Convolutional neural networks have the potential to predict penumbra volumes within acute ischemic stroke patients to determine their eligibility for mechanical thrombectomy based on the Defuse 3 clinical trial. Currently, computed tomography perfusion is the main method used to quantify penumbra volumes but not all stroke centers have this modality available. In this study, 2 networks were developed to automatically segment penumbra using FLAIR and DWI and performance metrics comparing each network’s predictions with ground truth penumbra (dual network: Dice=0.61, sensitivity=0.68, PPV=0.59, multi-input network: Dice=0.61, sensitivity=0.62, PPV=0.64) indicate a multi-input network is the most capable of segmenting penumbra tissue.
This abstract and the presentation materials are available to members only; a login is required.