语义分割用于分割弥散加权成像中脑梗死病灶

Objective To observe the effectiveness of semantic segmentation for segmenting brain infarct lesions on diffusion weighted imaging (DWI). Methods DWI data of 675 patients with newly occurred stroke were retrospectively analyzed. Taken manually depicted ROI of brain infarct lesions as gold standards, the effectiveness of semantic segmentation, threshold segmentation and local entropy information segmentation for segmenting brain infarct lesions on DWI were evaluated with Dice similarity coefficient (DSC) under 10-fold cross-validation and the area under the curve (AUG) of receiver operating characteristic (ROC) curve. Results DSC of semantic segmentation, threshold segmentation and local entropy information segmentation was 0.822. 0.647 and 0.728. with AUG of 0.905. 0.778 and 0.849. respectively. Conclusion Semantic segmentation had curtain clinical application value for segmenting brain infarct lesions on DWI.