核磁共振图像（Magnetic Resonance Image, MRI）作为判断心脏结构和功能的“金标准”，针对人工方法分析MRI图像手动分割费时费力，且结果可能因人而异的问题。本文提出了基于深度学习的短轴心脏MRI分割方法。分割方法包括预处理和分割两个步骤。预处理包括：首先对短轴心脏MRI数据进行感兴趣区域（Region of Interest，ROI）检测，使用了Canny边缘检测和霍夫变换，然后进行数据标准化和数据增强。分割使用深度学习中的语义分割模型，损失函数使用加权交叉熵损失、Dice损失、L2正则化的组合，对ROI的结果进行分割。本文的实验基于ACDC数据集，实验结果表明，本文提出的方法成功地从心脏MRI中自动分割出左心室、右心室和心肌，与不同的损失函数和挑战中领先的几种分割方法进行对比，取得了很好的结果。

As the "gold standard" to judge the structure and function of the heart, cardiac MRI requires experts to manually depict the contours of left ventricle, right ventricle and myocardium to provide guidance for the diagnosis of heart disease. However, manual segmentation is time-consuming and laborious, and the results may vary from person to person. This paper proposes a short axis cardiac MRI segmentation method based on deep learning. The segmentation method consists of two steps: preprocessing and segmentation. Preprocessing includes region of interest (ROI) detection of short axis cardiac MRI data normalization and data augmentation. Canny edge detection and Hough transform is used in ROI detection. The semantic segmentation model in deep learning is used for segmentation, and the combination of weighted cross entropy loss, dice loss and L2 regularization is used for loss function to segment the ROI results. The experiments in this paper are based on ACDC data sets. The experimental results show that the proposed method can automatically segment left ventricle, right ventricle and myocardium from cardiac MRI successfully. Compared with several leading segmentation methods in different loss functions and challenges, excellent results are obtained.