Irreversible infarcts are critical for the assessment of potential risk and benefit pertaining to thrombolysis in hyper-acute ischemic stroke. It is a challenging work to segment infarct at hyper-acute stage due to the substantial variability. A general abnormal tissue segmentation method is proposed and applied to segment hyper-acute ischemic infarct in this paper. Multiple features are designed to train a random forest classifier for voxel classification. Sparse coding based bag-of-features is adopted to train a region classifier for infarct region recognition. The proposed method has been validated on 98 consecutive patients recruited within 6 hours from onset and achieved a higher Dice coefficient 0.774±0.117 than the other two existing methods (0.755±0.118; 0.597±0.204). It could provide a potential tool to quantify infarcts from diffusion weighted imaging at hyper-acute stage with accuracy to assist the decision making especially for thrombolytic therapy.
展开▼