Towards optimal human magnetic resonance diffusion tensor imaging (DTI) protocols with wild bootstrap analysis