Acute myocardial infarction (AMI) is one of the most important diseases threatening human life. The existing MI prognosis prediction scales mostly predict the incidence of death, recurrent MI and heart failure through 6-8 clinical text indicators, and the data are collected relatively simply. Myocardial remodeling, as an adverse pathological change that can start and continue to progress in the early stage after myocardial infarction, is the main pathological mechanism of heart failure and death. However, there is no quantitative early-warning model of myocardial remodeling, and the clinical guidance of early intervention is lacking. Our previous study found that cardiac magnetic resonance imaging can accurately quantify the necrotic area and recoverable myocardium in the edematous myocardium after myocardial infarction. In this study, machine learning algorithm, variable convolution network (DCN) and capsule network (capsnet) are used to build a new neural network architecture. Structural feature extraction of multi-modal clinical image data such as MRI and ultrasound is realized. Combined with the established database of 3000 patients with myocardial infarction, the multimodal feature matrix will be constructed, and a variety of classifiers such as support vector machine (SVM) and random forest (RF) will be used for quantitative prediction of myocardial remodeling, and the effects of different classifiers were evaluated. It is expected that this project will establish a quantitative early warning model of myocardial remodeling after acute myocardial infarction in line with the characteristics of Chinese people. The same type of data outside the database will be used for verification to establish an efficient and stable early warning model.