Remember a spline's "radius" is infinitely changing. That's what makes it a spline. So at any one single point along a spline, the more accurate you try to be the more the radius tends toward infinity - i.e. a straight line.
It's only once you reduce the accuracy that the radius starts to make some sense. That's because of how a radius at a point along a curve is calculated. You pick the point, then calculated 2 points either side of it equi-distant from the first. Then calculate the centre point and the radius. Simple way is to bisect a line between the mid and each side, calculate a perp line to those and see where these intersect - that would be the centre point. As sample:
http://www.lee-mac.com/3pointarccircle.html And from that you should understand why a spline's "exact" radius appears to be infinite - as if it's a straight line: As the measuring distance between the picked point and its temporary measure points either side gets ever smaller, the radius increases, up to the point where the floating point number used in ACad for stuff like distances, angles, XYZ values, etc. gets to its minimum possible value (called epsilon).
The calculations become more difficult once you throw a z-value into the mix as well. This means you first need to define a UCS going through those 3 points (in order to get a flat pane between them). Only then can you properly calculate the radius in 3d space between those 3 points.
However, remember you're only approximating such "radius", since strictly speaking a spline has "no" radius at any point. And worst is that this technique wont work (without some modifications) on the start and end points of a non-closed spline.
There are other ways to calculate the same. Some math matrix calcs or trig may be used. But for starters I'd simply go with the geometric calc explained above. Just choose some chord distance between the 3 points which is appropriate for your scenario.
