); yylval.str = new std::string(); }
<bold>"__" { yy_pop_state(); return BOLD_TOKEN; }
<bold>.|\n { yylval.str += yytext; }