Automatic ischemic stroke lesion segmentation from computed tomography perfusion images by image synthesis and attention-based deep neural networks

•Fully automated framework for ischemic stroke lesion segmentation form CT perfusion images.•Segmentation through DWI synthesis with CNN and novel lesion-aware loss functions.•Efficient use of raw 4D CT images through high-level compact representation by a feature extractor.•Top performance achieved on MICCAI ISLES challenge 2018.