Algorithm Of Stroke Onset Time And Region Based On Perfusion Imaging

In China,the incidence,disability rate,recurrence rate and mortality of stroke remain high.Rapid and effective diagnosis of stroke is an important measure to ensure the life safety of stroke patients.Timely treatment within 4.5 hours after onset of acute ischemic stroke will significantly reduce the disability rate and mortality rate of stroke patients,which is of great significance for clinical treatment of cerebral infarction.In this thesis,the classification of stroke onset time and the segmentation algorithm of focal region were studied :(1)After registration,the range of penumbra was calculated according to the Tmax and DWI.Then the feature extraction method was used to extract the feature which can guide the classification.(2)In order to reduce the workload of data annotation,segmentation network based on multi-task learning was used to achieve automatic segmentation of ROI.The segmentation network consists of encoder,decoder and fully connected classifier.In the experimental stage,the classification model was built according to the six groups of selected features combined with LR and SVM.The experimental results of the five-fold cross-validation prove that the fusion image of DWI and Tmax combined with the feature extraction of penumbra ROI makes the model have higher performance and anti-interference.Moreover,its specificity was greater than sensitivity and ACC reached 0.805,which was more suitable for the classification of onset time of stroke.The segmentation performance of 2DUnet,3DUnet,Vnet,Attention Unet and X-Net proposed in this thesis was compared,which proved that the new network performed better than other networks in the segmentation of lesions.There are fewer false positive and false negative segmentation areas,and the Dice coefficient of the new network for DWI and Tmax reaches 80.76% and 82.93% respectively,which completes the segmentation of lesion regions adapted to different scales and images in the same network.The experimental results show that the proposed algorithm has good performance in the classification of stroke onset time and segmentation of lesion region,and has completed the automatic ROI acquisition and classification of onset time.