imfindcircles uses circular Hough transform underneath and detecting circles with small radius (usually < 5) is always challenging. I don't see a simple way to play with the parameters of imfindcircles to get rid of the boundary circles. There's a parameter called 'EdgeThreshold' that you can try but the idea there is to keep the strong edges in the image for circle detection and ignore the weak ones. But in this case I won't be surprised if the edges at the boundary of the region are the strongest, in which case this parameter won't be able to remove the ones at the edges (at least not in one call). This means that we need to do some post-processing to identify and eliminate the spurious circles at the edges. If you are using imfreehand to create the region you can get the binary mask of the region as follows:
h = imfreehand();
% Create a binary mask
binaryImage = h.createMask();
Now you can get a mask for the boundary of the region as follows:
rad = 5;
boundaryMask = imdilate(binaryImage, 5) - imerode(binaryImage, 5);
'boundaryMask' will highlight pixels that are within a 'rad' distance from the boundary, on either side. From this point on, it should be easy to check whether a circle (its center maybe) lies within this boundary region or not.
