Aiming at the poor segmentation effect of left ventricle due to the co mplex physiological structure of the left ventricle,uneven texture,high noise, and fuzzy boundary caused by interference of cardiac muscle in cardiac magnetic resonance imaging,this paper proposed a left ventricle segmentation algorithm combining the features of contextual multispectral spatial channels.The algorit hm uses U-shaped structure and jump connection to fuse the feature maps of diff e rent scales to ensure the secondary use of local feature information and improve the feature utilization.The depthwise pooling attention pyramid (DPAP) module is u sed to optimize the useful information of multi-scale features.Then the channe l space attention module in frequency domain is used to optimize the channel and space dimensions of the feature.Finally,combined loss function is used to alle viate the imbalance of positive and negative samples.The results show that on t he MICCAI2009Magnetic Resonance Left Ventricular Challenge Data Set,the simila rity coefficient of the algorithm is 93.25% on the verification set and 93.33% o n the online set, which is 1.6points higher than the mainstream segmentation al gorithm DeepLabV3+,compared to the medical classics The U-Net algorithm has a 6.9point improvement,which can effectively segment the left ventricle.The segm entation results have continuous edges and clear outlines.