Semi-automatic segmentation of core and penumbra regions in acute ischemic stroke: preliminary results

Acute ischemic stroke is one of the most prevalent vascular diseases in the world. The early diagnosis of perfusion deficits is preferable due to the limited therapeutic window (3h), for the application of thrombolytics. For early diagnosis, it is necessary to calculate perfusion maps which will allow the physician to identify the two ischemic regions: core, irreversible ischemic tissue; and penumbra, salvageable tissue. Perfusion computed tomography (PCT) is a convenient method for diagnosis of perfusion deficits. The goal of this work is the development of a semi-automatic software for the segmentation of both ischemic regions in PCT images. We implemented an algorithm called Local Statistics (LS) that obtains a regional mean and standard deviation extracted from a sample delineated by the user on the region of interest (core or penumbra) of a specific perfusion map. The algorithm expands the selected region accepting boundary pixels from an 8- neighbourhood only if they pertain to the interval defined by the distribution parameters obtained from the sample. The process stops when all boundary pixels on the last ischemic region obtained do not satisfy the acceptation criteria. The segmentation results were compared with the manual segmentation performed by a clinician (CB), considered as the reference standard for our work. Visual inspection showed that the software provided similar results to the manual methodology. Implementation of quantitative performance metrics is needed for an accurated validation of our method.