Fig. 1: Overview of the DSL architecture.
From: Mining multi-center heterogeneous medical data with distributed synthetic learning
a The architecture contains one central generator and multiple distributed discriminators, each located in a medical entity. Then the well-trained generator can be used as an image provider to build a synthetic database for downstream machine-learning tasks. The following three rows show three different scenarios for heterogeneous medical data. In (b), the generator learns and generates multi-modality synthetic images at the same time. In (c), the data centers provide data with misaligned modalities. To highlight, the proposed DSL framework could leverage the misaligned information from multiple data centers and synthesize unified multi-modality images. In (d), the temporal data centers can only be accessed sequentially. DSL can learn from each temporal dataset without catastrophically forgetting what the model has previously learned.
