Symmetry determined superpixels for efficient lesion segmentation of ischemic stroke from MRI

Non-invasive, quantitative and robust identification of ischemic stroke and estimation of injury extent is essential for assisting neuroradiologists. Manual lesion delineation techniques are susceptible to subjective errors and therefore computer aided preliminary screening of necrosis is warranted. Superpixel based segmentation has gained importance in the recent past by reducing the computational complexity and preserving the characteristics of a group of pixels with similar properties. Axial and coronal MR images of brain exhibit the important feature of symmetry which was integrated with superpixels for segmenting ischemic lesion. This method was evaluated on a challenging 10 patient data set along with MICCAI challenge data of 28 patients yielding promising results. Proposed symmetry determined superpixel based method demonstrated accuracy close to manual lesion demarcation with high performance indices with average sensitivity of 82.32%, specificity of 93.7% and Dice similarity score of 81.14%.