The dimensions of the .ddt file must be powers of two (i.e. 1024, 512, 256, etc), though not necessarily square (i.e. 1024x512 is alright) - that's primarily a limitation of graphics cards, which (typically) only support power-of-two textures, because it's much easier for the hardware to do calculations with those.
The only specific requirement of the .ddt format is that the dimensions must be multiples of 4 (so 12x12 is okay, 10x10 is not) (caused by the compression method splitting the image into 4x4 blocks - though it cheats a bit to get the 1x1 and 2x2 mipmaps), but the game requires power-of-two sizes and I guess the .ddt converter enforces that.
The four numbers in brackets are '(usage, alpha, format, levels)', where 'usage' is
* 0, 1 - no idea what that means, but 0 seems common for standard textures
* 6 - bump map
* 7 - no idea
* 8 - cube map (for reflections, etc)
and 'alpha' is 0, 1, 4 or 8, but I don't know what those mean (though there seems to be some tendency towards 0 being no alpha);
and 'format' is:
* 1 = uncompressed 32-bit-per-pixel BGRA (highest quality, highest filesize)
* 4 = DXT1 compressed, 4-bit-per-pixel, 1-bit alpha channel (low quality because of compression artifacts, but small file and saves video memory)
* 7 = greyscale, 8-bit-per-pixel
* 8 = DXT3 compressed, 8-bit-per-pixel, 4-bit alpha channel (similar to DXT1 but can handle smooth alpha blending, and larger file)
and 'levels' is the number of mipmap levels (smaller versions of the texture which are used when it's further away from the camera), which is typically either 1 (if the image is never resized smaller, e.g. the UI) or equal to log2(max(width, height)), i.e. a 1024x512 would have log2(max(1024,512)) = log2(1024) = 10 (because 1024 = 2^10). So if you double the size of an image, and its 'levels' value was not initially 1, you should add one to that value.
... So 0,0,4,1 is sensible for a loading screen - it's a plain texture, with no alpha, with the maximum compression, and only a single mipmap level. You could get better quality by using 0,0,1,1 instead, at the expense of the file being 8 times the size, but usually the compression doesn't look bad and is worthwhile