You have to remember that the FFT operation treats your data as if it's a periodic function. This means that, if you think of your image being defined over the range $x=0..(X-1)$ and $y=0..(Y-1)$ (so the image size is $X\times Y$) then the image should be though of as though it's just one rectangular $X\times Y$ sample taken from an infinite array of images that repeat every $X$ samples in the $x$ direction, and every $Y$ samples in the $y$ direction.
Your filtering operation will 'spread' each pixel form the input image over a larger areas in the output image, and this spreading operation will also span across these periodic boundaries. So, the pixel at $(7,X-1)$ will spread in all directions, and in particular, the 'spreading' to the right will affect pixel $(7,X)$, which is really $(7,0)$ (because the image 'repeats' every $X$ pixels).
If you want to use the FFT2 function to do your filtering, and you want to avoid this periodic 'wrap-around' effect, you need to enlarge your image first. If your Gaussian filter kernel is of size $2*N+1$ (it's usually odd), then you need to add a border of $N$ pixels on the left, right, top and bottom of the image. You also have to choose what pixel values to put in these border pixels. A good choice is to replicate the left-most pixel of each row in the image across the entire row in the left border (and do a similar replication operation into the other borders).
I hope this answers your question