Another unbreakable encryption from Bruce! His words.
Yes the boxes come from π. However, it's not the way you think. It's not a literal base change of the decimal expansion. You have to calculate π from scratch in hexadecimal. If you were to use your approach and simply change the base of the decimal expansion, you'd be putting 10 values into 16 bins for each digit. In hex, π is just:-
3.243F6 A8885 A308D 31319 8A2E0 37073 44A40 93822 299F3 1D008 2EFA9 8EC4E 6C894 52821 E638D 01377 BE546 6CF34 E90C6 CC0AC
which are your digits.
You do understand that the P and S boxes don't contain π? They are initialised with π digits, but then amended in a key related fashion. The π bit is what's know as nothing up your sleeve numbers (NUMSN), and there are many answers here relating to such initialisation. In summary, you have to initialise a variable with something or zero. The purpose of NUMSN is to create a dissociation between the numbers you start with and the way the algorithm works so that there cannot be any possible adverse relationship. There is some expansion here.
On that basis, yes you could initialise with anything. Even just zeros. Although, I've never seen security analysis of an algorithm with and without the NUMSN.
After all, I saw that sbox values were supposed to be random.
This is a contentious issue. For a super dooper efficient algorithm, s boxes are very cleverly designed with several security parameters in mind. There is an s-box tag that will highlight a pile of appropriate questions. Uber efficiency isn't always necessary. Brute force can be used instead by increasing the number of rounds. As Blowfish recalculates it's S and P boxes dynamically, such security pre design clearly isn't used in that case. I believe that this technique works for both the S and P boxes.
