Personally, I would try this:
1) Introduce an intensity threshold. All pixels below will be treated as 0, all above as 1.
2) Now, start from center of the image and move to the image boundary in multiple rays with different angle (0 to 360°) - like a star shape - and stop as soon as a 0-pixel is hit.
3) Collect all ray lengths. The minimum gives you the shortest distance (from center) where a void exists and thus the radius.
4) Play around with angle step / number of rays and intensity threshold to adjust processing time to needed detection coverage.
What do you think? Do you get my point?