Synthetic-to-real domain adaptation with deep learning for fitting the intravoxel incoherent motion model of diffusion-weighted imaging.