Take logical() of the array, to reduce the question to 0 vs 1 (pixel non-zero).
Use conv2() to pass in the logical array together with an array in which array elements are set to 1 in locations that need to be non-zero. If the status of the center pixel is not relevant be sure not to set it to 1 in the mask.
Now compare the result of the conv2 to nnz() of the mask. If the two are the same then all of the pixels in the mask region were set.
This process is essentially the same thing as "image dilation".