We provide a tool for detection and segmentation of ischemic acute and sub-acute strokes in brain diffusion weighted MRIs. The deep learning networks were trained and tested on a large dataset of 2,348 clinical images, and further tested on 280 images of an external dataset. Our proposed model outperformed generic nets and patch-wise approaches, particularly in small lesions, with lower false positive rate, balanced precision and sensitivity, and robustness to data perturbs (artefacts). The agreement with human delineation rivaled the inter-evaluator agreement. The method has minimal computational requirements and is fast: the lesion inference takes 20~30 seconds in CPU and the total processing, including registration and generation of results/report takes ~ 2.5 mins. The outputs, provided with a single command line, are: the predicted lesion mask, the lesion mask and the image inputs (DWI, B0, ADC) in standard space, MNI, and the quantification of the lesion per brain structure and per vascular territory.
GNU General Public License v3
GNU GENERAL PUBLIC LICENSE Version 3, 29 June 2007
Copyright (C) 2007 Free Software Foundation, Inc. <http://fsf.org/>
Everyone is permitted to copy and distribute verbatim copies
of this license document, but changing it is not allowed.
See <https://www.gnu.org/licenses/gpl-3.0.de.html> for full text.