Every cluster in your case is gaussian probability distribution density function. In onedimentional case its parameters are mean and variance.
In multidimentional case variance becomes covariance matrix.It describes ellipsoid axis directions and size.
You can reduce covariance to variance but you'll get circle or sphere instead ellipse or ellipsoid.
So ellipse axis directions will be eigenvectors of covariance matrix, and their halflengths will be square root of eigenvalues. Once you know ellipse axis you should deside the way you will convert it to circle. The radius you accept will be square root of variance.
But if you'll need compute probability you'll should compute covariance matrix from your variance by scaling identity matrix by factor equal to variance.