Abstract
Stroke is the second leading cause of death worldwide, and is increasinglyprevalent in low- and middle-income countries (LMICs). Timely interventions cansignificantly influence stroke survivability and the quality of life aftertreatment. However, the standard and most widely available imaging method forconfirming strokes and their sub-types, the NCCT, is more challenging andtime-consuming to employ in cases of ischemic stroke. For this reason, wedeveloped an automated method for ischemic stroke lesion segmentation in NCCTsusing the nnU-Net frame work, aimed at enhancing early treatment and improvingthe prognosis of ischemic stroke patients. We achieved Dice scores of 0.596 andIntersection over Union (IoU) scores of 0.501 on the sampled dataset. Afteradjusting for outliers, these scores improved to 0.752 for the Dice score and0.643 for the IoU. Proper delineation of the region of infarction can helpclinicians better assess the potential impact of the infarction, and guidetreatment procedures.