Traditionally, file systems are hierarchical trees. Each directory has a 0-n other sub-directories these also have 0-n subdirectories, these too, and so on.
Relational databases aren't very good for expressing tree data structures, because it requires you to do an unknown amount of self-joins to traverse the tree.
It would of course be theoretically possible to design a filesystem which isn't a directory tree. I could imagine a filesystem without directories where every filename in the whole filesystem must be unique so it can be used as a primary key (that would be like having every single file of a system in the same directory), or where the primary key for each file is application, user, filename and revisionNumber (so you essentially have a fixed tree depth). But that would be a concept which is completely different than Windows, POSIX or any other widely-used operating system standard. It would be unintuitive to use for users who come from these platforms and it would be considerable effort to port existing software to such a filesystem or make it interface with a system which uses a hierarchical filesystem.