Imperfect RF pulses in a multi-spin-echo (MSE) sequence disturb prediction of relaxation times. Provided are a magnetic resonance imaging (MRI) apparatus and method of operating the same, whereby a characteristic parameter value may be acquired from MR signal data via training using an artificial neural network (ANN) and a parametric map may be generated based on the acquired characteristic parameter value. The ANN may be trained to compensate for imperfect RF pulses while providing reduced computation times to produce an output image.