We provide a DL based tool for detection and segmentation of ischemic acute/sub-acute strokes, trained and tested in 2,628 brain MRIs. Using the original DWI as input, this fully automated system outputs 3D digital infarct mask, volume, and the feature vectors of regions affected by the infarct in two parcellation schemes: structural anatomy and arterial territories. The method is fast (the lesion inference takes 20~30 seconds in CPU; the total processing, including image registration and generation of reports take 3-7 mins, depending on the choice for registration algorithm). ADSv1 includes outputs of the brains and infarct masks mapped to a common space (MNI), ASPECTS calculation, automated radiological reports, with interpretable descriptions of the models' predictions. This system is publicly available, real time, run on local computers, with minimal computational requirements, and accessible to non-expert users.
All software is under the JHU Academic Software License.