LSRML: A latent space regularization based meta-learning framework for MR image segmentation

•This work incorporates domain discriminator and image reconstruction into a meta-learning framework for prostate MRI segmentation, and achieves better segmentation and generalization performance than state-of-the-art methods.•This work constructs a large-scale multimodal prostate dataset, which has significant advantages in terms of the quantity of data, data type, and data annotation.•This work analyzes the impact of domain shift on deep learning models through extensive experiments on the proposed and the benchmark datasets.