Optimizing 'spread' in Matlab's built-in radial basis network function

In later versions of Matlab, there are functions named newrb and newrbe to design radial basis networks. For both functions, there is an argument called "spread" which is a user-defined constant. It represents the measure of standard deviation of the Gaussian kernel for radial basis function, and all of the neurons will share same "spread" for a given configuration. In most cases, it needs to be optimized by trial and error. However, in my opinion, the network would be able to approximate desired function with much better capability if it could optimize the "spread" itself for each individual neuron (instead of having one shared by all neurons). Is there any way to achieve that using a built-in function or customize pieces of functions that could be stitched together inside these functions (newrb and newrbe)?