The purpose of this competition is to develop a more detailed medical image classification model. Because the diagnosis of congenital heart disease mainly relies on cardiac ultrasound examination, the accuracy of cardiac ultrasound is related to the level of ultrasound doctors. In underdeveloped regions, children with congenital heart disease often miss diagnosis. In the diagnosis guide, chest x-ray has a supporting role in diagnosis. We plan to use CXR to assist in the screening of congenital heart disease in underdeveloped regions, but this requires deep learning models to better pay attention to the details around the heart, and it is best to enhance doctors' confidence in the model through attention heat maps.