Numerical Methods of Optimum Experimental Design Based on a Second-Order Approximation of Confidence Regions